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 .
The status of the Claims
Claims 1-5, 7-12, 14-18 and 20 are pending for examination.
Claims 1, 8 and 14 are independent Claims.
Claims 1-5, 7-12, 14-18 and 20 are rejected under 35 U.S.C. §103
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.
Claim(s) 1-3, 7-10, 14-16 and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arslan et al. (U.S. 2017/0270921 hereinafter Arslan) in view of Duong et al. (U.S. 2021/0065709 hereinafter Duong).
As Claim 1, Arslan teaches a computer-implemented method utilizing representation learning to replace manual feature engineering comprising:
obtaining, by one or more processors (Arslan (¶0004 line 3), computer system), dialog data (Arslan (¶0125, fig. 7 item S1000), system receives an utterance from a user), the dialog data including heterogeneous data items (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words in a written text) including, at least, intents (Arslan (¶127 line 1-3, ¶¶ 0133-0134), matching the written text to a predefine phrase in a menu tree for selecting the operation according to the phrase matched node) and entities (Arslan (¶0126), sequence of words);
generating, by the one or more processors, a heterogeneous network (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree. System generates a heterogenous network which is construed as a network of matched parent and leaf nodes) based on the dialog data (Arslan (¶0126, fig. 7 item S2000), system converts user utterance (dialog data) into words or phrases in a written text) by combining two or more bipartite subnetworks (Arslan (¶0105 fig. 4, ¶0109, ¶0111), first subnetwork (figure 4) maps synonym terms of the menu tree. Arslan (¶0129), “corresponding predefined terms in sequence, wherein different words indicating the same intent are converted into a same term (using the term table of figure 4); each of the predefined terms corresponds to a pare it (parent) node or a leaf node”. Arslan (¶127 line 1-3, ¶¶ 0133-0134), second subnetwork is a menu tree), wherein each of the two or more bipartite subnetworks represents a different relationship between (Arslan (¶0109, fig. 4, ¶0127 fig. 7 item S3000, ¶0129), “corresponding predefined terms in sequence, wherein different words indicating the same intent are converted into a same term (using the term table of figure 4); each of the predefined terms corresponds to a pare it (parent) node or a leaf node (of the menu tree)”) the heterogenous data items in the dialog data (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree.), and wherein each node of the two or more bipartite subnetworks correspond to the data items in the dialog data (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree)
determining, by the one or more processors, automatically node representations for the nodes of the two or more bipartite subnetworks representing the heterogenous data items (Arslan (¶0129), “corresponding predefined terms in sequence, wherein different words indicating the same intent are converted into a same term (using the term table of figure 4 – first bipartite sub network); each of the predefined terms corresponds to a pare it (parent) node or a leaf node (using menu tree or second bipartite subnetwork)”.) through representation learning (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree. The operation is construed as a representation learning because user utterance is converted into its representation in the decision tree), the representation learning providing for automatic recovery of representations need for feature detection of the node representations (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree.), allowing the one or more processor to utilize the representation learning to both automatically learn features of the node representations of the dialog learning system (Arslan (¶0142, ¶0143), if multiple matches happen, system jump to the next step and memorize (or learn) matching nodes) and use the feature to perform a specific task (Arslan (¶0190, ¶0191), matched menu is passed to the terminal and executed on the terminal).
Arslan may not explicitly disclose:
at least one of the two or more bipartite subnetwork representing the intents in the heterogenous data items and at least one of the two or more bipartite subnetworks representing the entities in the heterogenous data items; and
Duong teaches:
at least one of the two or more bipartite subnetwork representing the intents in the heterogenous data items and at least one of the two or more bipartite subnetworks representing the entities in the heterogenous data items (Duong (¶0051 line 7-12), system uses a bipartite structure including topic and action attributes); and
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify term/action graph of Arslan instead be a bipartite structure taught by Duong, with a reasonable expectation of success. The motivation would be to “enable dialog system, specifically the dialog manager, to efficiently and effectively handle follow-up requests” (Duong (¶0054 line 4-7)).
As Claim 2, besides Claim 1, Arslan in view of Duong teaches further comprising:
obtaining, by the one or more processors, a downstream model by using the node representations as data representations of the data items in the dialog data to perform model learning (Arslan (¶0123 line 7-22), system accepts node representations as dynamic field into a model training integration).
