DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of Claims
This action is in reply to the amendment filed on 02/26/26.
Claims 1, 9 have been amended and are hereby entered.
Claims 15-17 have been canceled.
Claims 21-23 have been added.
Claims 1-14, 18-23 are currently pending and have been examined.
This action is made final.
Continuity/Priority Date
Status of this application as a 371 of PCT/EP2021/062085 is acknowledged. Accordingly, a priority date of 05/06/2021 has been given to this application.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-14, 18-23 are rejected under 35 U.S.C.101 because the claimed invention is directed to a judicial exception (an abstract idea) without significantly more.
Step 1
Claim 1 is drawn to a system, Claims 2-11, 21-23 are drawn to a method, and Claims 12-14, 18-20 are drawn to a non-transitory computer-readable medium, each of which are within the four statutory categories. Claims 1-14, 18-23 are further directed to an abstract idea on the grounds set out in detail below.
Claim 1 recites implementing the steps of:
receiving a first user action during a conversational interaction;
determining a first feature input based on the first user action in response to receiving the first user action, wherein the first feature input is a vectorization of a conversational detail or information from a user account of the user;
inputting the first feature input into a first model;
receiving the first output from the first model, the first output indicative of a selected context of the plurality of contexts;
selecting a second model, from a plurality of models, based on the selected context, wherein each context of the plurality of contexts corresponds to a respective model from the plurality of models;
inputting the first output into the second model;
receiving, using the control circuitry, a second output from the second model; and
selecting a mental health disorder recommendation from a plurality of mental health disorder recommendations based on the second output; and
transmitting the mental health disorder recommendation following the conversational interaction.
These steps amount to managing personal behavior or relationships or interactions
between people and therefore recite certain methods of organizing human activity. Collecting and processing data associated with a patient using a series of models to select and transmit a mental health disorder recommendation for the patient is a personal behavior that may be performed by a mental healthcare provider.
Claims 2 and 12 recite implementing the steps of:
receiving a first user action during a conversational interaction;
in response to receiving the first user action, determining a first feature input based on the first user action;
using the first feature input in a first model to select a context from a plurality of contexts based on user actions, wherein the first model is trained to select a context from a plurality of contexts based on user actions, wherein each context of the plurality of contexts corresponds to a respective emotional state of a user
receiving a first output from the first model;
using the first output in a second model to select an emotional state from a plurality of emotional states of the selected context based on the first output, wherein the second model is trained to select an emotional state from a plurality of emotional states of the selected context based on the first output, and wherein each emotional state of the plurality of emotional states corresponds to a respective emotional state of the user;
receiving a second output from the second model;
selecting a mental health disorder recommendation from a plurality of mental health disorder recommendations based on the second output; and
generating the mental health disorder recommendation following the conversational interaction.
These steps amount to managing personal behavior or relationships or interactions
between people and therefore recite certain methods of organizing human activity. Collecting and processing data associated with a patient using a series of models to select and generate a mental health disorder recommendation for the patient is a personal behavior that may be performed by a mental healthcare provider.
The above claims are therefore directed to an abstract idea.
Step 2A Prong 2
This judicial exception is not integrated into a practical application because the additional
elements within the claims only amount to:
A. Instructions to Implement the Judicial Exception. MPEP 2106.05(f)
The independent claims additionally recite:
cloud-based control circuitry as implementing the steps of the abstract idea (Claim 1)
a first user interface as implementing the step of receiving a first user action during a conversational interaction (Claim 1)
a second user interface as the entity to which the mental health disorder recommendation is transmitted (Claim 1)
a user interface as implementing the step of receiving a first user action during a conversational interaction (Claims 2, 12)
control circuitry as implementing the steps of determining a first feature input based on the first user action, inputting the first feature input into a first model, receiving a first output, inputting the first output into a second model, receiving a first output, receiving a second output from the second model, selecting a mental health disorder recommendation from a plurality of mental health disorder recommendations based on the second output (Claim 2)
first/second machine learning models as implementing the steps of selecting a context from a plurality of contexts and selecting an emotional state from a plurality of emotional states, respectively (Claims 1, 2, 12)
a non-transitory computer-readable medium comprising instructions that are executed by one or more processors as implementing the steps of the abstract idea (Claim 12)
The broad recitation of the aforementioned general purpose computing elements at a high level of generality only amounts to mere instructions to implement the abstract idea using computing components as tools. Regarding the cloud-based control circuitry, per [0032], [0037] these are understood to be general purpose computing elements functioning in their ordinary capacities to implement the steps of the abstract idea. Regarding the user interfaces, this is understood to be the display of an electronic mobile device such as a smartphone with a touchscreen for inputting/outputting information per paras. [0033], [0034]. Regarding the control circuitry, this element is understood to amount to “any suitable processing, storage, and/or input/output circuitry” per [0033] and therefore is given its broadest reasonable interpretation as general purpose computing components functioning in their ordinary capacities to implement the abstract idea. Regarding the machine learning models, the claim does not disclose the particulars of the machine learning model (e.g., the specific type of machine learning model, the data used for training it). The broad recitation of first and second machine learning models, in this case to determine a context and an emotional state, respectively, only amount to using the machine learning model as a tool to apply data to a model and generate a result (see MPEP 2106.05(f)(2)). Regarding the non-transitory CRM comprising instructions for execution by a processor, per [0035], these are understood to be general purpose computing elements functioning in their ordinary capacities.
B. Insignificant Extra-Solution Activity. MPEP 2106.05(g)
Claim 1 additionally recites:
cloud-based storage circuitry configured to: store a first machine learning model, wherein the first machine learning model is trained to select a context from a plurality of contexts based on user actions, and wherein each context of the plurality of contexts corresponds to a respective emotional state of a user following a first user action; and store a second machine learning model, wherein the second machine learning model is trained to select an emotional state from a plurality of emotional states of a selected context based on a first output, and wherein each emotional state of the plurality of emotional states corresponds to a respective emotional state of the user;
This element amounts to insignificant extra-solution activity. As explained above, Claim 1 is directed to an abstract idea in the form of collecting and processing data associated with a patient using a series of models to select and generate a mental health disorder recommendation for the patient. As stated in MPEP 2106.05(g), "[t]he term "extra-solution activity" can be understood as activities incidental to the primary process or product that are merely a nominal or tangential addition to the claim." In the present claim, the function of storing the first and second machine learning models in cloud-based circuitry is only nominally or tangentially related to the process of collecting and processing data associated with a patient using those first and second machine learning models to select and generate a mental health disorder recommendation for the patient, and accordingly constitutes insignificant extra-solution activity.
Thus, taken alone, the additional elements do not integrate the above-identified judicial exception into a practical application. Looking at the limitations as an ordered combination adds
nothing that is not already present when looking at the elements taken individually. Their
collective functions merely provide conventional computer implementation.
The above claims, as a whole, are therefore directed to an abstract idea.
Step 2B
The present claims do not include additional elements that are sufficient to amount to
more than the abstract idea because the additional elements or combination of elements amount to no more than a recitation of:
A. Instructions to Implement the Judicial Exception. MPEP 2106.05(f)
As explained above, claims 1, 2, 12 only recite the aforementioned computing elements as tools for performing the steps of the abstract idea, and mere instructions to perform the abstract idea using computing elements operating in their ordinary capacities is not sufficient to amount to significantly more than the abstract idea. MPEP 2106.05(f).
B. Insignificant Extra-Solution Activity. MPEP 2106.05(g)
Likewise, as explained above, the element of cloud-based storage circuitry configured to: store a first machine learning model, wherein the first machine learning model is trained to select a context from a plurality of contexts based on user actions, and wherein each context of the plurality of contexts corresponds to a respective emotional state of a user following a first user action; and store a second machine learning model, wherein the second machine learning model is trained to select an emotional state from a plurality of emotional states of a selected context based on a first output, and wherein each emotional state of the plurality of emotional states corresponds to a respective emotional state of the user, only amounts to insignificant extra-solution activity.
