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 . This action is response to the application filed on 11/21/2023.
Claims 1-14 are pending in the application and have been examined.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 11/21/2023, and 08/22/2023 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 4 and 12 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 4 and 12 recite the limitation “generating, according to a plurality of triples associated with the retrieved node and the input text template to generate a plurality of consecutive records of input text and the corresponding output text”. It does not distinctly define the meets and bounds of what is being generated here because there is no object to the action of “generating” the second time the word generating is used. For purpose of the examination this limitation is interpreted as generating additional training data according to the plurality of triples.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1,2,9,10 are rejected under 35 U.S.C. 103 as being unpatentable over Bakis (U.S PG Pub 2020/0042624), in view of Liu et al. “Dynamic Knowledge Graph Reasoning Based on Deep Reinforcement Learning” (hereinafter Liu (1)).
Regarding Claim 1, Bakis discloses,
A training system for a domain-specific data model, the training system comprising a computing device including at least one processor and a storage unit, (Bakis, paragraph 25, “In this illustrative example, data processing system 200 includes communications fabric 202, which provides communications between processor unit 204, memory 206, persistent storage 208, communications unit 210, input/output (I/O) unit(s) 212, and display 214.”)
wherein the storage unit stores a data model, (Bakis, paragraph 45 “The deep learning (DL) process 303 is shown as one box, but in preferred embodiments, it will have separate sub-components which will be trained separately. For example, one DL component will be trained to generate a better system response given a set of user questions; another DL component will be trained for generating new triples given existing triples and so forth. Each of the construction modules 305-309 provide respective data models for storage in the conversational knowledge data models store 313.”)
a domain knowledge graph, (Bakis, figure 6, shows the diagram of the knowledge graph and the figure shows an example of specific domain i.e. various insurance)
a training set generation module, (Bakis, paragraph 49, “The Deep Learning process 303 receives the data from domain specific corpus 301 and also consumes the conversational service 323 run-time dialog logs continuously so that both data sources continually construct a corpus 301 for a conversation flow self-adaptive process. In preferred embodiments, the Deep Learning process 323 learns the conversational flows directly from the data (e.g., pairs of user input and system output” suggesting the presence of a training set, making the deep learning process a training set generation module.)
and the computing device is configured to perform the following processes: generating, by the training set generation module, a training data set based on the domain knowledge graph, wherein the training data set includes at least one record of input text and corresponding output text that correspond to the domain knowledge graph (Bakis, paragraph 49, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module. The deep learning (DL) process 303 is shown as one box, but in preferred embodiments, it will have separate sub-components which will be trained separately. For example, one DL component will be trained to generate a better system response given a set of user questions; another DL component will be trained for generating new triples given existing triples and so forth” where the triples suggest the presence of knowledge graph)
updating the data model based on the training data set (Bakis, paragraph 49, “The Deep Learning process 303 receives the data from domain specific corpus 301 and also consumes the conversational service 323 run-time dialog logs continuously so that both data sources continually construct a corpus 301 for a conversation flow self-adaptive process”)
generating, by the training set generation module, training input text corresponding to the domain knowledge graph; (Bakis, paragraph 45, “The new triple construction module 307 provides the triple input used to construct a knowledge graph model 317. The new table construction module 309 constructs the tables which will be used to answer the user queries.” which includes having input corresponding to the knowledge graph)
inputting the training input text into the data model to obtain training output text; (Bakis, paragraph 49, “The Deep Learning process 323 learns the conversational flows directly from the data (e.g., pairs of user input and system output”)
However, Bakis fails to disclose reinforcement module, evaluation module, score generation and parameter adjustment.
