DETAILED ACTION
This communication is in response to the Amendments and Arguments filed on 2/10/2026. Claims 1-20 are pending and have been examined. Hence, this action has been made FINAL.
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 .
Response to Arguments
The reply filed on 2/10/2026 has been entered. Applicant’s arguments with respect to claims 1-20 have been considered but are moot in view of new ground(s) of rejection caused by the amendments.
With respect to the applicant’s arguments to claim rejections under 35 U.S.C § 101, Applicant has amended each of the independent claims and asserts that “the claimed system is not a generic computer implementation, it is a specific architecture for hierarchical personalization of a machine learning model.” The examiner agrees that these newly added limitations overcome the rejection under 35 U.S.C. 101.
With respect to the applicant’s arguments to claim rejections under 35 U.S.C § 103, the applicant’s arguments with respect to claims 1-20 have been considered but are moot in view of new ground(s) of rejection caused by the amendments.
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 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.
Claims 1-20 are rejected under 35 U.S.C. 103 as obvious over US Patent Publication 20240311577 A1, (Sinha et al.) in view of US Patent Publication 20250005629 A1 (Tan et al.) in view of "Transformer Hawkes Process" (Zuo et al.) in view of "Unlocking the Potential of User Feedback: Leveraging Large Language Model as User Simulator to Enhance Dialogue System" (Hu et al.).
Claim 1
Regarding claim 1, Sinha et al. disclose a system comprising at least one processor (Sinha et al. ¶ [0053], "This system of method 200 includes one or more processors") to:
determine first conversation history with a user (Sinha et al. ¶ [0042], " For example, the context engine 113 can determine a current context based on a current state of a dialog session (e.g., considering one or more recent inputs provided by a user during the dialog session)" A dialog session state is considered analogous to a first conversation history); and
determine a response by applying the first conversation history with the user to a machine learning model (Sinha et al. ¶ [0055]-[0057], “the system can augment the first NL based input (e.g., augment the explicit NL based input) with additional information, such as one or more past or current contexts of the client device and/or a user of the client device (e.g., via the context engine 113). … At block 220, the system generates, based on the first NL based input and using at least one large language model (LLM), one or more instances of first LLM output.” Use of context, which includes the current state of the dialog session, to generate an LLM output is considered analogous to applying a first conversation history with a user to a ML model), the machine learning model configured with a first set of pre-trained parameters (Sinha et al. ¶ [0057], "each of the at least one LLM can include billions of weights and/or parameters that are learned through training the LLM on enormous amounts of diverse data.") [corresponding to a plurality of users and a second set of personalization parameters specific to the user], wherein:
[the second set of personalization parameters corresponds to an adaptation component configured to modify, specific to the user and different from the plurality of users, at least one weight or activation associated with the first set of pre-trained parameters and that generates a modulation output based at least on a time-dynamic representation of at least one interest of the user];
[the modulation output is applied to the first set of pre-trained parameters to generate] generating a plurality of candidate responses to a question (Sinha et al. ¶ [0059], "At block 230, the system determines, based on the one or more instances of first LLM output, at least three responses to the first NL based input.");
the machine learning model is updated using the user input indicative of an interest level of the user for at least a subset of a plurality of candidate responses to a question (Sinha et al. ¶ [0069]-[0072], "At block 270, the system receives user input associated with the client device, the user input indicating a user selection of a particular response ... the system can cause the update engine 134 to update the state of the LLM based on the particular response that was selected at block 270." LLM is considered analogous to a machine learning model);
the machine learning model is updated using the user input as a reward signal (Sinha et al. ¶ [0069]-[0072], "At block 270, the system receives user input associated with the client device, the user input indicating a user selection of a particular response ... the system can cause the update engine 134 to update the state of the LLM based on the particular response that was selected at block 270."), [the reward signal applied to update the second set of personalization parameters, the machine learning model analyzes temporal dynamics in user engagement of the user for updating the at least one weight or activation.]
Sinha et al. do not explicitly disclose all of a first and second set of parameters corresponding to a machine learning model.