C. Well-Understood, Routine and Conventional Activities. MPEP 2106.0S(d)
In addition to amounting to insignificant extra-solution activity the elements in Section B above constitute well-understood, routine and conventional activity. The element of cloud-based storage circuitry configured to: store a first machine learning model, wherein the first machine learning model is trained to select a context from a plurality of contexts based on user actions, and wherein each context of the plurality of contexts corresponds to a respective emotional state of a user following a first user action; and store a second machine learning model, wherein the second machine learning model is trained to select an emotional state from a plurality of emotional states of a selected context based on a first output, and wherein each emotional state of the plurality of emotional states corresponds to a respective emotional state of the user to a user only amounts to storing/retrieving data in memory, which has been previously held to be well-understood, routine and conventional when claimed at a high level of generality or as insignificant extra-solution activity. See MPEP 2106.05(d)(II).
Thus, taken alone, the additional elements do not amount to significantly more than the
above-identified judicial exception. Looking at the limitations as an ordered combination adds
nothing that is not already present when looking at the elements taken individually. Their
collective functions merely provide conventional computer implementation.
Dependent claims recite additional subject matter which further narrows or defines the abstract idea embodied in the claims. For example, Claims 5, 6, 7, 9, 19 recite limitations which further narrow the scope of the independent claims. Claims 3, 4, 8, 10, 11, 13, 14, 18, further recite limitations that are certain methods of organizing human activity but for recitation of general purpose computing elements consistent with those discussed above with respect to the independent claims.
Claims 10 and 20 additionally recite, in part, limitations pertaining to determining a network location of the user information and generating a network pathway to the user information which amounts to mere instructions to apply the abstract idea using general purpose computing elements, e.g., using a computer to determine and generate a pathway to communicate electronically with the user. This is not sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception.
Claims 11 and 20 additionally recite, in part, limitations pertaining to automatically retrieving the user information from the network location based the mental health disorder recommendation, which comprises an additional element in the form of insignificant extra-solution activity. As stated in MPEP 2106.05(g), "[t]he term "extra-solution activity" can be understood as activities incidental to the primary process or product that are merely a nominal or tangential addition to the claim." In the present claim, the function of retrieving the user information from a network location is only nominally or tangentially related to the process of collecting and processing data associated with a patient using those first and second machine learning models to select and generate a mental health disorder recommendation for the patient, and accordingly constitutes insignificant extra-solution activity. This element also amounts to well-understood, routine and conventional activity, as it only amounts to storing/retrieving data in memory and/or transmitting data over a network, which have been previously held to be well-understood, routine and conventional when claimed at a high level of generality or as insignificant extra-solution activity. See MPEP 2106.05(d)(II). Claims 11 and 20 also recite generating for display the user information on a second user interface which only amounts to mere instructions to apply the abstract idea on a computer, e.g., using a computer interface to display the results. This is not sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception.
Claim 21 recites limitations pertaining to wherein the second machine learning model comprises: a shared encoder jointly trained across (i) a categorical emotional-state framework and (ii) a dimensional valence-arousal-dominance emotional framework; and a plurality of separate prediction heads corresponding respectively to the categorical emotional- state framework and the dimensional valence-arousal-dominance emotional framework, which only amounts to mere instructions to apply the abstract idea on a general purpose computer. This is not sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception.
Claim 22 recites limitations pertaining to wherein the second machine learning model is configured to output a probability score for each emotional state of the plurality of emotional states, and wherein selecting the mental health disorder recommendation is based on a probability distribution over the plurality of emotional states, which only amounts to mere instructions to apply the abstract idea, e.g., using a machine learning model on a general purpose computer to output a probability score and select a recommendation based on probability distribution. This is not sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception.
Claim 23 recites limitations pertaining to updating one or more parameters of the second machine learning model using backpropagation based on differences between predicted emotional states and labeled emotional states during training of the second machine learning model, which only amounts to mere instructions to apply the abstract idea on a computer, e.g., updating model parameters using a known algorithm (backpropagation). This is not sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception.
Dependent claims 3-11, 13-14, 18-23 have been given the full two-part analysis including analyzing the additional limitations both individually and in combination. The dependent claims, when analyzed individually, and in combination, are also held to be patent ineligible under 35 U.S.C. 101 as they include all of the limitations of claim 2 or claim 12 respectively. The additional recited limitations of the dependent claims fail to establish that the claims do not recite an abstract idea because the additional recited limitations of the dependent claims merely further narrow the abstract idea. Beyond the limitations which recite the abstract idea, the claims recite additional elements consistent with those identified above with respect to the independent claims which encompass adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f). Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Dependent claims 3-11, 13-14, 18-23 recite additional subject matter which amounts to additional elements consistent with those identified in the analysis the independent claims above. As discussed above with respect to the independent claims and integration of the abstract idea into a practical application, recitation of these additional elements (e.g., user interface, cloud circuitry, machine learning models, etc.) only amounts to invoking computers as a tool to perform the abstract idea. Looking at the limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. There is no indication that the combination of elements improves the functioning of a computer or improves any other technology. Their collective functions merely provide conventional computer implementation.
Dependent claims 3-11, 13-14, 18-23, when analyzed as a whole, are held to be patent ineligible under 35 U.S.C. 101 because the additional recited limitation(s) fail(s) to establish that the claim(s) is/are not directed to an abstract idea without significantly more. These claims fail to remedy the deficiencies of their parent claims above, and are therefore rejected for at least the same rationale as applied to their parent claims above, and incorporated herein.
For the reasons stated, Claims 1-14, 18-23 fail the Subject Matter Eligibility Test and are consequently rejected under 35 U.S.C. 101.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1 is/are rejected under 35 U.S.C. 103 as being unpatentable over Generoso et. al. (US Publication 20200020447A1) in view of Osotio et. al. (US Publication 20180101776A1), and further in view of Alikov et. al. (US Publication 20220215452A1).
