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 .
General Remarks
This communication is considered fully responsive to Applicant’s filing filed 02/17/2026.
Application filed: 01/31/2024
Applicant’s PgPUB: 2025/0245116
Claims:
Claims 1, 3, 5, 6, 8-18 and 21 are pending.
Claims 1, 17 and 21 are independent.
Claims 1 and 17 are amended.
Claims 8-13 dependency have changed due to claim 7 being canceled. Dependencies changed to claim 1.
Claims 2, 4, 7, 19 and 20 are canceled.
Claim 21 is new.
IDS:
Previous IDS:
IDS filed 05/07/2025 has been considered.
CN110955774 not provided.
International Search Report not provided.
Note: these references are provided in the PCT application (PCT/US25/14108).
Continuity/Priority Data:
International Patent Application PCT/US2025/014108 filed 01/31/2025 claims priority to this Application.
Response to Arguments
Applicant's arguments filed 02/17/2026 have been fully considered but they are not persuasive.
Applicant argues the cited prior art does not teach a validation test.
Examiner respectfully disagrees with Applicant’s assertions.
Regarding Ma:
Ma teaches running a variety of analysis (i.e., test) and then making comparisons to a threshold (i.e., validation). For example, paragraph 0136 of Ma states, “… by analyzing many traces to identify event sequences that are frequent (e.g. occurring at least as frequently as a minimum frequency threshold). If a number of identified event subsequences meets or exceeds the minimum frequency threshold”. This demonstrates a validation test by making an analysis and making a comparison with a threshold.
Regarding Kim:
Kim teaches running a variety of analysis (i.e., test) and then making comparisons to a threshold (i.e., validation). For example, paragraph 0057 of Kim states, “… The clustered events may then be analyzed to identify sub-tasks in the clustered events by identifying sequences of one or more clustered events (e.g., sequences of 1, 2, 3, 4, 5, or 6 clustered events) that repeat a threshold number of times in the computer usage data”. This demonstrates a validation test by making an analysis and making a comparison with a threshold.
Regarding Wilson:
Wilson teaches running hypothesis testing and making comparisons to thresholds. The Abstract of Wilson states, “event features. Clusters of related events are determined, and a state automaton is employed to determine a strength of temporal “bursty” activity for each cluster. Hypothesis testing is performed on each cluster to determine a likelihood that the cluster is a temporally emergent cluster. Clusters with a bursting likelihood above a threshold are determined to be an emergent cluster associated with an anomalous issue”. This demonstrates a validation test by making an analysis and making a comparison with a threshold.
Each cited priority reference teaches a form of a validation test. Ma, as shown in rolled-up claim 2, covers the specifics of the validation test as stated in the claims. As such, the teachings of the cited prior art covers the claim language as they are currently stated.
Claim Objections
Claim 21 is objected to because of the following informalities:
Claim 21 states in part, “ … a count matrix representing counts between one activity doe a given resource;” The word “doe” appears to not be intended. Examiner will interpret “doe” as “for” for the purposes of this Office Action.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
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 may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
Claims 1, 8-12, 17, 18 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication No. 2022/0171988 A1 to Ma et al. (“Ma”) in view of U.S. Patent Application Publication No. 2018/0113781 A1 to Kim et al. (“Kim”) in further view of U.S. Patent Application Publication No. 2020/0314127 A1 to Wilson et al. (“Wilson”).