However, Tan et al. disclose determining a response [by applying the first conversation history] to a machine learning model (Tan et al. ¶ [0039], "The LLM is configured to receive a prompt and generate a response to the prompt."), the machine learning model configured with a first set of pre-trained parameters (Tan et al. ¶ [0081], "The machine-learned language model (i.e., the original model) may be a large language model (LLM), which is fine-tuned to generate personalized responses for customer service at the online concierge system 140." ¶ [0066], “Each machine-learning model includes a set of parameters. The set of parameters for a machine-learning model are parameters that the machine-learning model uses to process an input to generate an output.”) [corresponding to a plurality of users] and a second set of personalization parameters specific to the user (Tan et al. ¶ [0081], "During the fine-tuning process in FIG. 3C, parameters of the original LLM 340 may be frozen, while parameters of the personalization model 330 are first copied based on the original LLM 340, but are updated through the fine-tuning process."), wherein:
the second set of personalization parameters corresponds to an adaptation component configured to modify, specific to the user and different from the plurality of users, at least one weight or activation associated with the first set of pre-trained parameters (Tan et al. ¶ [0081], "In one or more embodiments, the personalization module 270 provides a set of training examples to the model deployment system 150 for fine-tuning the personalization model 330. The training examples include, for each user, a user representation and user input including a prompt/request.") and that generates a [modulation] output based at least on a time-dynamic representation of at least one interest of the user (Tan et al. ¶ [0071], "FIG. 3A illustrates an example algorithmic flow of training a user representation model 310 to user behaviors 305 to output a user representation 315 at an online concierge system 140 … the input to the user representation model may include parameters associated with the user behaviors in the sequence, such as … purchasing time, purchasing history, etc.");
the [modulation] output is applied to the first set of pre-trained parameters to generate a plurality of candidate responses to a question (Tan et al. ¶ [0073], " the user representation model 310 may use a loss function to evaluate the prediction that indicates a difference between the expected token and the predicted token for each training instance. The parameters of the user representation model 310 are updated based on backpropagating error terms from the loss function.");
the machine learning model is updated using the user input indicative of an interest level of the user for at least a subset of a plurality of candidate responses to a question (Tan et al. ¶ [0077]-[0078], "the training example may be associated with a set of responses that are ranked based on the level of personalization/customization for a user … The loss function is computed that indicates a difference between the estimated evaluation scores between the higher ranked and lower ranked responses. The parameters of the evaluation model 320 are updated to backpropagate error terms from the loss function, such that the difference in evaluation scores between the two are maximized. … the training examples used for training the evaluation model 320 may be collected based on user reactions to LLM generated responses as well as human evaluations.").
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Sinha et al.’s personalized multi-response dialog system to incorporate Tan et al.’s machine-learning model training.
The suggestion/motivation for doing so would have been that, “personalized customization/optimization and general LLM evaluation may be balanced and align the personalization model 330 towards personalized experiences,” as noted by the Tan et al. disclosure in paragraph [0085].
Sinha et al. in view of Tan et al. do not explicitly disclose all of a modulation output.
However, Zuo et al. disclose a machine learning model configured with a first set of pre-trained parameters corresponding to a plurality of users (Zuo et al. pg. 11, Section 5.1, Paragraph 1, "The 911-Calls dataset contains emergency phone call records. Calling time, location of the caller, and nature of the emergency are logged for each record. ... We treat location of callers (given by zipcodes) as vertices on a relational information graph." pg. 12, Section 5.2, Paragraph 1, "Models are trained on 911-Calls... with different number of training events.") [and a second set of personalization parameters specific to the user], wherein:
[the second set of personalization parameters corresponds to] an adaptation component is configured to modify, [specific to the user and different from the plurality of users,] at least one weight or activation associated with the first set of pre-trained parameters and that generates a modulation output based at least on a time-dynamic representation (Zuo et al. pg. 5-6, Section 3.1, Paragraph 1-2, "our model still needs to be aware of the temporal information of inputs, i.e., time stamps. Therefore, analogous to the original positional encoding method (Vaswani et al., 2017), we propose to use a temporal encoding procedure, defined by [Equation 2]. Eq. 2 uses trigonometric functions to define a temporal encoding for each time stamp, i.e., for each
t
j
, we deterministically computes
z
t
j
ϵ
R
M
, where
M
is the dimension of encoding." Equation 2's trigonometric functions are considered analogous to a modulation output based on a time-dynamic representation) of at least one interest of the user (Zuo et al. pg. 11, Section 5.1, Paragraph 1, "We adopt several datasets to evaluate the models. ... The Retweets dataset contains sequences of tweets, where each sequence contains an origin tweet (i.e., some user initiates a tweet), and some follow-up tweets. We record the time and the user tag of each tweet.");
the modulation output is applied to the first set of pre-trained parameters (Zuo et al. pg. 6, Section 3.1, Paragraph 3, "Besides temporal encoding, we train an embedding matrix
U
∈
R
M
×
K
for the event types ... Embedding of the event sequence
S
=
{
t
j
,
k
j
}
j
=
1
L
is then specified by
X
=
(
U
Y
+
Z
)
T
(Equation 3) ... After the initial encoding and embedding layers, we pass
X
through the self-attention module. ... The attention output
S
is then fed through a position-wise feed-forward neural network, generating hidden representations
h
(
t
)
of the input event sequence" See Equation 5, which illustrates applying attention output
S
to neural network parameters) [to generate a plurality of candidate responses to a question]; …
the machine learning model analyzes temporal dynamics in user engagement of the user for updating the at least one weight or activation (Zuo et al. pg. 4, Section 1, Paragraph 10, "In social networks, each user has her own sequence of events, like tweets and comments. Sequences among users can be related, for example, a tweet from a user may trigger retweets from her followers. We can use graphs to model these follower-followee relationships (Zhou et al., 2013; Farajtabar et al., 2017), where each vertex corresponds to a specific user and each edge represents connections between the two associated users. We propose an extension to [Transformer Hawkes Process] that integrates these relational graphs (Borgatti et al., 2009; Linderman and Adams, 2014) into the self-attention module via a similarity metric among users.").
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Sinha et al. in view of Tan et al. to incorporate Zuo et al.’s modulation output.
The suggestion/motivation for doing so would have been that, “the Transformer Hawkes Process (THP) model [is] able to capture both short-term and long-term dependencies whilst enjoying computational efficiency,” as noted by Zuo et al. on pg. 3, Section 1, Paragraph 7.
Sinha et al. in view of Tan et al. in view of Zuo et al. do not explicitly disclose all of updating a machine learning model using a reward signal.
However, Hu et al. disclose a machine learning model configured with a first set of pre-trained parameters [corresponding to a plurality of users] and a second set of personalization parameters specific to a user (Hu et al. pg. 3, Section 3.3, Paragraph 1-6, "a key design goal in our approach [is] employing LLMs as a user simulator to provide satisfaction feedback and leveraging this feedback to optimize the fine-tuning TOD model, so as to leverage the LLM’s knowledge, understanding and reasoning capabilities. This user feedback guides the TOD model toward producing responses that better satisfy the user. ... We utilize Flan-T5 (large version) [4] as the fine-tuning TOD model. This model has been trained extensively on instruction tasks and is well-suited for domain-supervised fine-tuning."), wherein: …
the machine learning model is updated using the user input as a reward signal, the reward signal applied to update the second set of personalization parameters (Hu et al. pg. 2, Section 3.1, Paragraph 1, "we fine-tune the TOD models through supervised training with response data and optimize it using the PPO algorithm and satisfaction rewards." Fine-tuning with response data is considered analogous to updating a second set of personalization parameters).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Sinha et al. in view of Tan et al. in view of Zuo et al. to incorporate Hu et al.’s policy updating using a reward signal.
The suggestion/motivation for doing so would have been that, “the optimization goal during this phase is to achieve higher satisfaction scores. … The aforementioned optimization process can be complex and unstable for the TOD model. Therefore, we approach the TOD optimization as a reinforcement learning problem and employ the proximal policy optimization (PPO) algorithm,” as noted by the Hu et al. disclosure in pg. 3, Section 3.3, Paragraph 3-4.
Claim 2
Regarding claim 2, the rejection of claim 1 is incorporated. Sinha et al. in view of Tan et al. in view of Hu et al. disclose all the elements of the claimed invention as stated above. Sinha et al. further disclose wherein the at least one processor is to determine the plurality of candidate responses using second conversation history with the user and user data corresponding to the user (Sinha et al. ¶ [0078], "the system can augment the second NL based input (e.g., augment the explicit NL based input) with additional information, such as one or more past or current contexts of the client device and/or a user of the client device (e.g., via the context engine 113)." ¶ [0083], "At block 330, the system determines... at least three responses to the second NL based input." Current device context after a first LLM output is considered analogous to a second conversation history. User context is considered analogous to user data. See Figs. 2 and 3 for a flowchart illustration).
Claim 3
Regarding claim 3, the rejection of claim 1 is incorporated.
Tan et al. further disclose wherein:
the machine learning model comprises an encoder (Tan et al. ¶ [0072], “the user representation model 310 may take the sequence of user behaviors 305 and generate a contextual vector. The contextual vector may be used as the condensed user representation for user intentions and preferences.” User representation model 310 is considered analogous to an encoder), a base policy network (Tan et al. ¶ [0081], "the personalization model 330 may be trained using reinforcement leaning policy iterations" Personalization model 330 is considered analogous to a base policy network), and a personalization network (Tan et al. ¶ [0080], "personalization model 330 integrates a machine learned language model (e.g., LLM of the model serving system 150) with the user representations and the evaluation model into a reinforcement learning framework that is iteratively trained based on individual user's intentions and preferences." Personalization model 330 is considered analogous to a personalization network);
the encoder is to determine at least one vector representation of user data of the user by encoding user data (Tan et al. ¶ [0072], "the user representation model 310 may take the sequence of user behaviors 305 and generate a contextual vector. The contextual vector may be used as the condensed user representation for user intentions and preferences.");
the personalization network is to determine at least one weight by inputting the at least one vector representation into the personalization network (Tan et al. ¶ [0085], "the personalization model 330 can learn the user's preferences and intentions and adjust the differences from the original LLM using reinforcement learning." ¶ [0072], “The contextual vector may be used as the condensed user representation for user intentions and preferences.” A user's preferences and intentions are represented by a contextual vector. Therefore, the personalization model receives an input of a vector representation); and
the at least one weight is applied to the base policy network (Tan et al. ¶ [0085], "The parameters of the personalization model 330 may be updated by backpropagating through the personalization model 330 based on the total loss" Backpropagating is considered analogous to applying weights).
Claim 4
Regarding claim 4, the rejection of claim 3 is incorporated.
Tan et al. further disclose updating the base policy network and the personalization network (Tan et al. ¶ [0085], "The parameters of the personalization model 330 may be updated by backpropagating through the personalization model 330 based on the total loss" Backpropagating is considered analogous to applying weights).
Hu et al. further disclose [the base policy network and the personalization network are updated] updating policy networks using proximal policy optimization to increase the reward signal (Hu et al. pg. 2, Section 1, Paragraph 4, "the satisfaction score predicted by the LLM will be utilized as a reward to optimize the fully supervised trained TOD system. This will be accomplished by employing the Proximal Policy Optimization (PPO) algorithm, which allows us to capitalize on user feedback.").
Claim 5
Regarding claim 5, the rejection of claim 3 is incorporated.
Tan et al. further disclose wherein:
the personalization network employs a Multilayer Perceptron (MLP) network (Tan et al. ¶ [0080], "The personalization model 330 integrates a machine learned language model (e.g., LLM of the model serving system 150) with the user representations and the evaluation model into a reinforcement learning framework that is iteratively trained based on individual user's intentions and preferences." ¶ [0065], "The machine-learning models may also include neural networks, such as perceptrons [and] multilayer perceptrons") to generate a plurality of weights based on user data (Tan et al. ¶ [0085], "the personalization model 330 can learn the user's preferences and intentions and adjust the differences from the original LLM using reinforcement learning. The parameters of the personalization model 330 may be updated by backpropagating through the personalization model 330 based on the total loss." Updating a set of machine learning model parameters is considered analogous to generating a plurality of weights); and
the plurality of weights are applied to the base policy network to personalize responses (Tan et al. ¶ [0085], "the personalization model 330 can learn the user's preferences and intentions and adjust the differences from the original LLM using reinforcement learning. The parameters of the personalization model 330 may be updated by backpropagating through the personalization model 330 based on the total loss.").
Claim 6
Regarding claim 6, the rejection of claim 1 is incorporated.
Tan et al. further disclose wherein the machine learning model is a large language model (LLM), and wherein the machine learning model is updated using reinforcement learning from human feedback (RLHF) (Tan et al. ¶ [0080], "The personalization model 330 integrates a machine learned language model (e.g., LLM of the model serving system 150) with the user representations and the evaluation model into a reinforcement learning framework that is iteratively trained based on individual user's intentions and preferences." ¶ [0078], "the training examples used for training the evaluation model 320 may be collected based on user reactions to LLM generated responses" A reinforcement learning framework that is trained on individual user’s intentions and preferences is considered analogous to RLHF, since a user’s feedback is collected in order to generate the training examples).
Claim 7
Regarding claim 7, the rejection of claim 1 is incorporated.
Sinha et al. further disclose wherein the at least one processor is comprised in at least one of:
…
a system implementing language models (Sinha et al. ¶ [0035], "The example environment includes a client device 110 and a natural language (NL) based output system 120."); … .
Claim 8
Regarding claim 8, the rejection of claim 1 is incorporated.
Tan et al. further disclose wherein the user input comprises a first user input [indicating positive interest] and a second user input [indicating negative interest] (Tan et al. ¶ [0077], "For a pair of responses (with one being higher ranked than the other), the model serving system 150 applies estimated parameters of the evaluation model 320 to both the lower ranked and higher ranked responses. The loss function is computed that indicates a difference between the estimated evaluation scores between the higher ranked and lower ranked responses. " A higher ranked response is considered analogous to a first user input. A lower ranked response is considered analogous to a second user input).
Hu et al. further disclose wherein the user input comprises a first [user] input indicating positive interest and a second [user] input indicating negative interest (Hu et al. pg. 2, Section 1, Paragraph 4, "the satisfaction score predicted by the LLM will be utilized as a reward to optimize the fully supervised trained TOD system. This will be accomplished by employing the Proximal Policy Optimization (PPO) algorithm, which allows us to capitalize on user feedback." See Figure 2, which illustrates user feedback taking the form of a five-star rating system. A 1 or 2 star rating out of 5 is considered analogous to negative interest; a 3 or higher rating out of 5 is considered analogous to a positive interest).
Claim 9
Regarding claim 9, the rejection of claim 1 is incorporated. Sinha et al. in view of Tan et al. in view of Hu et al. disclose all the elements of the claimed invention as stated above. Hu et al. further disclose wherein:
the machine learning model comprises a base policy network (Hu et al. pg. 3, Section 3.3, Paragraph 4, "We initialize the policy network
π
0
using the TOD model, denoted as
p
T
O
D
.");
the at least one processor updates the base policy network using a reward function (Hu et al. pg. 3, Section 3.3, Paragraph 5, "To incorporate penalty rewards, we also utilize the KL-divergence and set the hyperparameter
β
for training. Consequently, the final reward is calculated as follows:
r
x
,
y
=
L
L
M
(
x
,
y
)
-
β
l
o
g
π
(
y
|
x
)
p
T
O
D
(
y
|
x
)
"
r
x
,
y
is considered analogous to a reward function);
the at least one processor applies the [user] input as the reward signal to the reward function (Hu et al. pg. 2, Section 1, Paragraph 4, "the satisfaction score predicted by the LLM will be utilized as a reward to optimize the fully supervised trained TOD system. This will be accomplished by employing the Proximal Policy Optimization (PPO) algorithm, which allows us to capitalize on user feedback."); and
the machine learning model is updated by increasing an expected reward for the base policy network (Hu et al. pg. 3, Section 3.3, Paragraph 4-5, "we approach the TOD optimization as a reinforcement learning problem and employ the proximal policy optimization (PPO) algorithm. ... the final reward is calculated as follows:
r
x
,
y
=
L
L
M
(
x
,
y
)
-
β
l
o
g
π
(
y
|
x
)
p
T
O
D
(
y
|
x
)
" pg. 2, Section 3.2, Paragraph 3, "the satisfaction score predicted by the LLM [is] denoted as
S
.
S
=
L
L
M
(
x
,
y
)
"
r
x
,
y
increases if satisfaction score increases."). Sinha et al. and Tan et al. disclose all of a user input selecting LLM responses.
Claim 10
Regarding claim 10, Sinha et al. disclose a system comprising at least one processor (Sinha et al. ¶ [0053], "This system of method 200 includes one or more processors") to:
cause a user interface to display a question and a plurality of candidate responses to the question (Sinha et al. ¶ [0145], "responsive to the NL based input 1065 of “What should I visit this weekend in Kentucky?”, the candidate segment engine 132 may determine five responses to the NL based input 1065." See Fig. 10, which illustrates the user interface displaying a question and a plurality of responses), wherein a content of the question is determined (Sinha et al. ¶ [0057], "At block 220... the system can cause the LLM engine 131 to process, using at least one LLM stored in the LLM(s) database 131A, the first NL based input to generate one or more instances of first LLM output. … the one or more instances of first LLM output can be considered a stream in that, as each word or phrase of the first NL based input is being processed using the LLM, the probability distribution over the sequence of words or phrases that are predicted to be responsive to the first NL based input can be continuously updated and with respect any previously selected segments for a stream of NL based output.") using the at least one processor (Sinha et al. ¶ [0053], "This system of method 200 includes one or more processors");
determine user input of a user via the user interface, the user input indicative of an interest level of the user for [each of] the plurality of candidate responses (Sinha et al. ¶ [0069]-[0072], "At block 270, the system receives user input associated with the client device, the user input indicating a user selection of a particular response ... the system can cause the update engine 134 to update the state of the LLM based on the particular response that was selected at block 270." LLM is considered analogous to a machine learning model); and
update one or more parameters of a machine learning model using the user input (Sinha et al. ¶ [0069]-[0072], "At block 270, the system receives user input associated with the client device, the user input indicating a user selection of a particular response ... the system can cause the update engine 134 to update the state of the LLM based on the particular response that was selected at block 270.") [as a reward signal].
Tan et al. disclose determining user input of a user via the user interface, the user input indicative of an interest level of the user for each of the plurality of candidate responses (Tan et al. ¶ [0077]-[0078], "the training example may be associated with a set of responses that are ranked based on the level of personalization/customization for a user … the training examples used for training the evaluation model 320 may be collected based on user reactions to LLM generated responses as well as human evaluations.").
The remaining limitations of claim 10 are similar in scope to that of claim 1 and therefore are rejected for similar reasons as described above.
Claim 11
Regarding claim 11, the rejection of claim 10 is incorporated. The limitations of claim 11 are similar in scope to that of claim 2 and therefore are rejected for similar reasons as described above.
Claim 12
Regarding claim 12, the rejection of claim 10 is incorporated. Sinha et al. in view of Tan et al. in view of Hu et al. disclose all the elements of the claimed invention as stated above. Sinha et al. further disclose wherein the question and the plurality of candidate responses are displayed simultaneously in a same screen of the user interface (Sinha et al. Fig. 10 illustrates the user interface displaying a question and a plurality of responses simultaneously).
Claim 13
Regarding claim 11, the rejection of claim 10 is incorporated. The limitations of claim 13 are similar in scope to that of claim 3 and therefore are rejected for similar reasons as described above.
Claim 14
Regarding claim 14, the rejection of claim 13 is incorporated. The limitations of claim 14 are similar in scope to that of claim 4 and therefore are rejected for similar reasons as described above.
Claim 15
Regarding claim 15, the rejection of claim 13 is incorporated. The limitations of claim 15 are similar in scope to that of claim 5 and therefore are rejected for similar reasons as described above.
Claim 16
Regarding claim 16, the limitations of claim 16 are similar in scope to that of claim 1 and therefore are rejected for similar reasons as described above.
Claim 17
Regarding claim 17, the rejection of claim 16 is incorporated. The limitations of claim 17 are similar in scope to that of claim 2 and therefore are rejected for similar reasons as described above.
Claim 18
Regarding claim 18, the rejection of claim 10 is incorporated. The limitations of claim 18 are similar in scope to that of claim 3 and therefore are rejected for similar reasons as described above.
Claim 19
Regarding claim 19, the rejection of claim 18 is incorporated. the limitations of claim 19 are similar in scope to that of claim 4 and therefore are rejected for similar reasons as described above.
Claim 20
Regarding claim 20, the rejection of claim 18 is incorporated. The limitations of claim 20 are similar in scope to that of claim 5 and therefore are rejected for similar reasons as described above.
Reference Cited
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
“Reinforcing User Retention in a Billion Scale Short Video Recommender System” to Cai et al. disclose training a reinforcement learning model based on user engagement with video content.
US Patent Publication 20250181475 A1 to Edwards et al. disclose training a machine learning model to identify user engagement with textual content. In particular, the art describes generating a plurality of candidate responses in order to measure user engagement as well as a distinct separation between a first set of pre-trained parameters and a second set of personalization parameters.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, 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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JACOB B VOGT whose telephone number is (571)272-7028. The examiner can normally be reached Monday - Friday 9:30am - 7pm 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, Paras D Shah can be reached at (571)270-1650. 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.
/JACOB B VOGT/ Examiner, Art Unit 2653
/Paras D Shah/ Supervisory Patent Examiner, Art Unit 2653
04/02/2026
Prosecution Insights