Regarding Claim 1, Generoso discloses:
cloud-based storage circuitry configured to ([0031], [0035], [0082]): store a first machine learning model, wherein the first machine learning model is trained to select a context from a plurality of contexts based on user actions (Abstract teaches on using a first machine learning model to generate a user analysis profile; [0048] teaches on machine learning models are supervised models initially generated through a training process; see also [0063] regarding training of ML models; [0058] teaches on the prediction module generating the user analysis profile including a personality profile, which may include data about posting frequency, comments on friends’ posts, interactions with friends, group posts, e.g., user actions; [0059], [0060] respectively teach on the prediction module generating second and third elements of the user analysis profile, a mood score and identified bullying; [0061] teaches on the prediction module generating the fourth element of user analysis, psychological symptoms, by considering data about external events in the user’s life such as death of a loved one, reduction in social network activity, reduction in movement; where [0038] teaches on the prediction module being configured to use machine learning to identify potential anomalies in the user’s health, including the user’s mental health, based on data ingested; anomalies identified are interpreted as a “contexts”), and wherein each context of the plurality of contexts corresponds to a respective emotional state of a user following a first user action ([0061] teaches on the prediction module identifying symptoms of psychological issues such as anxiety or depression, interpreted as “emotional states”; [0062] provides an example of identifying anger via social media content posted); and
store a second machine learning model, wherein the second machine learning model is trained to [identify potential health anomalies to identify a potential health condition] based on the first output,; (Abstract teaches on a second machine learning model which classifies the user analysis profile to determine a weighted score associated with potential health issues for the user; [0038] teaches on the action module 240 (interpreted as second machine learning model) is configured to use further machine learning to analyze the results of the prediction module (first output from first machine learning model); [0050] teaches on the action module using machine learning techniques to determine a weighted score related to detecting a possible health issue using output from the prediction module; [0051] teaches on the collaboration and deployment driver 242 and analytical behavior module, which are part of the action module 240, taking the identified potential anomalies to identify potential health issue of the user);
cloud-based control circuitry configured to ([0031], [0035], [0082]): receive the first user action during a conversational interaction with a first user interface ([0005] teaches on receiving data related to social media activity for a user; [0015] teaches on ingesting information about a user’s social media activity including frequency of posts, participation in groups, content of user’s posts, frequency of likes on friends’ posts; [0058] further teaches on generating a user analysis profile by using user activity on a social network such as participation in groups and interaction with friends and comments on friends’ posts; [0059] teaches on analyzing “tone of user’s voice in conversations” which indicates conversational interaction; the former examples of interacting with social media are interpreted as “user action” received during “conversational interaction with a user interface” as comments/interactions are inputted via user interface; if the user activity is used to generate a user analysis profile, it is understood that the user actions have been received in order to do so);
determine a first feature input based on the first user action in response to receiving the first user action, wherein the first feature input is a vectorization of a conversational detail or information from a user account of the user ([0044] teaches on a real-time analytics engine which can process data to put it in a more suitable form for use by the prediction module (first machine learning model); the real time analytics engine can generate feature vectors related to the input data; in this embodiment, a feature vector is a vector having any number of dimensions which represents a numerical representation of the associated electronic data; this “vectorization” of electronic data may improve the speed, efficiency and/or accuracy of the system);
input the first feature input into the first machine learning model ([0005] teaches on generating a user analysis profile for a user by analyzing the received data using a first machine learning model; per [0015], information about a user’s social media activity (frequency of posts, participation in groups, etc.) can be ingested and classified using a trained ML model to generate a personality profile for the user; [0059], the received social media data includes the first feature input – laughter, positive/negative comments, voice tone in conversations);
receive the first output from the first machine learning model, the first output indicative of a selected context of the plurality of contexts ([0061] teaches on outputting the results (blocks 504, 506, 508, 510) of the prediction module, interpreted as the first machine learning to the action module per [0038]; the output corresponds to identification of an anomaly which is interpreted as “context” as indicated above));
input the first output into the second machine learning model (Abstract teaches on a second machine learning model which classifies the user analysis profile to determine a weighted score associated with potential health issues for the user; [0038] teaches on the action module 240 (interpreted as second machine learning model) is configured to use further machine learning to analyze the results of the prediction module (first output from first machine learning model); [0050] teaches on the action module using machine learning techniques to determine a weighted score related to detecting a possible health issue using output from the prediction module; [0051] teaches on the collaboration and deployment driver 242 and analytical behavior module, which are part of the action module 240, taking the identified potential anomalies to identify potential health issue of the user);
receiving, using the control circuitry, a second output from the second machine learning model ([0005] teaches on determining a weighted score associated with detecting a potential health issue for the user by classifying the user analysis profile using a second machine learning model; the weighted score is interpreted as “output”; [0050] teaches on the action module determining a weighted score related to detecting a potential health issue using output from the prediction module and further machine learning techniques); and
select a mental health disorder recommendation from a plurality of mental health disorder recommendations based on the second output ([0055] teaches on the system comparing the weighted score (“second output”) with a predetermined threshold; if the threshold is exceeded, the system recommends one or more actions; e.g. identifying and contacting a contact person such as parent, friend sibling to notify them of the health concern; potential further remedial actions can be provided); and
cloud-based input/output circuitry configured to ([0031], [0035], [0082]): transmit, to a second user interface, the mental health disorder recommendation following the conversational interaction ([0055] teaches on a recommendation as to the nature of the possible health concern and potential further remedial actions being provided; a contact person such as parent/guardian, friend, sibling may be notified; the social network could be notified to provide different content to the user – other parties are interpreted as utilizing a different (second) user interface than the first user interface on which the user actions were obtained).
Generoso teaches on using first and second machine learning models, in which output from the first machine learning model is input into a second machine learning model to determine appropriate remedial actions for potential anomalies in a user’s mental health identified from user context, but does not disclose that the second machine learning model is trained to select an emotional state of a selected context. Osotio which is directed to a method of extracting an emotional state of a user from device data, teaches:
wherein a machine learning model is trained to select an emotional state from a plurality of emotional states of the selected context based on [user data], and wherein each emotional state of the plurality of emotional states corresponds to a respective emotional state of the user ([0023], the user’s emotional state is extracted through a personalized emotional state model, created through supervised or unsupervised machine learning; [0045] teaches on the emotional model being created through one or more supervised or unsupervised machine learning processes, each having particular training requirements; initial training can be accomplished either by starting with a standard model over time or by using a batch of collected data; [0064] teaches on emotional engine using data from a variety of sources such as data from a user including profile/user data; the information can also be based on current context – this data is interpreted as user data which is the basis for selecting an emotional state; [0030]-[0038] teach on an example of a team of employees participating in a conference call; [0034] teaches on the system evaluating data captured from team member’s devices, tones of voices, and content of communications to identify participant Max is “frustrated” (emotional); the system also identifies the “various emotional states” of the other team members (plurality of emotional states).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify Generoso with these teachings of Osotio, so that the second machine learning model of Generoso is trained to select an emotional state from a plurality of emotional states of a select context based on the determined context of Generoso (first output), with the motivation of using the emotional state of a user to determine an appropriate action to be performed to change the emotional state of the user (Osotio [0017]).
Generoso/Osotio do not explicitly teach the following. Alikov, which is directed to systems and methods for generating machine searchable keywords, teaches:
select the machine learning model, from a plurality of machine learning models, based on the context selected from the plurality of contexts ([0005] teaches on receiving a context and generating a plurality of fields based on the context; selecting, for each of the plurality of fields, a machine learning model from a plurality of machine learning models (Examiner equates “plurality of fields” with “plurality of contexts” as each field of Alikov has a machine learning model selected, and the instant claim has a machine learning model selected for each context; in Alikov and the instant claim, a particular data label (context/field) of a plurality of data label (contexts/fields) is used as a basis for selecting one machine learning model from a plurality of machine learning models), wherein each context of the plurality of contexts corresponds to a respective machine learning model from the plurality of machine learning models ([0005], a machine learning model is selected from a plurality of machine learning models for each individual field of the plurality of fields, which is interpreted as each field (“context”, as explained above) corresponding to the respective machine learning model that is selected for use with that particular field).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify the combined teachings of Generoso/Osotio with these teachings of Alikov, to select the second machine learning model of Generoso by selecting a machine learning model from a plurality of machine learning models based on the selected context, wherein each context corresponds to a respective machine learning model, with the motivation of using machine learning models that have been trained using data containing relevant information for a specific purpose (Alikov [0080]).
Claim(s) 2, 4-8, 12, 14-18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Generoso et. al. (US Publication 20200020447A1) in view of Osotio et. al. (US Publication 20180101776A1).