Liu (1), in the same field of endeavor, teaches,
a reinforcement learning module based on a reward model, (Liu (1), section 3.5 , “To get the model with dynamic reasoning ability, we construct the dynamic reward for model training” )
an evaluation module (Liu (1), section 3.4, “To get rid of the pre-trained model, we design a simpler scoring function based on the dynamic reasoning hypothesis, where we obtain the reward value by computing the similarity of the embedding vectors of entities and relations”)
evaluating and generating a score by the evaluation module based on a correlation between the training output text and the domain knowledge graph; (Liu (1), section 3.4 “If the predicted results are incorrect throughout the path, we consider the cosine similarity of three kinds of embeddings to design the reward: (1) embeddings of path relations and query relation; (2) embeddings of the predicted entity and target entity; (3) embeddings of the predicted entity, source entity, and query relation”)
and adjusting, by the reinforcement learning module, parameters of the data model according to the score (Liu (1), section 3.6, “To solve the optimization problem, we use an off-policy soft actor–critic algorithm [26] to adequately explore the action space and update model parameters θ”
and an optimization goal of the reward model until the score meets a training completion condition, and then taking the data model as the domain-specific data model (Liu (1), section 3.7, “To tackle the fixed-step reasoning problem during the testing process, we propose the judgment condition to achieve dynamic knowledge graph reasoning” .... “If the condition holds, we then consider the reasoning successful and stop reasoning, otherwise step forward when t ≤ T.”)
Bakis and Liu (1) are both considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis and Liu (1) to use a evaluation module and adjust the parameters of the data model by the reinforcement module. The motivation to use reinforcement module for parameter adjustment is to have good results. (Liu (1), abstract, “a dynamic knowledge graph reasoning framework is proposed based on deep reinforcement learning, which learns to navigate the graph to find the promising target answer conditioned on the input query.”)
Regarding Claim 2,
the Bakis/Liu (1) combination of claim 1 teaches the training system according to claim 1 (and thus the rejection of claim 1 is incorporated). The combination via Bakis further teaches,
wherein the process of generating the training data set further includes: generating the at least one record of input text and the corresponding output text according to one or more triples in the domain knowledge graph, (Bakis, paragraph 45, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module….. For example, one DL component will be trained to generate a better system response given a set of user questions; another DL component will be trained for generating new triples given existing triples and so forth….The new triple construction module 307 provides the triple input used to construct a knowledge graph model 317. )
Regarding Claim 9, Bakis discloses,
A training method for a domain-specific data model, configuring a computing device including at least one processor and a storage unit to perform the following processes: (Bakis, paragraph 25, “In this illustrative example, data processing system 200 includes communications fabric 202, which provides communications between processor unit 204, memory 206, persistent storage 208, communications unit 210, input/output (I/O) unit(s) 212, and display 214.”)
generating, by the training set generation module, a training data set based on the domain knowledge graph, wherein the training data set includes at least one record of input text and corresponding output text that correspond to the domain knowledge graph (Bakis, paragraph 49, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module. The deep learning (DL) process 303 is shown as one box, but in preferred embodiments, it will have separate sub-components which will be trained separately. For example, one DL component will be trained to generate a better system response given a set of user questions; another DL component will be trained for generating new triples given existing triples and so forth” where the triples suggest the presence of knowledge graph)
updating the data model based on the training data set (Bakis, paragraph 49, “The Deep Learning process 303 receives the data from domain specific corpus 301 and also consumes the conversational service 323 run-time dialog logs continuously so that both data sources continually construct a corpus 301 for a conversation flow self-adaptive process”)
generating, by the training set generation module, training input text corresponding to the domain knowledge graph; (Bakis, paragraph 45, “The new triple construction module 307 provides the triple input used to construct a knowledge graph model 317. The new table construction module 309 constructs the tables which will be used to answer the user queries.” which includes having input corresponding to the knowledge graph)
inputting the training input text into the data model to obtain training output text; (Bakis, paragraph 49, “The Deep Learning process 323 learns the conversational flows directly from the data (e.g., pairs of user input and system output”)
While Bakis fails to teach the score generation and parameters adjustment, Liu (1) discloses,
evaluating and generating a score by the evaluation module based on a correlation between the training output text and the domain knowledge graph; (Liu (1), section 3.4 “If the predicted results are incorrect throughout the path, we consider the cosine similarity of three kinds of embeddings to design the reward: (1) embeddings of path relations and query relation; (2) embeddings of the predicted entity and target entity; (3) embeddings of the predicted entity, source entity, and query relation”)
and adjusting, by the reinforcement learning module, parameters of the data model according to the score (Section 3.6, “To solve the optimization problem, we use an off-policy soft actor–critic algorithm [26] to adequately explore the action space and update model parameters θ”
and an optimization goal of the reward model until the score meets a training completion condition, and then taking the data model as the domain-specific data model (Liu (1), section 3.7, “To tackle the fixed-step reasoning problem during the testing process, we propose the judgment condition to achieve dynamic knowledge graph reasoning” .... “If the condition holds, we then consider the reasoning successful and stop reasoning, otherwise step forward when t ≤ T.”)