As to claim 1, Ma discloses:
a method of task identification in a workforce (¶0065 – Ma teaches (a computer-implemented method of identifying one or more processes for robotic automation (RPA)), comprising:
gathering time series data of resources, computer apps, and computer screens for desktop activities of a workforce (¶0022 Ma teaches a plurality of event streams, each event stream corresponding to a human user interacting with a computing device to perform one or more tasks, are recorded in operation 302; ¶0072 – Ma teaches recording a plurality of event streams constitutes gathering time series data of resources, computer apps, and computer screens for desktop activities of a workforce);
performing pre-processing of time series data, including filtering the time series data (the event recording function may be configured, e.g. in a configuration file, to filter different types of events, time periods, applications, etc. that may be relevant or irrelevant to a given type of task. In this manner, the recording function may be tailored to facilitate recording only events relevant to a given task, or at least reduce the proportion of irrelevant events included in an overall event stream, para 0103, [filtering different types of events constitutes performing pre-processing including filtering]; Cleaning essentially seeks to reduce the data set by eliminating irrelevant and/or misleading information from the event streams, para 0111) and performing at least one data transformation operation to convert the time series data into a data representing desktop activities, distances between activities, and counts (¶0021-¶0022 – Ma teaches a "task" in the context of robotic process automation (RPA) is to be understood as referring to an objective typically performed by a human interacting with a computing device such as a desktop, laptop, tablet, smartphone, etc. that must be accomplished on a recurring basis. Exemplary "tasks" in accordance with the present descriptions may include, without limitation, data entry, customer service interactions (e.g. chat sessions, call center interactions), business transactions (e.g. invoice processing, product acquisition, sales, new account processing, etc.), performing processing workflows of any type (e.g. image processing, data analytics, statistical analyses, etc.) or any other type of human/computer interaction that may be represented by a human interacting with one or more user interfaces (Uls) of a computing device; Fig. 3, ¶0027 – Ma teaches preferably based on a critical, often minimal, set of interactions, "traces" are built, in accordance with the inventive concepts described herein. A "trace" is to be understood as a segment that accomplished a particular task in a particular instance (e.g., tasks and traces constitute desktop activities); Fig. 3, ¶0123 – Ma shows each recorded event stream is concatenated. The concatenation process preferably also includes some parsing, such that for example each event stream is preferably organized according to individual events (e.g. via line breaks, tab or other delimiters, etc. as would be known by a skilled artisan upon reading the present disclosure). Accordingly, the result of concatenation may be a more organized structure than simply a single string of all events in a recorded stream; Fig. 3, ¶0129 – Ma teaches segmenting some or all of the concatenated event streams to generate one or more individual traces performed by the user interacting with the computing device, each trace corresponding to a particular task (e.g., concatenating and segmenting the event streams constitutes at least one data transformation operation that converts the time series data into a data representing desktop activities since the concatenating and segmenting organizes the event stream data into traces/tasks); ¶0149 – Ma teaches the hybrid segmenting and clustering approach includes: identifying, among the concatenated event streams, one or more segments characterized by ... a frequency of occurrence within the concatenated event streams greater than or equal to a predetermined frequency threshold, (e.g., the segmenting and clustering including identifying a frequency occurrence constitutes counts); ¶0152 – Ma teaches the concatenated event streams are parsed into subsequences using the sliding window length N and feature vectors are calculated for each subsequence starting at each position within the event stream; ¶0154 – Ma teaches that regardless of the particular manner in which feature vectors are generated, a distance matrix is computed for all pairs of subsequences, (e.g., the concatenated streams being parsed into subsequences and then a distance matrix being computed for all pairs of subsequences constitutes distances between activities));
generating a matrix of activities from the data representing desktop activities, distances between activities, and counts (¶0154 – Ma teaches that regardless of the particular manner in which feature vectors are generated, a distance matrix is computed for all pairs of subsequences (e.g., a distance matrix computed for pairs of subsequences constitutes generating a matrix of activities from the data representing desktop activities, distances between activities, and counts));
determining an optimum number of clusters for the filtered and transformed data (¶0173 – Ma teaches a certain minimum number of clusters should not be violated, i.e. a number of clusters including subsequence(s) not truly belonging to the cluster should be an amount less than or equal to a threshold minimum number of clusters (e.g., a number of clusters that is less than or equal to a threshold minimum number of clusters constitutes determining an optimum number of clusters));
clustering the filtered and transformed time series data into the optimum number of clusters (¶0173 – Ma teaches a certain minimum number of clusters should not be violated, i.e. a number of clusters including subsequence(s) not truly belonging to the cluster should be an amount less than or equal to a threshold minimum number of clusters; ¶0174 – Ma teaches that subsequence is then assigned to the appropriate cluster to improve overall clustering quality, and the process is repeated until each subsequence has been assigned to a cluster (i.e. no further subsequences belonging to more than one cluster await clustering) (e.g., assigning a subsequence to the appropriate cluster where the number of clusters is less than or equal to the threshold minimum number of clusters constitutes clustering the filtered and transformed time series data into the top)); and
the validation test based on a set of quality metrics associated with real processes for robotics process automation including a frequency threshold for combinations of activities within a cluster, a minimum number of activities threshold, and a median inner distance within a cluster being below a threshold (¶0136 – Ma teaches that If a number of identified event subsequences meets or exceeds the minimum frequency threshold, the subsequence is extended and the search performed again; ¶0157 – Ma teaches a number of identified subsequences meeting or exceeding a minimum frequency threshold constitutes a minimum number of activities threshold]; wherein the maximum clustering distance threshold is a value of at least 2 standard deviations smaller than a median distance between the members of the subsequence pair and any other subsequence in the plurality of event streams; ¶0203 – Ma teaches since clustering is based on being 2 standard deviations smaller than a median distance this constitutes a median distance threshold]; the traces clustered according to task type are characterized by: appearing within the recorded event streams at least as frequently as a predetermined frequency threshold (e.g., a predetermined frequency threshold constitutes frequency threshold criteria)).
generating a list of candidate tasks from the clusters (¶0065 – Ma teaches clustering the traces according to a task type; identifying, from among some or all RPA of the clustered traces, one or more candidate processes for robotic automation, (e.g., the candidate processes constitute generating a list of candidate tasks from the clusters)).