Regarding Claim 2, Generoso discloses
receiving a first user action during a conversational interaction with a user interface ([0005] teaches on receiving data related to social media activity for a user; [0015] teaches on ingesting information about a user’s social media activity including frequency of posts, participation in groups, content of user’s posts, frequency of likes on friends’ posts; [0058] further teaches on generating a user analysis profile by using user activity on a social network such as participation in groups and interaction with friends and comments on friends’ posts; [0059] teaches on analyzing “tone of user’s voice in conversations” which indicates conversational interaction; the former examples of interacting with social media are interpreted as “user action” received during “conversational interaction with a user interface” as comments/interactions are inputted via user interface; if the user activity is used to generate a user analysis profile, it is understood that the user actions have been received in order to do so);
in response to receiving the first user action, determining, using control circuitry, a first feature input based on the first user action ([0005] teaches on analyzing the received social media data using one or more machine learning models; [0059] teaches on analyzing social media data to identify recent laughter, positive comments, or negative comments, image and voice recognition; tone of user’s voice in conversations could be analyzed to determine a positive or negative tone/inflection; content of spoken/written conversations could be analyzed – any of the aforementioned responses are interpreted as reading on “first feature input” during the user action of interacting with social media);
inputting, using the control circuitry, the first feature input into a first machine learning model ([0005] teaches on generating a user analysis profile for a user by analyzing the received data using a first machine learning model; per [0015], information about a user’s social media activity (frequency of posts, participation in groups, etc.) can be ingested and classified using a trained ML model to generate a personality profile for the user; [0059], the received social media data includes the first feature input – laughter, positive/negative comments, voice tone in conversations), wherein the first machine learning model is trained to select a context from a plurality of contexts based on user actions (Abstract teaches on using a first machine learning model to generate a user analysis profile; [0048] teaches on machine learning models are supervised models initially generated through a training process; see also [0063] regarding training of ML models; [0058] teaches on the prediction module generating the user analysis profile including a personality profile, which may include data about posting frequency, comments on friends’ posts, interactions with friends, group posts, e.g., user actions; [0059], [0060] respectively teach on the prediction module generating second and third elements of the user analysis profile, a mood score and identified bullying; [0061] teaches on the prediction module generating the fourth element of user analysis, psychological symptoms, by considering data about external events in the user’s life such as death of a loved one, reduction in social network activity, reduction in movement; where [0038] teaches on the prediction module being configured to use machine learning to identify potential anomalies in the user’s health, including the user’s mental health, based on data ingested; anomalies identified are interpreted as a “contexts”) and wherein each context of the plurality of contexts corresponds to a respective emotional state of a user ([0061] teaches on the prediction module identifying symptoms of psychological issues such as anxiety or depression, interpreted as “emotional states”; [0062] provides an example of identifying anger via social media content posted);
receiving, using the control circuitry, a first output from the first machine learning model ([0061] teaches on outputting the results (blocks 504, 506, 508, 510) of the prediction module, interpreted as the first machine learning to the action module per [0038]);
inputting, using the control circuitry, the first output into a second machine learning model, wherein the second machine learning model is trained to [identify potential health anomalies to identify a potential health condition] based on the first output,; (Abstract teaches on a second machine learning model which classifies the user analysis profile to determine a weighted score associated with potential health issues for the user; [0038] teaches on the action module 240 (interpreted as second machine learning model) is configured to use further machine learning to analyze the results of the prediction module (first output from first machine learning model); [0050] teaches on the action module using machine learning techniques to determine a weighted score related to detecting a possible health issue using output from the prediction module; [0051] teaches on the collaboration and deployment driver 242 and analytical behavior module, which are part of the action module 240, taking the identified potential anomalies to identify potential health issue of the user);
receiving, using the control circuitry, a second output from the second machine learning model ([0005] teaches on determining a weighted score associated with detecting a potential health issue for the user by classifying the user analysis profile using a second machine learning model; the weighted score is interpreted as “output”; [0050] teaches on the action module determining a weighted score related to detecting a potential health issue using output from the prediction module and further machine learning techniques);
selecting, using the control circuitry, a mental health disorder recommendation from a plurality of mental health disorder recommendations based on the second output ([0055] teaches on the system comparing the weighted score (“second output”) with a predetermined threshold; if the threshold is exceeded, the system recommends one or more actions; e.g. identifying and contacting a contact person such as parent, friend sibling to notify them of the health concern; potential further remedial actions can be provided); and
generating the mental health disorder recommendation following the conversational interaction ([0055] teaches on determining health anomalies indicate that a concern has been identified; responsively, the system recommends one or more actions; a recommendation as to the nature of the possible health concern and potential further remedial actions can be provided).
Generoso teaches on using first and second machine learning models, in which output from the first machine learning model is input into a second machine learning model to determine appropriate remedial actions for potential anomalies in a user’s mental health identified from user context, but does not disclose that the second machine learning model is trained to select an emotional state of a selected context. Osotio which is directed to a method of extracting an emotional state of a user from device data, teaches:
wherein a machine learning model is trained to select an emotional state from a plurality of emotional states of the selected context based on [user data], and wherein each emotional state of the plurality of emotional states corresponds to a respective emotional state of the user ([0023], the user’s emotional state is extracted through a personalized emotional state model, created through supervised or unsupervised machine learning; [0045] teaches on the emotional model being created through one or more supervised or unsupervised machine learning processes, each having particular training requirements; initial training can be accomplished either by starting with a standard model over time or by using a batch of collected data; [0064] teaches on emotional engine using data from a variety of sources such as data from a user including profile/user data; the information can also be based on current happenings, e.g., context – this data is interpreted as user data which is the basis for selecting an emotional state; [0030]-[0038] teach on an example of a team of employees participating in a conference call and how emotional state is selected; [0032] teaches on the conference call starting off smooth with team agreeing on content (context); [0034] teaches on the system evaluating data captured from team member’s devices later in the conference call, e.g., tones of voices and content of communications, to identify participant Max is “frustrated” (emotional state) in a context of a conference call becoming more serious and tense; the system also identifies the “various emotional states” of the other team members (plurality of emotional states)).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify Generoso with these teachings of Osotio, so that the second machine learning model of Generoso is trained to select an emotional state from a plurality of emotional states of a select context based on the determined context of Generoso (first output), with the motivation of using the emotional state of a user to select an appropriate action to be performed to change the emotional state of the user (Osotio [0017]).
Regarding Claim 4, Generoso/Osotio teach the limitations of Claim 2. The method of claim 2, further comprising:
receiving a second user action during the conversational interaction with the user interface ([0005] teaches on receiving data related to social media activity for a user; [0015] teaches on ingesting information about a user’s social media activity including frequency of posts, participation in groups, content of user’s posts, frequency of likes on friends’ posts; [0058] further teaches on generating a user analysis profile by using user activity on a social network such as participation in groups and interaction with friends and comments on friends’ posts; [0059] teaches on analyzing “tone of user’s voice in conversations” which indicates conversational interaction; the former examples of interacting with social media are interpreted as “user action” received during “conversational interaction with a user interface” as comments/interactions are inputted via user interface; if the user activity is used to generate a user analysis profile, it is understood that the user actions have been received in order to do so); in response to receiving the second user action, determining a second feature input for the first machine learning model based on the second user action ([0005] teaches on analyzing the received social media data using one or more machine learning models; [0059] teaches on analyzing social media data to identify recent laughter, positive comments, or negative comments, image and voice recognition; tone of user’s voice in conversations could be analyzed to determine a positive or negative tone/inflection; content of spoken/written conversations could be analyzed – any of the aforementioned responses are interpreted as reading on “first feature input” during the user action of interacting with social media); inputting the second feature input into the first machine learning model ([0005] teaches on generating a user analysis profile for a user by analyzing the received data using a first machine learning model; per [0015], information about a user’s social media activity (frequency of posts, participation in groups, etc.) can be ingested and classified using a trained ML model to generate a personality profile for the user; [0059], the received social media data includes the first feature input – laughter, positive/negative comments, voice tone in conversations); receiving a different output from the first machine learning model, wherein the different output corresponds to a different context from the plurality of contexts ([0061] teaches on outputting the results (blocks 504, 506, 508, 510) of the prediction module, interpreted as the first machine learning to the action module per [0038]); and inputting the different output into the second machine learning model (Abstract teaches on a second machine learning model which classifies the user analysis profile to determine a weighted score associated with potential health issues for the user; [0038] teaches on the action module 240 (interpreted as second machine learning model) is configured to use further machine learning to analyze the results of the prediction module (first output from first machine learning model).