Bakis and Liu (1) are both considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis and Liu (1) to use a evaluation module and adjust the parameters of the data model by the reinforcement module. The motivation to use reinforcement module for parameter adjustment is to have good results. (Liu (1), abstract, “a dynamic knowledge graph reasoning framework is proposed based on deep reinforcement learning, which learns to navigate the graph to find the promising target answer conditioned on the input query.”)
Regarding Claim 10,
the Bakis/Liu (1) combination of claim 9 teaches the training method according to claim 9 (and thus the rejection of claim 9 is incorporated). The combination via Bakis further teaches,
wherein the process of generating the training data set further includes: generating the at least one record of input text and the corresponding output text according to one or more triples in the domain knowledge graph, (Bakis, paragraph 45, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module….. For example, one DL component will be trained to generate a better system response given a set of user questions; another DL component will be trained for generating new triples given existing triples and so forth….The new triple construction module 307 provides the triple input used to construct a knowledge graph model 317. )
Claims 5,7,8,13,14 are rejected under 35 U.S.C. 103 as being unpatentable over Bakis, in view of Liu (1), and further in view of Sun et al. “Topic Model based Knowledge Graph for Entity Similarity Measuring”
Regarding Claim 5,
the Bakis/Liu (1) combination of claim 1 teaches
The training system according to claim 1, wherein the step of evaluating and generating the score by the evaluation module (and thus the rejection of claim 1 is incorporated). However, Bakis fails to mention text parsing algorithm, vector space, and calculating the vector space.
Liu (1) discloses,
executing a text parsing algorithm on the training output text, so as to extract entities and relationships of the entities of the training output text to establish a training output text triple structure (Liu (1), section 3.3, “For knowledge graph embedding, the TransE [25] model forms relations by interpreting them as translation operations on the low-dimensional vector space of the entities” suggesting that TransE model is a text parsing algorithm.)
mapping a plurality of nodes of the training output text triple structure to a vector space of the domain knowledge graph, so as to calculate and obtain a plurality of space vectors of the plurality of nodes (Liu(1), section 3.3, “For example, the result of vector (Paris)− vector (France)+ vector (Italy) is closest to the word vector representation vector (Queen) of the word Queen [24], where vector (・) is a function to obtain the embedding information of an entity or relation. For knowledge graph embedding, the TransE [25] model forms relations by interpreting them as translation operations on the low-dimensional vector space of the entities…. For example, vector (Paris) + vector (is − capital − of) ≈ vector (France)... If (es, rq, eo) holds, the embedding of the target entity should be close to the embedding of the source entity plus the embedding that depends on query relation.” showing that the nodes and relations of the knowledge graph are mapped low-dimensional vector space which are used to calculate the relationship of the nodes)
It would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to use a text parsing algorithm of Liu(1) on the output text to extract nodes and mapping the nodes to a vector space of the knowledge graph. The motivation to do so would be to calculate the distance between the nodes and predict correlations between them. (Liu(1), section 3.3, “They believe that when (h, l, t) holds, h+ l ≈ t and t should be a nearest neighbor of h+ l, otherwise h+ l should be far away from t. For example, vector (Paris) + vector (is − capital − of ) ≈ vector (France).
Bakis/Liu (1) fails to teach the average distance and the correlation between output text and the domain knowledge graph.