Kim discloses what Ma does not expressly disclose.
Kim discloses:
clustering the filtered and transformed time series data into the optimum number of clusters as a set of clusters corresponding to an initial set of candidate tasks (Fig. 3, 302-306, ¶0101-¶0106– Kim teaches processing of events (i.e., tasks) and clustering events and then identifying sub-tasks in the plurality of clustered events (i.e., candidate tasks);
generating a list of validated candidate tasks from the validated clusters (Fig. 3, 308, ¶0107 – Wilson teaches identifying tasks (i.e., generate list) in the plurality of cluster events using the identified sub-tasks).
Ma and Kim are analogous arts because they are from the same field of endeavor with respect to data clustering.
Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to incorporate tasking clustering and sub-clustering as discussed in Kim with method of task identification in a workforce as discussed in Ma by adding the functionality of Kim to the system/method of Ma in order to discover automatable task (Kim, ¶0013).
Wilson discloses what Ma and Kim do not expressly disclose.
Wilson discloses:
performing a validation test for each cluster in the set of clusters, the validation test based on at least one quality metric indicative of a real task (Abstract – Wilson teaches hypothesis testing is performed on each cluster to determine a likelihood that the cluster is a temporally emergent cluster. Clusters with a bursting likelihood above a threshold are determined to be an emergent cluster associated with an anomalous issue.);
Ma, Kim and Wilson are analogous arts because they are from the same field of endeavor with respect to data clustering.
Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to incorporate testing as discussed in Wilson with incorporate tasking clustering and sub-clustering as discussed in Kim with method of task identification in a workforce as discussed in Ma by adding the functionality of Wilson to the system/method of Ma and Kim in order to approximate real-time detection and identification of emerging and related events for performance (Wilson, ¶0001).
As to claim 8, Ma, Kim and Wilson discloses:
method of claim 1, and
Ma discloses:
wherein the resources comprise at least one of human resources and robotics process automation resources (¶0020 – Ma teaches a "task" in the context of robotic process automation (RPA) is to be understood as referring to an objective typically performed by a human interacting with a computing device such as a desktop, laptop, tablet, smartphone, etc. that must be accomplished on a recurring basis (e.g., a task being performed by a human interacting with a computing device constitutes at least one human resource)).
As to claim 9, Ma, Kim and Wilson discloses:
method of claim 1, and
Ma discloses:
wherein the filtering of the time series data comprises removing non-work activities and duplicative activities (¶0111 – Ma teaches that cleaning essentially seeks to reduce the data set by eliminating irrelevant and/or misleading information from the event streams. In preferred approaches, cleaning may involve analyzing the text of the event stream records, and identifying redundant irrelevant events, streams; ¶0112 – Ma teaches a given event stream may include redundant events such as clicking on the same element of a UI multiple times consecutively (consider an impatient user interacting with a slow or unresponsive system) (e.g., irrelevant events constitutes non-work activities and redundant events constitutes duplicate activities)).
As to claim 10, Ma, Kim and Wilson discloses:
method of claim 1, and
Ma discloses:
wherein the filtering comprises removing activities only encountered by a pre-selected threshold low number of resources (¶0182 – Ma teaches that If there are no subsequences that occur with "sufficient" frequency, another search is performed with the next smaller length (L-1). Here "sufficient" frequency is preferably defined in terms of a number of tasks per person per unit of time, and may be defined according to enterprise policy, e.g. a policy of automating tasks performed greater than or equal to a predetermined number of times per day per person (e.g., a predetermined number of times per day per person constitutes a pre-selected threshold low number of resources where tasks are not considered as tasks to automate if they are not performed greater than or equal to the sufficient frequency and therefore constitutes being removed)).
As to claim 11, Ma, Kim and Wilson discloses:
method of claim 10, and
Ma discloses:
wherein the pre-selected threshold low number is one (¶0137 – Ma teaches a minimum frequency threshold value in a range from about 1.0 to about 10.0 may be used for smaller data sets; (e.g., the minimum frequency threshold being in a range from 1-10 constitutes a pre-selected threshold low number of one)).