While Examiner has provided citations above, Examiner notes that the limitations of claim 4 mirror those recited in previous claim 2 except for performing each function a second time, e.g. receiving a second user action, determining a second feature input to input into the first machine learning model, to receive a different output corresponding to a different context than what was received in Claim 2, for the second input of Claim 4. Generoso discloses obtaining data of a user interacting with social media ([0057]-[0060], e.g., user actions), and discloses determining different user actions at para. [0059], e.g., laughter/positive comments vs. negative comments which are interpreted as first and second user actions, respectively; As the system of Generoso uses the different user actions in the first machine learning model which predicts whether the user is experiencing a possible mental health issue (output) based on the user actions fed into the model, it is interpreted that this influences the output as laughter leads to a higher mood score while a negative comment could indicate a higher bullying score per [0063]). The limitations of Claim 4 amount to duplication of parts, which has no patentable significance unless a new and unexpected result is produced, see MPEP 2144.04(VI)(B). Claim 4 is merely repeating the same steps as Claim 2 with a different input to start the process, and does not yield a new or unexpected result.
Regarding Claim 5, Generoso/Osotio teach the limitations of Claim 2. Generoso further discloses wherein the first machine learning model is a supervised machine learning model, and wherein the second machine learning model is a supervised machine learning model ([0048], the one or more machine learning models are supervised learning models).
Regarding Claim 6, Generoso/Osotio teach the limitations of claim 2. Generoso does not disclose, but Osotio further teaches wherein the first machine learning model is a support vector machine classifier ([0023] teaches on a support vector machine (SVM) technique), and wherein the second machine learning model is an artificial neural network model ([0023] teaches on convolutional neural network and deep neural network; both of which are types of artificial neural networks).
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of Generoso to incorporate SVM and ANN models as taught by Osotio as the first and second machine learning models, since the claimed invention is only a combination of these old and well known elements which would have performed the same function in combination as each did separately. In the present case Generoso already discloses using supervised machine learning models ([0048]). Incorporating different machine learning models such as SVM or neural networks as taught by Osotio would perform their same functions within the system of Generoso, as they are merely different types of machine learning models, making the results predictable to one of ordinary skill in the art (MPEP 2143).
Regarding Claim 7, Generoso/Osotio teach the limitations of Claim 2. Generoso further discloses wherein the first feature input is a vectorization of a conversational detail or information from a user account of the user ([0044] teaches on a real-time analytics engine which can process data to put it in a more suitable form for use by the prediction module (first machine learning model); the real time analytics engine can generate feature vectors related to the input data; in this embodiment, a feature vector is a vector having any number of dimensions which represents a numerical representation of the associated electronic data; this “vectorization” of electronic data may improve the speed, efficiency and/or accuracy of the system).
Regarding Claim 8, Generoso/Osotio teach the limitations of claim 2. Generoso further discloses: further comprising: receiving a first labelled feature input, wherein the first labelled feature input is labelled with a known context for the first labelled feature input ([0063] teaches on generating and updating a supervised machine learning model; supervised/trained ML model refer to machine learning that utilizes exemplars and predetermined attribute scores to train the model (interpreted as labeled feature input with known context); a corpus of training data is converted to feature vectors with associated training data); and training the first machine learning model to classify the first labelled feature input with the known context ([0063], teaches on the model training component uses supervised machine learning techniques to generate and update a trained machine learning model to process new electronic data using classification techniques).
Regarding Claim 12, Generoso/Osotio teach the limitations of claim 1. Claim 12 recites limitations that are the same or substantially similar to Claim 1, and the discussion above with respect to Claim 1 is equally applicable to Claim 12. Claim 12 is rejected for the same reasons as Claim 1. Claim 12 additionally recites the following, which are also taught by Generoso: a non-transitory computer-readable medium comprising of instructions that are executed by one or more processors ([0020]-[0025]).
Regarding Claim 14, Generoso/Osotio teach the limitations of claim 4. Claim 14 recites limitations that are the same or substantially similar to Claim 4, and the discussion above with respect to Claim 4 is equally applicable to Claim 14. Claim 14 is rejected for the same reasons as Claim 4.
Regarding Claim 15, Generoso/Osotio teach the limitations of claim 5. Claim 15 recites limitations that are the same or substantially similar to Claim 5, and the discussion above with respect to Claim is equally applicable to Claim 15. Claim 15 is rejected for the same reasons as Claim 5.
Regarding Claim 16, Generoso/Osotio teach the limitations of claim 6. Claim 16 recites limitations that are the same or substantially similar to Claim 6, and the discussion above with respect to Claim 6 is equally applicable to Claim 16. Claim 16 is rejected for the same reasons as Claim 6.
Regarding Claim 17, Generoso/Osotio teach the limitations of claim 7. Claim 17 recites limitations that are the same or substantially similar to Claim 7, and the discussion above with respect to Claim 7 is equally applicable to Claim 17. Claim 17 is rejected for the same reasons as Claim 7.
Regarding Claim 18, Generoso/Osotio teach the limitations of claim 8. Claim 18 recites limitations that are the same or substantially similar to Claim 8, and the discussion above with respect to Claim 8 is equally applicable to Claim 18. Claim 18 is rejected for the same reasons as Claim 8.
Claim(s) 3 and 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over Generoso et. al. (US Publication 20200020447A1) in view of Osotio et. al. (US Publication 20180101776A1) as applied to Claim 2 and 12 respectively above, and further in view of Alikov et. al. (US Publication 20220215452A1).
Regarding Claim 3 and Claim 13, Generoso/Osotio teach the limitations of Claims 2 and 12, respectively, but do not explicitly teach the following. Alikov, which is directed to systems and methods for generating machine searchable keywords, teaches: further comprising of selecting the machine learning model, from a plurality of machine learning models, based on the context selected from the plurality of contexts ([0005] teaches on receiving a context and generating a plurality of fields based on the context; selecting, for each of the plurality of fields, a machine learning model from a plurality of machine learning models (Examiner equates “plurality of fields” with “plurality of contexts” as each field of Alikov has a machine learning model selected, and the instant claim has a machine learning model selected for each context; in Alikov and the instant claim, a particular data label (context/field) of a plurality of data label (contexts/fields) is used as a basis for selecting one machine learning model from a plurality of machine learning models), wherein each context of the plurality of contexts corresponds to a respective machine learning model from the plurality of machine learning models ([0005], a machine learning model is selected from a plurality of machine learning models for each individual field of the plurality of fields, which is interpreted as each field (“context”, as explained above) corresponding to the respective machine learning model that is selected for use with that particular field).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify the combined teachings of Generoso/Osotio with these teachings of Alikov, to select the second machine learning model of Generoso by selecting a machine learning model from a plurality of machine learning models based on the selected context, wherein each context corresponds to a respective machine learning model, with the motivation of using machine learning models that have been trained using data containing relevant information for a specific purpose (Alikov [0080]).
Claim(s) 9 and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over
Generoso et. al. (US Publication 20200020447A1) in view of Osotio et. al. (US Publication 20180101776A1) as applied to Claims 2 and 12 respectively above, and further in view of Pourmohammad et. al. (US Publication 20190096217A1).
Regarding Claim 9 and Claim 19, Generoso/Osotio teach the limitations of claims 2 and 12, respectively, but do not teach the following. Pourmohammad, which is directed to a risk management system, teaches
wherein the first feature input is a vectorization of an n-grams corresponding to [text] ([0271] teaches on an NLP engine vectorizing extracted n-grams (unigrams and bigrams); the extracted n-grams can be vectorized in a high-dimensional vector space), (Per claim construction “or”, the second limitation is not required).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify the combined teachings of Generoso/Osotio with these teachings of Pourmohammad, to use a feature input of a vectorization of n-grams corresponding to the first user action as taught by Generoso, with the motivation of enabling the system to work with numbers instead of words and indicating the frequency at which words occur (Pourmohammad [0271]).
Claim(s) 10, 11, 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over
Generoso et. al. (US Publication 20200020447A1) in view of Osotio et. al. (US Publication 20180101776A1) as applied to Claim 2 and 12 respectively above, and further in view of Le et. al. (US Publication 20220091713A1).