However, Sun discloses,
and calculating a vector distance of each of the nodes based on the plurality of space vectors, and calculating an average distance between any adjacent two of the nodes, wherein the average distance is used to represent a correlation between the training output text and the domain knowledge graph (Sun, pg. 99, “In terms of text, we use the difference of the probability distribution on topics to calculate the similarity. Since the probability distribution on topics is a vector, we use cosine similarity to measure their differences. For nodes i, j
PNG
media_image1.png
72
217
media_image1.png
Greyscale
”
which computes the similarity between the vectors of two nodes. “At last, we define the average distance between two nodes i,j:
PNG
media_image2.png
158
562
media_image2.png
Greyscale
”
which calculates the average distance between two adjacent nodes. “In terms of relationship, the average distance calculated above is the measure of similarity. Therefore, the similarity of any two nodes i, j in the graph is defined as
PNG
media_image3.png
43
319
media_image3.png
Greyscale
” and pg. 95, “we construct an knowledge graph containing academic papers, scholars and conferences. Then we propose an unsupervised generative model, called AgTM, to extract topics on textual aspect and quantify the strength on relationship aspect.
which combines vector similarity with graph average distance to compute similarity between graph entities where output model (AgTM) is used together with the knowledge graph.)
Bakis, Liu (1), and Sun are considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis, Liu (1) and Sun to use a text parsing algorithm to extract nodes and edges, map the nodes to a vector space to calculate the vector distance of the nodes based on space vectors and use average distance to represent the relation between output text and knowledge graph. The motivation to get the average distance is to get precise data as “the amount of data and the scale of the knowledge graph are usually large, and most of the entities that are not closely related are not necessary to calculate the similarity, we first select some entities and relationships that need to be calculated by similarity among each other” (Sun, Pg.98)
Regarding Claim 7,
the Bakis/Liu (1)/Sun combination of claim 5 teaches,
The training system according to claim 5, (and thus the rejection of claim 5 is incorporated).
Liu(1) discloses,
wherein the vector space of the domain knowledge graph is established by executing a mapping algorithm on all of the triples of the domain knowledge graph (Liu(1), section 3.3, “For example, the result of vector (Paris)− vector (France)+ vector (Italy) is closest to the word vector representation vector (Queen) of the word Queen [24], where vector () is a function to obtain the embedding information of an entity or relation. For knowledge graph embedding, the TransE [25] model forms
relations by interpreting them as translation operations on the low-dimensional vector space of the entities” where the knowledge graph includes the triples of the graph)
It would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to add the mapping algorithm of Liu(1) on the triples of the knowledge graph to establish the vector space of the knowledge graph in the Bakis/Liu(1)/Sun combination. The motivation to do so would be (Liu(1), section 3.3, “ to obtain the embedding information of an entity or relation”)
Regarding Claim 8,
the Bakis/Liu (1)/Sun combination of claim 5 teaches,
The training system according to claim 5(and thus the rejection of claim 5 is incorporated).Bakis further teaches,
wherein the training input text includes a plurality of consecutive records of input text, the training output text is a plurality of records of output text that respectively correspond to the plurality of consecutive records of input text, and the step of generating the score further includes (Bakis, paragraph 45, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module.”)
However, Bakis fails to teach executing text parsing algorithm on the training output text.
Liu (1) discloses,
executing the text parsing algorithm on the training output text to extract entities and relationships of the entities of the plurality of records of output text, so to establish the training output text triple structure (Liu(1), section 3.1, “ A knowledge graph G = (E, R) is defined as a collection of triples (e1, r, e2), where E and R are respectively made up of entities e and relations r” showing that knowledge graph is a combination of triples consisting of entities and relations, section 3.3, “For knowledge graph embedding, the TransE [25] model forms relations by interpreting them as translation operations on the low-dimensional vector space of the entities” showing that TransE model can be sued for extracting entities and relations and puts it in triple form for evaluation and can be used in a training output text to training output triple text structure.
It would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to use a text parsing algorithm like that of Liu(1) to extract nodes and edges of the nodes of the plurality of output text triple structure. The motivation to do so would be (Liu(1), section 3.3 “to form relations by interpreting them as translation operations on the low-dimensional vector space of the entities.”
Regarding Claim 13,
the Bakis/Liu (1) combination of claim 9 teaches
The training method according to claim 9, wherein the step of evaluating and generating the score by the evaluation module (and thus the rejection of claim 9 is incorporated). However, Bakis fails to mention text parsing algorithm, vector space, and calculating the vector space.