As to claim 12, Ma, Kim and Wilson discloses:
method of claim 1, and
Ma discloses:
wherein the transformation process comprises determining minimum distances between activities and a count (Fig. 3, ¶0162 – Ma teaches starting with the pair of subsequences with minimal distance (e.g., minimal distance between subsequences constitutes determining minimum distances between activities]); Fig. 3, ¶0136 – Ma teaches employing segmentation in operation 306 of method 300 by analyzing many traces to identify event sequences that are frequent (e.g. occurring at least as frequently as a minimum frequency threshold). If a number of identified event subsequences meets or exceeds the minimum frequency threshold, the subsequence is extended and the search performed again (e.g., a number of identified event subsequences constitutes a count)).
As to claim 17, Ma discloses:
a method of task identification (¶0065 – Ma teaches (a computer-implemented method of identifying one or more processes for robotic automation (RPA)), comprising:
monitoring desktop activities of a workforce, including monitoring time-series data of workers, applications, and screens used (¶0022 Ma teaches a plurality of event streams, each event stream corresponding to a human user interacting with a computing device to perform one or more tasks, are recorded in operation 302; ¶0072 – Ma teaches recording a plurality of event streams constitutes gathering time series data of resources, computer apps, and computer screens for desktop activities of a workforce);
filtering the time-series data to filter out non-work related desktop activities (¶0103, ¶0111 – Ma teaches, [filtering different types of events constitutes performing pre-processing including filtering]; Cleaning essentially seeks to reduce the data set by eliminating irrelevant and/or misleading information from the event streams);
transforming the time series data into a matrix representation indicative of a distance between activities and frequency of occurrence (Fig. 3, ¶0123 – Mat teaches recorded event stream is concatenated. The concatenation process preferably also includes some parsing, such that for example each event stream is preferably organized according to individual events (e.g. via line breaks, tab or other delimiters, etc. as would be known by a skilled artisan upon reading the present disclosure). Accordingly, the result of concatenation may be a more organized structure than simply a single string of all events in a recorded stream; Fig. 3, ¶0129 – Ma teaches segmenting some or all of the concatenated event streams to generate one or more individual traces performed by the user interacting with the computing device, each trace corresponding to a particular task; ¶0149 – Ma teaches the hybrid segmenting and clustering approach includes: identifying, among the concatenated event streams, one or more segments characterized by ... a frequency of occurrence within the concatenated event streams greater than or equal to a predetermined frequency threshold (e.g., the segmenting and clustering including identifying a frequency occurrence constitutes frequency of occurrence); ¶0152 – Ma teaches the concatenated event streams are parsed into subsequences using the sliding window length N and feature vectors are calculated for each subsequence starting at each position within the event stream; ¶0154 – Ma teaches a distance matrix is computed for all pairs of subsequences (e.g.., the concatenated streams being parsed into subsequences and then a distance matrix being computed for all pairs of subsequences constitutes distances between activities); and
generating an initial set of candidate tasks by performing a clustering process to generate a set of clusters and (¶0065 – Ma teaches clustering the traces according to a task type; identifying, from among some or all of the clustered traces, one or more candidate processes for robotic automation (e.g., identifying one or more candidate processes from the clusters constitutes performing clustering to generate a candidate list of tasks); 0217 – Ma teaches that not all processes for RPA will merit implementing a software robot. Using predefined policies specifying the weight (which may be defined according to any suitable measure, such as time (e.g. person-hours), resources (e.g. compute resources, physical resources, financial resources, etc.) required to perform tasks, and/or frequency of performing various tasks, segments and/or traces suitable for automation may be prioritized and a select number of segments/traces chosen for building RP A models based on overall efficiency improvements conveyed upon the enterprise (e.g., specifying weights and prioritizing the segments and traces in order to select a number of traces for building RP A models constitutes generating candidate tasks by performing one other operation to validate valid tasks, where the selected segments for modeling constitute valid tasks since they are valid for modeling according to the predefined policies)).
Kim discloses what Ma does not expressly disclose.
Kim discloses:
identifying validated candidate tasks from the validated clusters tasks (Fig. 3, 302-306, ¶0101-¶0106– Kim teaches processing of events (i.e., tasks) and clustering events and then identifying sub-tasks in the plurality of clustered events (i.e., candidate tasks; Fig. 3, 308, ¶0107 – Wilson teaches identifying tasks (i.e., generate list) in the plurality of cluster events using the identified sub-tasks).
The suggestion/motivation and obviousness rejection is the same as in claim 1.
Wilson discloses what Ma and Kim does not expressly disclose.