Regarding Claim 10, Generoso/Osotio teach the limitations of claim 2. Generoso further discloses determining user information corresponding to the mental health disorder recommendation ([0055] teaches on determining a possible health anomaly, which per [0038] is understood to include mental health; [0055], when a pre-determined threshold is exceeded, the system recommends one or more remedial actions, e.g., a contact person is notified about the potential health concern; [0056] teaches on sending therapeutic messages to the user when the threshold is exceeded or notifying the social network to modify the user’s experience to avoid problematic people/content or block posts/content which could be potentially problematic).
Generoso/Osotio do not teach the following, but Le, which is directed to generating dynamic interface options using machine learning models, teaches:
determining a network location of the user information ([0065] teaches on determining a network location of the user information) and generating a network pathway to the user information ([0066] teaches on generating a network pathway to the user information).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify the combined teachings of Generoso/Osotio with these teachings of Le, to determine a network location of the user information and generate a network pathway to the user information, with the motivation of using the determined network location of the user information for facilitating a modification of the user’s account (Le [0065] and to quickly and efficiently retrieve the user information (Le [0062]).
Regarding Claim 11, Generoso/Osotio teach the limitations of claim 2. Generoso/Osotio do not teach the following, but Le, which is directed to generating dynamic interface options using machine learning models, teaches:
further comprising: automatically retrieving the user information from the network location ([0067] teaches on automatically retrieving the user information from the network location in response to the user action); and
generating for display the user information on a second user interface ([0068] teaches on generating for display the user information on a second user interface).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify the combined teachings of Generoso/Osotio with these teachings of Le, to automatically retrieve the user information from the network location based on the mental health disorder recommendation as taught by Generoso, and generate for display the user information on a second user interface with the motivation of presenting the retrieved user information to a second user interface located remotely from a first user interface (Le [0068]).
Regarding Claim 20, Generoso/Osotio/Le teach the limitations of Claims 10 and 11. Claim 20 recites the same or substantially similar limitations as Claims 10 and 11 combined, and the discussions above with respect to Claims 10 and 11 are equally applicable to Claim 20. Claim 20 is rejected for the same reasons as Claim 10 and 11.
Claim(s) 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over
Generoso et. al. (US Publication 20200020447A1) in view of Osotio et. al. (US Publication 20180101776A1) as applied to Claim 2 above, and further in view of Shriberg et. al. US Publication 20200118458A1).
Regarding Claim 22, Generoso/Osotio teach the limitations of Claim 2 but do not teach the following. wherein the second machine learning model is configured to output a probability score for each emotional state of the plurality of emotional states ([0347] teaches on a model providing a score, typically a posterior probability over a set of predetermined emotions that describes how well a sample matches pre-trained models for each of the said amotions), and wherein selecting the mental health disorder recommendation is based on a probability distribution over the plurality of emotional states ([0329] teaches on a raw score output by a machine learning algorithm; if the score is a probability near 1, the classification module may apply a “severe” label to the score; [0421] teaches on the scores allowing clinicians to determine severities of one or more mental states of the patient, e.g., mild, moderate or severe depression).
It would have been obvious to one of ordinary skill in the art at the time the invention was filed, to further modify Generoso/Osotio with these teachings of Shriberg, to use a probability score for emotional states for selecting a mental health disorder recommendation, with the motivation of better allowing a clinician to determine the extent of the patient’s condition (Shriberg [0421]).
Claim(s) 23 is/are rejected under 35 U.S.C. 103 as being unpatentable over
Generoso et. al. (US Publication 20200020447A1) in view of Osotio et. al. (US Publication 20180101776A1) as applied to Claim 2 above, and further in view of Hissoiny (US Publication 20190175952 A1).
Regarding Claim 23, Generoso/Osotio teach the limitations of Claim 2. Osotio further teaches further comprising: updating one or more parameters of the second machine learning model ([0018] teaches on using implicit and/or explicit feedback to refine the emotional state model; see also [0048], [0061], [0074]).
It would have been obvious to one of ordinary skill in the art at the time the invention was filed, to further modify Generoso/Osotio with these teachings of Osotio, to update parameters of the second ML model based on differences between predicted and labeled states during training, with the motivation of refining the system’s ability to predict/appropriately respond to an extracted emotional state (Osotio [0018]).
Generoso/Osotio do not teach the following, but
updating a machine learning model using backpropagation ([0057] teaches on updating parameters of a model using backpropagation).
It would have been obvious to one of ordinary skill in the art at the time the invention was filed, to further modify Generoso/Osotio with these teachings of Hissoiny, to update the model parameters of Generoso/Osotio using backpropagation, with the motivation of updating the model based on corresponding errors (Hissoiny [0057]).
Subject Matter Free of the Prior Art
Claim 21 recites wherein the second machine learning model comprises: a shared encoder jointly trained across (i) a categorical emotional-state framework and (ii) a dimensional valence-arousal-dominance emotional framework; and a plurality of separate prediction heads corresponding respectively to the categorical emotional- state framework and the dimensional valence-arousal-dominance emotional framework.
A search of publicly available prior art fails to yield a reference or combination of references that would make the claimed combination of elements obvious when considered as a whole. Claim 21 is free of the prior art.
Response to Applicant’s Remarks/Arguments
Please note: When referencing page numbers of Applicant’s response, references are to page numbers as printed.
Claim Objections
The objections to Claims 1 and 9 for minor informalities are withdrawn in view of Applicant’s amendments to these claims.
35 USC 101 Rejections
Applicant’s remarks regarding the rejections of all pending claims under 35 USC 101 have been considered but are not persuasive. Applicant argues:
(a) The pending independent claims recite a specific, multi-tier machine-learning pipeline
that generates a vectorized feature input, uses a first machine-learning model to select a context from a plurality of contexts, uses that selected context to select a second-tier machine-learning model, optionally from a plurality of models, and uses the second-tier output to generate a mental health disorder recommendation. This is not mere mental analysis or an abstract idea performed on generic computers. Rather, the claims are directed to a concrete restructuring of how machine-learning models are trained and how inference is performed within a hierarchical architecture (page 9)
Regarding (a), Examiner respectfully submits that the abstract idea was not categorized as “mental processes”; it was categorized as certain methods of organizing human activity (see Non-Final Action dated 08/26/25 at page 5). The argument with respect to “mental analysis” is therefore not persuasive.
Regarding “abstract idea performed on a computer”, Examiner submits that the instant claims are directed to an abstract idea, specifically, certain methods of organizing human activity. MPEP 2106. 04(a)(2)(II) states that a claimed invention is directed to certain methods of organizing human activity if the identified claim elements contain limitations that encompass fundamental economic principles or practices, commercial or legal interactions, or managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions). The Examiner submits that the identified claim elements represent person behaviors, or a series of rules or instructions that a person or persons, with or without the aid of a computer, would follow to collect and process data associated with a patient using a series of models to select and provide a mental health disorder recommendation for the patient. As such, the claimed invention is directed to an abstract idea. This argument is not persuasive.
(b) The claims are not directed to a mental process. Although the claimed system ultimately generates a recommendation related to emotional state, the claims require computational operations that cannot be performed in the human mind … These operations are implemented in high-dimensional vector space using trained model parameters and probabilistic inference mechanisms that are not practically performable in the human mind. The claims therefore are not directed to a mental process but to a computer-implemented machine-learning architecture. (page 10)
Regarding (b), Examiner respectfully submits that the abstract idea was not categorized as “mental processes”; it was categorized as certain methods of organizing human activity (see Non-Final Action dated 08/26/25 at page 5); see preceding remarks regarding argument (a). Therefore, this argument is not persuasive. However, Examiner provides the following additional remarks: The abstract idea has been categorized as certain methods of organizing human activity. Per MPEP 2106.04(a)(2)(II), this grouping may encompass an individual interacting with a computer. Examiner submits that Applicant’s invention utilizes known methods of machine learning to apply data to an algorithm and output a result, which only amounts to mere instructions to apply the abstract idea. Examiner submits that “vectorization” falls within the scope of the abstract idea. Examiner submits that using a first machine learning model to partition feature space into one of a plurality of concepts only amounts to applying the abstract idea on a computer, e.g., using the machine learning model on a computer to partition feature space into a particular concept. Similarly, selecting a context-dependent second model and generation of probabilistic outputs falls within the scope of the abstract idea; recitation of “machine learning model” only amounts to using a computer/machine learning model as a tool to apply data to an algorithm and output a result. The same rationale applies to remarks pertaining to backpropagation and neural network inferences; Examiner submits that these are known concepts and Applicant is merely using them as a tool to output a result.
Regarding remarks pertaining to “Even if the claims were understood to involve mathematical concepts such as probability calculations or vector transformations, those concepts are integrated into a practical application under Step 2A, Prong 2, Examiner submits that no practical application is present. Applicant asserts, “User inputs regarding mental health are fluid, inconsistent, ambiguous, and not easily mapped to standardized diagnostic taxonomies”. However, the problem of user inputs being fluid, inconsistent, ambiguous and not easily mapped to standardized diagnostic taxonomies are not a problem caused by the technological environment of the claim (a general purpose computer). Examiner submits that implementation of machine learning models only amounts to mere instructions to apply the abstract idea; the steps of partitioning contexts and then constraining emotional state inference within the selected context falls within the scope of the abstract idea as these are steps that could be performed by a healthcare provider. Examiner submits that using two models in series, which are specialized to perform specific tasks, falls within the scope of an improvement to the abstract idea itself and is not a technological improvement per MPEP 2106.05(a) which states “It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements.” Applicant has not provided, nor can Examiner find evidence of, how any of the additional elements identified above in main 101 analysis section are providing an improvement over prior art systems. Therefore, this argument is not persuasive.
(c) Even if the claims were understood to involve mathematical concepts such as probability calculations or vector transformations, those concepts are integrated into a practical application under Step 2A, Prong 2 ... These disclosures reinforce that the invention is not an abstract idea divorced from implementation, but rather a technical improvement in how machine learning systems operate. (page 10-11)
Regarding (c), the Examiner respectfully disagrees. As discussed above with respect to (b), the problem of user inputs being fluid, inconsistent, ambiguous and not easily mapped to standardized diagnostic taxonomies are not a problem caused by the technological environment of the claim (a general purpose computer). Examiner submits that implementation of machine learning models only amounts to mere instructions to apply the abstract idea; the steps of partitioning contexts and then constraining emotional state inference within the selected context falls within the scope of the abstract idea as these are steps that could be performed by a healthcare provider. Examiner submits that using two models in series, which are specialized to perform specific tasks, falls within the scope of an improvement to the abstract idea itself and is not a technological improvement per MPEP 2106.05(a) which states “It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements.” Applicant has not provided, nor can Examiner find evidence of, how any of the additional elements identified above in main 101 analysis section are providing an improvement over prior art systems. Therefore, this argument is not persuasive.
(d) This analysis is consistent with the precedential decision in Ex parte Desjardins … Here, as in Desjardins, the claims recite a specific machine-learning architecture with defined interactions between model tiers, context dependent model selection, probabilistic outputs, and parameter updating during training. The multi-tier architecture improves internal model stability and classification precision under ambiguous input conditions. The mathematical operations are not claimed in the abstract but are integrated into a structured pipeline that improves the performance and reliability of the computer system (page 11)
Regarding (d), The Examiner respectfully submits that there is no improvement to machine learning. Initially, the Examiner notes that Ex parte Desjardins does not represent a substantive change in subject matter eligibility analysis; there has been no indication by the Office that this decision impacts the how claims involving machine learning are analyzed at the Examiner level. The decision is specific to the facts before the Appeals Review Panel and follows the subject matter eligibility analysis set forth in MPEP 2106. As found by the Panel, the claimed “training strategy allows the model to preserve performance on earlier tasks even as it learns new ones, directly addressing the technical problem of 'catastrophic forgetting' in continual learning systems" represents “technical improvements over conventional systems by addressing challenges in continual learning and model efficiency by reducing storage requirements and preserving task performance across sequential training.” This analysis represents implementation of the practical application-“improvement” analysis of MPEP 2106.04(d)(I) to the facts before the Panel.
Applicant’s claims do not provide such an improvement. As an initial matter, Examiner notes that the claim in Desjardins is specifically directed a method of training machine learning models, e.g., see preamble of Claim 1 in Desjardins. The instant claims are directed to a system, method and non-transitory CRM for “generating mental health disorder recommendations”. The machine learning models in the instant claims are merely tools by which to arrive at the output, e.g., mental health disorder recommendations. Applicant has not cited to, nor can examiner find, evidence of any training steps that provide an improvement over the prior art that are analogous to Desjardins. Applicant has not cited to, nor can Examiner find, an indication in the Specification that the claimed invention provides an improvement as to how a model is trained. Improving the accuracy of a mental health disorder recommendation by using multi-tier models rather than a singular model is not an improvement to how the model is trained within the meaning of Desjardins. Put another way, the particular way the machine learning model of applicant’s invention uses the data to train itself is not improved, which is the holding of Desjardins. Applicant is merely improving the accuracy of model output (e.g., an improved mental health recommendation) by optimizing the data selected and to be input to the second tier model. Improving the accuracy of a model based on the data used is not an improvement by any measure in MPEP 2106.
Examiner’s position is also supported by the decision in Recentive Analytics, Inc. v. Fox Corp. Recentive held that non-specifically claimed training of an AI/ML algorithm is insufficient to provide a practical application or significantly more because it does not result in “improving the mathematical algorithm or making machine learning better.” Recentive at 12. The decision further instructed that “[i]terative training using selected training material…are incident to the very nature of machine learning” and thus does not provide for an improvement. Recentive at 12.
This argument is not persuasive.
(e) Even assuming arguendo that the claims were found to recite an abstract idea, they
include significantly more than generic computer implementation under Step 2B (page 11)
Regarding (e), the Examiner respectfully disagrees. Examiner maintains the position that the computer system and machine learning models are recited merely as tools to perform the steps (e.g., vectorizing inputs, selecting contexts, selecting a second model based on contexts, generating probabilistic emotional-state outputs to produce recommendations). Regarding reference to joint training, probability-based output layers, and backpropagation-based parameter updates, Examiner respectfully submits that as discussed in the specification, these are known methods, and as such, only amounts to mere instructions to apply the abstract idea on a computer.
Regarding remarks directed to Desjardins, Examiner respectfully disagrees that enabling a neural network to retain performance across multiple tasks without catastrophic forgetting is analogous to the instant invention. Applicant has not cited to evidence in the instant specification to support this position.
Regarding remark to “ordered combination of elements (such as vectorization, contextual partitioning, dynamic model selection, probabilistic inference, and recommendation generation) amounts to significantly more than any alleged abstract idea”, Examiner submits that Applicant has identified elements falling within the scope of the abstract idea. The “ordered combination” pertains to additional elements. An “ordered combination” of steps/limitations isn’t sufficient to integrate the judicial exception into a practical application or amount to significantly more the elements are all within the abstract idea and merely implemented by generic computing elements or software. 2106.05(I)(B).
These arguments are therefore unpersuasive.