Liu (1) discloses,
executing a text parsing algorithm on the training output text, so as to extract entities and relationships of the entities of the training output text to establish a training output text triple structure (Liu (1), section 3.3, “For knowledge graph embedding, the TransE [25] model forms relations by interpreting them as translation operations on the low-dimensional vector space of the entities” suggesting that TransE model is a text parsing algorithm.)
mapping a plurality of nodes of the training output text triple structure to a vector space of the domain knowledge graph, so as to calculate and obtain a plurality of space vectors of the plurality of nodes (Liu(1), section 3.3, “For example, the result of vector (Paris)− vector (France)+ vector (Italy) is closest to the word vector representation vector (Queen) of the word Queen [24], where vector (・) is a function to obtain the embedding information of an entity or relation. For knowledge graph embedding, the TransE [25] model forms relations by interpreting them as translation operations on the low-dimensional vector space of the entities…. For example, vector (Paris) + vector (is − capital − of) ≈ vector (France)... If (es, rq, eo) holds, the embedding of the target entity should be close to the embedding of the source entity plus the embedding that depends on query relation.” showing that the nodes and relations of the knowledge graph are mapped low-dimensional vector space which are used to calculate the relationship of the nodes)
It would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to use a text parsing algorithm of Liu(1) on the output text to extract nodes and mapping the nodes to a vector space of the knowledge graph. The motivation to do so would be to calculate the distance between the nodes and predict correlations between them. (Liu(1), section 3.3, “They believe that when (h, l, t) holds, h+ l ≈ t and t should be a nearest neighbor of h+ l, otherwise h+ l should be far away from t. For example, vector (Paris) + vector (is − capital − of ) ≈ vector (France).
Bakis/Liu (1) fails to teach the average distance and the correlation between output text and the domain knowledge graph.
However, Sun discloses,
and calculating a vector distance of each of the nodes based on the plurality of space vectors, and calculating an average distance between any adjacent two of the nodes, wherein the average distance is used to represent a correlation between the training output text and the domain knowledge graph (Sun, pg. 99, “In terms of text, we use the difference of the probability distribution on topics to calculate the similarity. Since the probability distribution on topics is a vector, we use cosine similarity to measure their differences. For nodes i, j
PNG
media_image1.png
72
217
media_image1.png
Greyscale
”
which computes the similarity between the vectors of two nodes. “At last, we define the average distance between two nodes i,j:
PNG
media_image2.png
158
562
media_image2.png
Greyscale
”
which calculates the average distance between two adjacent nodes. “In terms of relationship, the average distance calculated above is the measure of similarity. Therefore, the similarity of any two nodes i, j in the graph is defined as
PNG
media_image3.png
43
319
media_image3.png
Greyscale
” and pg. 95, “we construct an knowledge graph containing academic papers, scholars and conferences. Then we propose an unsupervised generative model, called AgTM, to extract topics on textual aspect and quantify the strength on relationship aspect.
which combines vector similarity with graph average distance to compute similarity between graph entities where output model (AgTM) is used together with the knowledge graph.)
Bakis, Liu (1), and Sun are considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis, Liu (1) and Sun to use a text parsing algorithm to extract nodes and edges, map the nodes to a vector space to calculate the vector distance of the nodes based on space vectors and use average distance to represent the relation between output text and knowledge graph. The motivation to get the average distance is to get precise data as “the amount of data and the scale of the knowledge graph are usually large, and most of the entities that are not closely related are not necessary to calculate the similarity, we first select some entities and relationships that need to be calculated by similarity among each other” (Sun, Pg.98)
Regarding Claim 14,
the Bakis/Liu (1)/Sun combination of claim 13 teaches,
The training method according to claim 13 (and thus the rejection of claim 13 is incorporated).Bakis further teaches,
wherein the training input text includes a plurality of consecutive records of input text, the training output text is a plurality of records of output text that respectively correspond to the plurality of consecutive records of input text, and the step of generating the score further includes (Bakis, paragraph 45, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module.”)
However, Bakis fails to teach executing text parsing algorithm on the training output text.