Wilson discloses:
performing a validation test for each cluster in the set of clusters, the validation test based on at least one quality metric indicative of a real task (Abstract – Wilson teaches hypothesis testing is performed on each cluster to determine a likelihood that the cluster is a temporally emergent cluster. Clusters with a bursting likelihood above a threshold are determined to be an emergent cluster associated with an anomalous issue.);
The suggestion/motivation and obviousness rejection is the same as in claim 1.
As to claim 18, Ma, Kim and Wilson discloses:
method of claim 17, and
Ma discloses:
further comprising
generating a list of valid tasks (¶0065 – Ma teaches clustering the traces according to a task type; identifying, from among some or all of the clustered traces, one or more candidate processes for robotic automation (e.g., identifying one or more candidate processes from the clusters constitutes performing clustering to generate a candidate list of tasks); 0217 – Ma teaches that not all processes for RPA will merit implementing a software robot. Using predefined policies specifying the weight (which may be defined according to any suitable measure, such as time (e.g. person-hours), resources (e.g. compute resources, physical resources, financial resources, etc.) required to perform tasks, and/or frequency of performing various tasks, segments and/or traces suitable for automation may be prioritized and a select number of segments/traces chosen for building RP A models based on overall efficiency improvements conveyed upon the enterprise (e.g., specifying weights and prioritizing the segments and traces in order to select a number of traces for building RP A models constitutes generating candidate tasks by performing one other operation to validate valid tasks, where the selected segments for modeling constitute valid tasks since they are valid for modeling according to the predefined policies)).
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication No. 2022/0171988 A1 to Ma et al. (“Ma”) in view of U.S. Patent Application Publication No. 2018/0113781 A1 to Kim et al. (“Kim”) in further view of U.S. Patent Application Publication No. 2020/0314127 A1 to Wilson et al. (“Wilson”) in further view of U.S. Patent Application Publication No. 2022/0019888 A1 to Aggarwal et al. (“Aggarwal”).
As to claim 3, Ma, Kim and Wilson discloses:
method of claim 1,
Aggarwal discloses what Ma, Kim and Wilson does not expressly disclose.
Aggarwal discloses:
wherein determining an optimum number of clusters comprises plotting task candidates on a y-axis and an increasing number of clusters on an x-axis (¶0082 - Aggarwal teaches wherein determining an optimum number of clusters comprises plotting task candidates on a y-axis and an increasing number of clusters on an x-axis (an operator of the computer system 110 viewing the clustering output may observe a cluster affinity (indicated on the y-axis) for three clusters over the sequence of events (indicated by time of occurrence associated with each event on the x-axis) for a second cluster over a sequence of events increase while the cluster affinity for a first cluster decreases over the sequence of events and a cluster affinity for a third cluster remains stable over the sequence of events ... the number of clusters ("k") is preset to three (e.g., the sequence of events constitute task candidates and the number of clusters k constitutes an increasing number of clusters on the x-axis)).
Ma, Kim, Wilson and Aggarwal are analogous arts because they are from the same field of endeavor with respect to work analysis.
Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to incorporate clustering as discussed in Aggarwal with testing as discussed in Wilson with incorporate tasking clustering and sub-clustering as discussed in Kim with method of task identification in a workforce as discussed in Ma by adding the functionality of Aggarwal to the system/method of Ma, Kim and Wilson in order to provide supervised predictions (e.g. event type and time of occurrence predictions) and unsupervised dynamic clustering in a neural network using a unified model which enables systems to predict both an event type and associated time of occurrence for the next event in a sequence of events and to indicate an evolution of user behavior in the sequence of events through clustering over the sequence of events (Aggarwal, ¶0005).
Claims 5, 6 and 13-16 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication No. 2022/0171988 A1 to Ma et al. (“Ma”) in view of U.S. Patent Application Publication No. 2018/0113781 A1 to Kim et al. (“Kim”) in further view of U.S. Patent Application Publication No. 2020/0314127 A1 to Wilson et al. (“Wilson”) in further view of Chinese Patent Application No. CN110955774 A1 to Guo et al. (“Guo”).
As to claim 5, Ma, Kim and Wilson discloses:
method of claim 1, and
distance matrix (¶0154 – Ma teaches a distance matrix is computed for all pairs of subsequences)
Guo discloses what Ma, Kim and Wilson does not expressly disclose.
Guo discloses:
wherein the transformation operation comprises generating a count matrix and a [distance matrix] (¶0051- Guo discloses the system will count the frequency of the n selected keywords appearing in the training set at the same time, and construct the importance matrix V = of the keywords according to the frequency (e.g., the importance matrix V constitutes a count matrix since it is constructed based on a count of the frequency of the n selected keywords)).