Regarding remarks directed to new claims 21-23 at the bottom of page 12:
Regarding Claim 21 and features cited by Applicant, para. [0074] of instant specification discloses “For example, the system may use an encoder described above, though there are now two separate prediction heads for each type of emotional framework being predicted. This model therefore benefits from having a shared component with parameters that are trained jointly on both frameworks, and separate components with parameters that are specific to each task as well. This often improves performance relative to training on a single task alone, and aids in generalisation” (Emphasis Examiner). Examiner submits that as Applicant has disclosed “this often improves performance…”, this limitation only amounts to mere instructions to apply the abstract idea, e.g., Applicant is using a known method of joint training which is often known to improve performance. This only amounts to mere instructions to applying the abstract idea on a computer and is not sufficient to integrate the judicial exception into a practical application. Regarding remarks to Desjardins, this has previously been addressed; see above remarks directed to Desjardins. This argument is not persuasive.
Regarding Claim 22 and features cited by Applicant, this only amounts to mere instructions to apply the abstract idea on a computer. Outputting probability scores and making recommendations based on the resulting probability distribution fall within the scope of the abstract idea, e.g., these are steps that may be performed by a healthcare provider. Any purported improvements may be an improvement to the abstract idea itself of making a better mental healthcare recommendation, but are not technological improvements per MPEP 2106.05(a) as discussed above. This argument is not persuasive.
Regarding Claim 23 and features cited by Applicant, this only amounts to mere instructions to apply the abstract idea on a computer. Backpropagation is a known algorithm used in machine learning; Applicant has not invented a new or improved method of training a machine learning model. Applicant’s remark that it “is not a mental process” is not persuasive as the abstract idea was categorized as certain methods of organizing human activity, not as mental processes. This argument is not persuasive.
For all of the above reasons, Applicant’s arguments are not persuasive.
The rejections of Claims 1-14, 18-23 under 35 USC 101 are maintained.
35 USC 103 Rejections
Regarding the rejections under 35 USC 103, Applicant’s remarks have been fully considered but are not persuasive. Examiner has considered Applicant's explanation of the virtues of their invention at pages 15-16; however, while they are informative, they do not represent arguments applicable to the current rejection. The actual arguments that the Examiner can identify are addressed below.
Applicant traverses the rejection for at least the reason that the references relied upon in the Office Action, even if properly combinable as alleged, do not teach or suggest each and every feature of the claimed invention. In particular, even if the cited references generally discuss emotion detection systems or recommendation engines, the pending claims require a specific multi-tier machine-learning architecture that is neither disclosed nor suggested by the cited art, individually or in combination (bottom of page 14-top of page 15)
Regarding (a), Applicant has not provided specific remarks/arguments directed to specific limitations and the applied references, and will address this argument as best understood. Examiner respectfully disagrees with Applicant’s position, and submits that the combination of Generoso/Osotio/Alikov as presented in the non-final action teaches on the multi-tier machine learning architecture of the instant claim. Generoso teaches on using first and second machine learning models, in which a first machine learning model is used to receive a feature input and provide output that is indicative of a selected context of a plurality of contexts; this output from the first machine learning model is input into a second machine learning model to obtain an output to select a mental health disorder recommendation – e.g., multi-tier machine learning architecture. While Generoso does not teach on the particulars of the instant claim pertaining to how the second machine learning model is trained, Osotio teaches on the particulars of a machine learning model being trained to select an emotional state from a plurality of emotional states of the selected context, wherein each emotional state of the plurality of states corresponds to a respective emotional state of a user. Generoso/Osotio do not teach on how a second model is selected, but Alikov teaches on selecting a machine learning model based on a particular context – e.g., receiving a context and generating a plurality of fields based on the context; selecting, for each field, which is generated from a context, a ML model from a plurality of ML models. Examiner submits that modifying Generoso/Osotio’s multi-tier machine learning architecture with Alikov’s teachings of selecting a machine learning model based on a generated context teaches on the claimed multi-tier ML architecture of the instant claim. This argument is not persuasive.
The references relied upon in the Office Action, even if they describe general emotion detection or recommendation generation, do not teach or suggest this hierarchical routing constraint (page 15)
Regarding (b), Applicant has not provided specific arguments directed to particular limitations and applied references, and as such, will address this argument as best understood. Examiner respectfully disagrees with Applicant’s position, and submits that the combination of Generoso/Osotio/Alikov as presented in the non-final action teach on a “hierarchical routing constraint”. Generoso teaches on using first and second machine learning models, in which a first machine learning model is used to receive a feature input and provide output that is indicative of a selected context of a plurality of contexts; this output from the first machine learning model is input into a second machine learning model to obtain an output to select a mental health disorder recommendation – e.g., multi-tier machine learning architecture. While Generoso does not teach on the particulars of the instant claim pertaining to how the second machine learning model is trained, Osotio teaches on the particulars of a machine learning model being trained to select an emotional state from a plurality of emotional states of the selected context, wherein each emotional state of the plurality of states corresponds to a respective emotional state of a user. Generoso/Osotio do not teach on how a second model is selected, but Alikov teaches on selecting a machine learning model based on a particular context – e.g., receiving a context and generating a plurality of fields based on the context; selecting, for each field, which is generated from a context, a ML model from a plurality of ML models. Examiner submits that modifying Generoso/Osotio’s multi-tier machine learning architecture with Alikov’s teachings of selecting a machine learning model based on a generated context teaches on the claimed multi-tier ML architecture with a “hierarchical routing constraint” as the output of the first ML model is used to select a second ML model to utilize to produce an output. This argument is not persuasive.
The references relied upon in the Office Action do not disclose, teach or suggest this two-stage constraint in which context selection governs both the candidate emotional-state set and, optionally, the selection of the second-tier model itself (page 15)
Regarding (c), Applicant has not provided specific remarks/arguments directed to specific limitations and the applied references; as such, Examiner will address as best understood. Examiner respectfully disagrees with Applicant’s position, and submits that the combination of Generoso/Osotio/Alikov as presented in the non-final action teaches on the multi-tier machine learning architecture with a two-stage constraint. Generoso teaches on using first and second machine learning models, in which a first machine learning model is used to receive a feature input and provide output that is indicative of a selected context of a plurality of contexts (first stage constraint); this output from the first machine learning model is input into a second machine learning model to obtain an output to select a mental health disorder recommendation. While Generoso does not teach on the particulars of the instant claim pertaining to how the second machine learning model is trained, Osotio teaches on the particulars of a machine learning model being trained to select an emotional state from a plurality of emotional states of the selected context, wherein each emotional state of the plurality of states corresponds to a respective emotional state of a user. Generoso/Osotio do not teach on how a second model is selected, but Alikov teaches on selecting a machine learning model based on a particular context – e.g., receiving a context and generating a plurality of fields based on the context; selecting, for each field, which is generated from a context, a ML model from a plurality of ML models – selection of the second tier model itself (second stage constraint). Examiner submits that modifying Generoso/Osotio’s multi-tier machine learning architecture with Alikov’s teachings of selecting a machine learning model based on a generated context teaches on the claimed multi-tier ML architecture with a two-stage constraint. This argument is not persuasive.
Regarding the rejection of dependent claims, the Applicant has not offered any arguments with respect to these claims. As such, the rejection of these claims is also maintained.
The rejections of Claims 1-14, 18-23 under 35 USC 103 are maintained.
Conclusion
Examiner respectfully requests that Applicant provides citations to relevant paragraphs of specification for support for amendments in future correspondence.
The following relevant prior art not cited is made of record:
US Publication 20210133509A1, teaching on model optimization and data analysis using machine learning techniques for identifying and treating behavioral, neurological or mental health conditions or disorders
US Publication 20210265064A1, teaching on a system and method for providing model-based predictions of quality of life implications of a treatment via individual-specific machine learning models
US Publication 20200205709A1, teaching on a system for generating a mental state indicator for use in identifying a mental state of a subject by applying machine learning
THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANNE-MARIE K ALDERSON whose telephone number is (571)272-3370. The examiner can normally be reached on Mon-Fri 9:00am-5:00pm EST, and generally schedules interviews in the timeframe of 2:00-5:00pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Fonya Long, can be reached on 571-270-5096. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ANNE-MARIE K ALDERSON/Primary Examiner, Art Unit 3682