As Claim 3, besides Claim 2, Arslan in view of Duong teaches wherein the downstream model includes a model for a task selected from a group comprising: intent detection, slot filing, and natural language generation (Arslan (¶0127 fig. 7 item S3000), system creates a parent and leaf node relationship based on the matching of a written text and a predefined phrase in a menu tree. Leaf node is user’s intention to operate the menu).
As Claim 7, besides Claim 1, Arslan in view of Duong teaches wherein the dialog data are pre-processed by a data cleaner (Arslan (¶0126, ¶0129), system preprocesses utterance and provides sequential words in a written text. Words are converted to terms for clearer meaning).
As Claim 8-10, the Claims are rejected for the same reason as Claims 1-3, respectively.
As Claim 14-16 and 20, the Claims are rejected for the same reasons as Claims 1-3 and 7, respectively.
Claim(s) 4, 11 and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arslan in view of Duong in further view of Menon et al. (U.S. 2021/0051121 hereinafter Menon).
As Claim 4, besides Claim 4, Arslan in view of Duong may not explicitly disclose:
wherein the representation learning is a graph neutral network algorithm implemented on the heterogeneous network.
Menon teaches:
wherein the representation learning is a graph neutral network algorithm implemented on the heterogeneous network (Menon (¶0007 line 12-15 and 22-28), “computational models for the bipartite graphs, e.g., graph-based convolutional neural networks (CNNs) (a graph neural network algorithm implemented on the heterogenous network), are used to compute vector representations (representation learning) of the entity instances associated with the first and second items”).
Arslan discloses a system/method to generate a representation learning network based on the decision tree. Menon discloses a system/method to use graph based neural network as representation learning method. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify representation learning method of Arslan in view of Dong instead be a representation learning method using graph-based convolution neural network taught by Menon, with a reasonable expectation of success. The motivation would be to “automatically scores pairs of a first item and a second item according to the relevance of the second item to the first, enabling recommendations to be made based on the scores, without the need for hand designing features that capture the rich information reflected in the complex-entity definitions and graphs” (Manon (¶0007 line 5-10)).
As Claim 11 and 17, the Claims are rejected for the same reason as Claim 4.
Claim(s) 5, 12 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arslan and Duong in view of Manon in further view of Creed et al. (U.S. 2021/0051121 hereinafter Creed).
As Claim 5, besides Claim 4, Arslan in view of Duong in further view of Manon teaches wherein the objective function (Manon (¶0051 line 1-5), “the neural-network architecture 502 is trained on the current training dataset, in act 608, to minimize an objective function such as, e.g., the regularized form of the negative log likelihood ( corresponding to cross entropy) of the predicted score s("U , J )”) is a sum of two or more objective functions corresponding to the two or more bipartite subnetworks in the heterogeneous network respectively (Manon (¶0051 line 9-19), “the second term is a sum of 12-norms of all the weight matrices in the architecture 502. In some embodiments, the weights and subsequent embeddings from multiple bipartite graphs are learned using an Adam optimization algorithm (which is an extension to the stochastic gradient descent technique introduced by D. Kingma et al. in "Adam: A method for Stochastic Optimization," first published on arXiv in 2014 in TensorFlow (an open-source software library for dataflow and differentiable programming that is suitable for various machine-learning applications) with mini-batches.”).
Arslan in view of Duong in further view of Manon may not explicitly disclose while Jung teaches:
wherein the graph neutral network algorithm is implemented with a negative sample method by optimizing parameters in an objective function (Creed (¶0129 (¶0129 line 1-5), “Training the network, which includes the encoding and decoding/scoring networks, consists of minimizing the cross-entropy loss on the link classE { 0, 1}, enriched by negative samples created by corrupting either the subject or object of a relation with a random entity.”),
Arslan discloses a system/method for training a machine learning model by minimizing object function. Creed discloses a system and method for training machine learning with negative samples. 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 algorithm of Arslan in view of Duong in further view of Manon instead be an algorithm taught by Creed, with a reasonable expectation of success. The motivation would be to improve the performance of machine learning model (Creed (¶00129 last 2 lines)).
As Claim 12 and 18, the Claims are rejected for the same reasons as Claim 5.