Ma, Kim, Wilson and Guo are analogous arts because they are from the same field of endeavor with respect to work analysis
Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to incorporate classification as discussed in Guo with testing as discussed in Wilson with incorporate tasking clustering and sub-clustering as discussed in Kim with method of task identification in a workforce as discussed in Ma by adding the functionality of Guo to the system/method of Ma, Kim and Wilson in order to provide a word frequency distribution-based word classification method, device, equipment and medium, and aims to solve the technical problem that efficient and accurate automatic classification of words cannot be realized by reasonably constructing a vector space through the numerical value of the keyword frequency in a short message (Guo, ¶0008).
As to claim 6, Ma, Kim, Wilson and Guo discloses:
method of claim 5, and
distance matrix (¶0154 – Ma teaches a distance matrix is computed for all pairs of subsequences), and
Guo discloses:
further comprising
generating a single matrix from the count matrix and the [distance matrix] to generate a distance matrix having a weighted average (¶0049 – Guo teaches S20: selecting keywords from a training set, establishing a corresponding matrix according to the keywords; ¶0051 – Guo teaches the system will count the frequency of the n selected keywords appearing in the training set at the same time, and construct the importance matrix V = of the keywords according to the frequency (e.g., the importance matrix V constitutes a count matrix since it is constructed based on a count of the frequency of the n selected keywords); ¶0055 – Guo teaches establishing a weighted average spatial distance algorithm, acquiring the occurrence frequency vector of each keyword from the verification set according to the keywords, and calculating the weighted average spatial distance from the short message to the classification group in a vector space through the weighted average spatial distance algorithm according to the occurrence frequency vector and the standardized keyword frequency vector; ¶0057-¶0059 – Guo teaches that It should be understood that a weighted average spatial distance algorithm is established, wherein the weighted average spatial distance is the distance from the short information in the defined verification set to the classification group j in the vector space of the occurrence frequency of the keyword, and the algorithm is as follows: di,j = [(Ri *-Sj*)*V-1 *(Ri *-Sj*)T] 1/2 ... where V-lls the inverse of matrix V (e.g., the weighted average spatial distance being based on the matrix V and frequency vectors from the keyword matrix constitutes generating a single matrix from a count matrix and another matrix to generate a distance matrix having a weighted average)).
The suggestion/motivation and obviousness rejection is the same as in claim 5.
As to claim 13, Ma, Kim and Wilson discloses:
method of claim 1, and
distance matrix (¶0154 – Ma teaches a distance matrix is computed for all pairs of subsequences)
Guo discloses what Ma does not expressly disclose.
Guo discloses:
wherein the transformation process comprises generating a [distance matrix] and a count matrix (¶0051 – Guo teaches the system will count the frequency of the n selected keywords appearing in the training set at the same time, and construct the importance matrix V = of the keywords according to the frequency (e.g., the importance matrix V constitutes a count matrix since it is constructed based on a count of the frequency of the n selected keywords)).
The suggestion/motivation and obviousness rejection is the same as in claim 5.
As to claim 14, Ma, Kim, Wilson and Guo discloses:
method of claim 13, and
Ma discloses:
distance matrix (¶0154 – Ma teaches a distance matrix is computed for all pairs of subsequences)
; and
Guo discloses:
wherein the transformation process comprises generating a matrix of m activities by m activities (¶0051 – Guo teaches the system will count the frequency of the n selected keywords appearing in the training set at the same time, and construct the importance matrix V = of the keywords according to the frequency (e.g., the importance matrix V constitutes a count matrix since it is constructed based on a count of the frequency of the n selected keywords)).
The suggestion/motivation and obviousness rejection is the same as in claim 5.
As to claim 15, Ma, Kim, Wilson and Guo discloses:
method of claim 14, and
Guo discloses:
wherein the transformation process comprises generating a m x m matrix that includes a weight average of distances (¶0051 – Guo teaches the system will count the frequency of the n selected keywords appearing in the training set at the same time, and construct the importance matrix V = of the keywords according to the frequency (e.g., the importance matrix V constitutes a count matrix since it is constructed based on a count of the frequency of the n selected keywords)). The suggestion/motivation and obviousness rejection is the same as in claim 5.
As to claim 16, Ma, Kim, Wilson and Guo discloses:
method of claim 13, and
Ma discloses:
wherein the clustering comprises K-means clustering (¶0210 – Ma teaches clustering may be performed according to known techniques such as K-means clustering). The suggestion/motivation and obviousness rejection is the same as in claim 5.
Claim 21 is rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication No. 2022/0171988 A1 to Ma et al. (“Ma”) in view of U.S. Patent Application Publication No. 2018/0113781 A1 to Kim et al. (“Kim”) in further view of U.S. Patent Application Publication No. 2020/0314127 A1 to Wilson et al. (“Wilson”) in further view of Chinese Patent Application No. CN110955774 A1 to Guo et al. (“Guo”).
As to claim 21, Ma disclose:
a method of task identification in a workforce, comprising:
gathering time series data of resources, computer apps, and computer screens for desktop activities of a workforce (¶0022 Ma teaches a plurality of event streams, each event stream corresponding to a human user interacting with a computing device to perform one or more tasks, are recorded in operation 302; ¶0072 – Ma teaches recording a plurality of event streams constitutes gathering time series data of resources, computer apps, and computer screens for desktop activities of a workforce);
performing pre-processing of the time series data, including filtering of the time series data wherein the filtering of the time series data includes removing activities not related to work projects and removing duplicate activities (¶0103, ¶0111 – Ma teaches, [filtering different types of events constitutes performing pre-processing including filtering]; Cleaning essentially seeks to reduce the data set by eliminating irrelevant and/or misleading information from the event streams);
transforming the pre-processed time series data into a data representing desktop activities, distances between activities, and counts by generating a distance matrix representing average minimum distances from one activity to another for a given resource (¶0154 – Ma teaches a distance matrix is computed for all pairs of subsequences; ¶0021-¶0022 – Ma teaches a "task" in the context of robotic process automation (RPA) is to be understood as referring to an objective typically performed by a human interacting with a computing device such as a desktop, laptop, tablet, smartphone, etc. that must be accomplished on a recurring basis. Exemplary "tasks" in accordance with the present descriptions may include, without limitation, data entry, customer service interactions (e.g. chat sessions, call center interactions), business transactions (e.g. invoice processing, product acquisition, sales, new account processing, etc.), performing processing workflows of any type (e.g. image processing, data analytics, statistical analyses, etc.) or any other type of human/computer interaction that may be represented by a human interacting with one or more user interfaces (Uls) of a computing device; Fig. 3, ¶0027 – Ma teaches preferably based on a critical, often minimal, set of interactions, "traces" are built, in accordance with the inventive concepts described herein. A "trace" is to be understood as a segment that accomplished a particular task in a particular instance (e.g., tasks and traces constitute desktop activities); Fig. 3, ¶0123 – Ma shows each recorded event stream is concatenated. The concatenation process preferably also includes some parsing, such that for example each event stream is preferably organized according to individual events (e.g. via line breaks, tab or other delimiters, etc. as would be known by a skilled artisan upon reading the present disclosure). Accordingly, the result of concatenation may be a more organized structure than simply a single string of all events in a recorded stream; Fig. 3, ¶0129 – Ma teaches segmenting some or all of the concatenated event streams to generate one or more individual traces performed by the user interacting with the computing device, each trace corresponding to a particular task (e.g., concatenating and segmenting the event streams constitutes at least one data transformation operation that converts the time series data into a data representing desktop activities since the concatenating and segmenting organizes the event stream data into traces/tasks); ¶0149 – Ma teaches the hybrid segmenting and clustering approach includes: identifying, among the concatenated event streams, one or more segments characterized by ... a frequency of occurrence within the concatenated event streams greater than or equal to a predetermined frequency threshold, (e.g., the segmenting and clustering including identifying a frequency occurrence constitutes counts); ¶0152 – Ma teaches the concatenated event streams are parsed into subsequences using the sliding window length N and feature vectors are calculated for each subsequence starting at each position within the event stream; ¶0154 – Ma teaches that regardless of the particular manner in which feature vectors are generated, a distance matrix is computed for all pairs of subsequences, (e.g., the concatenated streams being parsed into subsequences and then a distance matrix being computed for all pairs of subsequences constitutes distances between activities))),
determining an optimum number of clusters for the filtered and transformed data (¶0173 – Ma teaches a certain minimum number of clusters should not be violated, i.e. a number of clusters including subsequence(s) not truly belonging to the cluster should be an amount less than or equal to a threshold minimum number of clusters (e.g., a number of clusters that is less than or equal to a threshold minimum number of clusters constitutes determining an optimum number of clusters));
clustering the filtered and transformed time series data into the optimum number of clusters as a set of clusters (¶0173 – Ma teaches a certain minimum number of clusters should not be violated, i.e. a number of clusters including subsequence(s) not truly belonging to the cluster should be an amount less than or equal to a threshold minimum number of clusters; ¶0174 – Ma teaches that subsequence is then assigned to the appropriate cluster to improve overall clustering quality, and the process is repeated until each subsequence has been assigned to a cluster (i.e. no further subsequences belonging to more than one cluster await clustering) (e.g., assigning a subsequence to the appropriate cluster where the number of clusters is less than or equal to the threshold minimum number of clusters constitutes clustering the filtered and transformed time series data into the top));
identifying task candidates for robotics process automation of clusters satisfying a validation test having a plurality of quality metrics including a minimum frequency threshold for combinations of activities within a cluster, a minimum threshold for a number of activities and a median inner distance within a cluster being below a threshold (¶0136 – Ma teaches that If a number of identified event subsequences meets or exceeds the minimum frequency threshold, the subsequence is extended and the search performed again; ¶0157 – Ma teaches a number of identified subsequences meeting or exceeding a minimum frequency threshold constitutes a minimum number of activities threshold]; wherein the maximum clustering distance threshold is a value of at least 2 standard deviations smaller than a median distance between the members of the subsequence pair and any other subsequence in the plurality of event streams; ¶0203 – Ma teaches since clustering is based on being 2 standard deviations smaller than a median distance this constitutes a median distance threshold]; the traces clustered according to task type are characterized by: appearing within the recorded event streams at least as frequently as a predetermined frequency threshold (e.g., a predetermined frequency threshold constitutes frequency threshold criteria)).
Kim discloses what Ma does not expressly disclose.
Kim discloses:
clustering the filtered and transformed time series data into the optimum number of clusters as a set of clusters corresponding to an initial set of candidate tasks (Fig. 3, 302-306, ¶0101-¶0106– Kim teaches processing of events (i.e., tasks) and clustering events and then identifying sub-tasks in the plurality of clustered events (i.e., candidate tasks);
Ma and Kim are analogous arts because they are from the same field of endeavor with respect to data clustering.
Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to incorporate tasking clustering and sub-clustering as discussed in Kim with method of task identification in a workforce as discussed in Ma by adding the functionality of Kim to the system/method of Ma in order to discover automatable task (Kim, ¶0013).
Guo discloses what Ma and Kim does not expressly disclose.
Guo discloses:
a count matrix representing counts between one activity doe a given resource (¶0049 – Guo teaches S20: selecting keywords from a training set, establishing a corresponding matrix according to the keywords; ¶0051 – Guo teaches the system will count the frequency of the n selected keywords appearing in the training set at the same time, and construct the importance matrix V = of the keywords according to the frequency (e.g., the importance matrix V constitutes a count matrix since it is constructed based on a count of the frequency of the n selected keywords); ¶0055 – Guo teaches establishing a weighted average spatial distance algorithm, acquiring the occurrence frequency vector of each keyword from the verification set according to the keywords, and calculating the weighted average spatial distance from the short message to the classification group in a vector space through the weighted average spatial distance algorithm according to the occurrence frequency vector and the standardized keyword frequency vector; ¶0057-¶0059 – Guo teaches that It should be understood that a weighted average spatial distance algorithm is established, wherein the weighted average spatial distance is the distance from the short information in the defined verification set to the classification group j in the vector space of the occurrence frequency of the keyword, and the algorithm is as follows: di,j = [(Ri *-Sj*)*V-1 *(Ri *-Sj*)T] 1/2 ... where V-lls the inverse of matrix V (e.g., the weighted average spatial distance being based on the matrix V and frequency vectors from the keyword matrix constitutes generating a single matrix from a count matrix and another matrix to generate a distance matrix having a weighted average))).
Ma, Kim and Guo are analogous arts because they are from the same field of endeavor with respect to work analysis
Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to incorporate classification as discussed in Guo with incorporate tasking clustering and sub-clustering as discussed in Kim with method of task identification in a workforce as discussed in Ma by adding the functionality of Guo to the system/method of Ma and Kim in order to provide a word frequency distribution-based word classification method, device, equipment and medium, and aims to solve the technical problem that efficient and accurate automatic classification of words cannot be realized by reasonably constructing a vector space through the numerical value of the keyword frequency in a short message (Guo, ¶0008).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TAYLOR A ELFERVIG whose telephone number is (571)270-5687. The examiner can normally be reached Monday (10:00 AM CST) - Friday (4:00 PM CST).
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, Oscar Louie can be reached at (571) 270-1684. 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.
/TAYLOR A ELFERVIG/Primary Examiner, Art Unit 2445
Prosecution Insights