Liu (1) discloses,
executing the text parsing algorithm on the training output text to extract entities and relationships of the entities of the plurality of records of output text, so to establish the training output text triple structure (Liu(1), section 3.1, “ A knowledge graph G = (E, R) is defined as a collection of triples (e1, r, e2), where E and R are respectively made up of entities e and relations r” showing that knowledge graph is a combination of triples consisting of entities and relations, section 3.3, “For knowledge graph embedding, the TransE [25] model forms relations by interpreting them as translation operations on the low-dimensional vector space of the entities” showing that TransE model can be sued for extracting entities and relations and puts it in triple form for evaluation and can be used in a training output text to training output triple text structure.
It would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to use a text parsing algorithm like that of Liu(1) to extract nodes and edges of the nodes of the plurality of output text triple structure. The motivation to do so would be (Liu(1), section 3.3 “to form relations by interpreting them as translation operations on the low-dimensional vector space of the entities.”
Claims 6 is rejected under 35 U.S.C. 103 as being unpatentable over Bakis, in view of Liu (1), and Sun, and further in view of Border et al. “Novelty Based Learning of Primitive Manipulation Strategies” and Kim (U.S PG Pub 2022/0156468)
Regarding Claim 6,
the Bakis/Liu (1)/Sun combination of claim 5 teaches
The training system according to claim 5, (and thus the rejection of claim 5 is incorporated). However, Bakis /Liu(1)/Sun fails to mention the training completion condition.
Border discloses,
wherein the training completion condition is met (Border, pg.3, section 2.3, “Conventional machine learning of all types, use a criterion to determine when learning should be stopped. This is typically when a specified number of iterations are completed or the learning error falls below a threshold”)
Bakis, Liu (1), Sun and Border are considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis, Liu(1), Sun and border to add training completion condition of Border to the data model. The motivation to do so would be to have a goal as to when the training is completed and the data model is ready.
Border fails to mention the criteria for the completion condition.
However, Kim discloses,
in response to the average distance being less than a target value (“When a calculated average distance between the two concept words is equal to or smaller than a threshold value, the computing device 100 may determine that the two concept words have the association relationship.”)
Bakis, Liu (1), Sun, Border and Kim are considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis, Liu (1) Sun, Border and Kim to have a completion condition where the average distance is less than any threshold or target value. The motivation to do so would be to keep the version of the model with the highest accuracy.
Claims 3,4, 11, 12 are rejected under 35 U.S.C. 103 as being unpatentable over Bakis, in view of Liu(1), and further in view of Liu, "Deep Learning in Natural Language Processing," (hereinafter Liu(2)).
Regarding Claim 3,
the Bakis/Liu (1) combination of claim 2 teaches,
the training system according to claim 2, wherein the step of generating the at least one record of input text and the corresponding output text according to the one or more triples in the domain knowledge graph further includes (and thus the rejection of claim 2 is incorporated).
Bakis further discloses,
generating the at least one record of input text and the corresponding output text based on an input text template and one or more triples associated with the node (Bakis, paragraph 45, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module….. For example, one DL component will be trained to generate a better system response given a set of user questions; another DL component will be trained for generating new triples given existing triples and so forth.”)
However, they fail to disclose node and its relationship with the knowledge graph.
Liu (2) discloses,
retrieving a node from the domain knowledge graph (Liu (2), section 5.1.1, “A typical KG is usually composed of two elements, entities (i.e., concrete entities and abstract concepts in real world) and relations between entities. Thus, it arranges all kinds of knowledge into large quantities of triple facts in the form of (e1, relation, e2) where e1 indicates the head entity and e2 indicates the tail entity.… Through all these triples, knowledge is thus represented as a huge directed graph, in which entities are considered as nodes and relations as edges” suggesting that the triple contains nodes.)
Bakis, Liu (1) and Liu (2) are all considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis, Liu(1) and Liu (2) to retrieve a node from the domain knowledge graph and generate input text and corresponding output text based on input text template and one or more triples associated with the node. The reason behind this combination would be to generate training data for training the data model.
Regarding Claim 4,
the Bakis/Liu (1)/Liu (2) combination of claim 3 teaches,
The training system according to claim 3 (and thus the rejection of claim 3 is incorporated)
Bakis further teaches,
wherein the process of generating the training data set that includes the at least one record of input text and the corresponding output text further includes: generating, according to a plurality of triples associated with the retrieved node and the input text template to generate a plurality of consecutive records of input text and the corresponding output text (Bakis, paragraph 49, “The Deep Learning process 303 receives the data from domain specific corpus 301 and also consumes the conversational service 323 run-time dialog logs continuously so that both data sources continually construct a corpus 301 for a conversation flow self-adaptive process. In preferred embodiments, the Deep Learning process 323 learns the conversational flows directly from the data (e.g., pairs of user input and system output)” showing the deep learning process and how it uses conversational knowledge data as input to multiple construction module which generates additional training data. The conversational knowledge data can have entities and relations which includes triples. Therefore, the deep learning process (training data set generation module) is generating additional training data according to the plurality of triples.)
Regarding Claim 11,
the Bakis/Liu (1) combination of claim 9 teaches,
the training method according to claim 9, wherein the step of generating the at least one record of input text and the corresponding output text according to the one or more triples in the domain knowledge graph further includes (and thus the rejection of claim 9 is incorporated).
Bakis further discloses,
generating the at least one record of input text and the corresponding output text based on an input text template and one or more triples associated with the node (Bakis, paragraph 45, “The deep learning process 303 provides tailored inputs to the new dialog question and answer construction module 305, the new triple construction module 307 and the new table construction module 309. The inputs to the construction modules vary according to the type output desired from the respective construction module….. For example, one DL component will be trained to generate a better system response given a set of user questions; another DL component will be trained for generating new triples given existing triples and so forth.”)
However, they fail to disclose node and its relationship with the knowledge graph.
Liu (2) discloses,
retrieving a node from the domain knowledge graph (Liu (2), section 5.1.1, “A typical KG is usually composed of two elements, entities (i.e., concrete entities and abstract concepts in real world) and relations between entities. Thus, it arranges all kinds of knowledge into large quantities of triple facts in the form of (e1, relation, e2) where e1 indicates the head entity and e2 indicates the tail entity.… Through all these triples, knowledge is thus represented as a huge directed graph, in which entities are considered as nodes and relations as edges” suggesting that the triple contains nodes.)
Bakis, Liu (1) and Liu (2) are all considered to be analogous to the claimed invention because they are in the same field. Therefore, it would have been obvious to someone of ordinary skill in the art before the filing date of the claimed invention to have combined the teachings of Bakis, Liu(1) and Liu (2) to retrieve a node from the domain knowledge graph and generate input text and corresponding output text based on input text template and one or more triples associated with the node. The reason behind this combination would be to generate training data for training the data model.
Regarding Claim 12,
the Bakis/Liu (1)/Liu (2) combination of claim 11 teaches,
The training method according to claim 11 (and thus the rejection of claim 11 is incorporated)
Bakis further teaches,
wherein the process of generating the training data set that includes the at least one record of input text and the corresponding output text further includes: generating, according to a plurality of triples associated with the retrieved node and the input text template to generate a plurality of consecutive records of input text and the corresponding output text (Bakis, paragraph 49, “The Deep Learning process 303 receives the data from domain specific corpus 301 and also consumes the conversational service 323 run-time dialog logs continuously so that both data sources continually construct a corpus 301 for a conversation flow self-adaptive process. In preferred embodiments, the Deep Learning process 323 learns the conversational flows directly from the data (e.g., pairs of user input and system output)” showing the deep learning process and how it uses conversational knowledge data as input to multiple construction module which generates additional training data. The conversational knowledge data can have entities and relations which includes triples. Therefore, the deep learning process (training data set generation module) is generating additional training data according to the plurality of triples.)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PRAYUKTA GHIMIRE whose telephone number is (571)270-5484. The examiner can normally be reached M-F, 9 am to 5 pm ET.
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, Kakali Chaki can be reached at (571) 272-3719. 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.
/P.N.G./Examiner, Art Unit 2122
/BRIAN M SMITH/ Primary Examiner, Art Unit 2122
Prosecution Insights