Response to Arguments
I. Ground of Rejection 1 (Claims Group A: Claims 1, 18, 14)
As Claims 1, 8 and 14, Applicants argue that Arslan does not disclose “generating ... two or more bipartite subnetworks” because “menu tree” is different than the two or more bipartite subnetworks as claimed (fifth paragraph of page 9 in the remarks).
PNG
media_image1.png
336
602
media_image1.png
Greyscale
Applicants’ arguments are not persuasive because Arslan teaches:
generating, by the one or more processors, a heterogeneous network (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree. System generates a heterogenous network which is construed as a network of matched parent and leaf nodes) based on the dialog data (Arslan (¶0126, fig. 7 item S2000), system converts user utterance (dialog data) into words or phrases in a written text) by combining two or more bipartite subnetworks (Arslan (¶0105 fig. 4, ¶0109, ¶0111), first subnetwork (figure 4) maps synonym terms of the menu tree. Arslan (¶0129), “corresponding predefined terms in sequence, wherein different words indicating the same intent are converted into a same term (using the term table of figure 4); each of the predefined terms corresponds to a pare it (parent) node or a leaf node”. Arslan (¶127 line 1-3, ¶¶ 0133-0134), second subnetwork is a menu tree), wherein each of the two or more bipartite subnetworks represents a different relationship between (Arslan (¶0109, fig. 4, ¶0127 fig. 7 item S3000, ¶0129), “corresponding predefined terms in sequence, wherein different words indicating the same intent are converted into a same term (using the term table of figure 4); each of the predefined terms corresponds to a pare it (parent) node or a leaf node (of the menu tree)”) the heterogenous data items in the dialog data (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree.), and wherein each node of the two or more bipartite subnetworks correspond to the data items in the dialog data (Arslan (¶0126, fig. 7 item S2000), system converts user utterance into words or phrases in a written text. Arslan (¶ 0133-0134), phrases are matched with parent and leaf nodes of the decision tree).
“https://math.stackexchange.com/questions/3065279/a-bipartite-graph” discloses that every tree is a bipartite graph.
PNG
media_image2.png
936
738
media_image2.png
Greyscale
As Claims 1, 8 and 14, Applicants argue that Duong does not disclose “generating ... two or more bipartite subnetworks” because “menu tree” is different than the two or more bipartite subnetworks as claimed (first paragraph of page 10 in the remarks).
Applicant’s arguments are moot because Arslan teaches the limitation(s).
As Claims 1, 8 and 14, Applicants argue that Non-final Office action of July 15, 2025 failed to address Appellant’s argument (second paragraph of page 11 in the remarks).
Applicant’s arguments are moot because current Office Action is issued to address the Applicant’s argument(s)/concern(s).
II. Ground of Rejection 2 (Claims Group B: Claims 4, 11 and 17)
Applicant argues that “menu trees” and “graph neural networks” are fundamentally different computational structures (second paragraph of page 12 in the remarks).
Applicant’s arguments are moot because new reference Manon teaches the limitation(s).
III. Ground of Rejection 3 (Claim Group C: Claims 7, 20)
Applicant disagree with Examiner’s interpretations of “data cleaner” (first paragraph of page 13 in the remarks).
PNG
media_image3.png
306
600
media_image3.png
Greyscale
Applicant’s arguments are not persuasive. Arslan (¶0126, ¶0129 teaches that system preprocesses utterance and provides sequential words in a written text. Words are converted to terms for clearer meaning. Further clarifying “data cleaner” might advance the prosecution.
IV. Ground of Rejection 4 (Claims Group D: Claims 5, 12, 18)
Applicant argues that Jung does not disclose “Negative Sampling method” (second paragraph of page 14 in the remarks).
PNG
media_image4.png
86
562
media_image4.png
Greyscale
Applicant’s arguments are moot because new reference Creed teaches the negative sample method.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Chang et al. (U.S. 2021/0049225) discloses a dynamic interactive graph used as negative samples for further training.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NHAT HUY T NGUYEN whose telephone number is (571)270-7333. The examiner can normally be reached M-F: 12:00-8:00 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, Viker Lamardo can be reached on 571-270-5871. 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.
/NHAT HUY T NGUYEN/Primary Examiner, Art Unit 2147
/VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights