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 .
Response to Arguments
Applicant’s arguments regarding the 103 rejection have been fully considered but are respectfully not persuasive. Applicant argues that the prior art references, particularly the Zhang reference, does not teach the limitations of learning a linking model, processing via the linking model, and generating based on the linking model (shortened for brevity).
More specifically, Applicant argues that the Zhang reference operates in the text domain and that the embeddings are contextual text embeddings and not knowledge graph embeddings. In response, Examiner argues that the knowledge graph embeddings are taught and introduced by the Saxena reference. Embeddings as a whole are vector representation of data so regardless of the type, the functionality is the same. Thus the combination of Saxena and Zhang have the KG embeddings from Saxena combined with the functionality of Zhang.
Applicant argues that the linking model operates in the knowledge graph embedding domain and is learned based on training data and an ontology. In response, Examiner argues that the Zhang reference states “Specifically, we use the AIDA-CoNLL dataset (Hoffart et al., 2011) as the in-domain dataset: we train EntQA on the training portion of AIDA, use the validation portion (AIDA-A) for development, and reserve the test portion (AIDA B) for in-domain test performance” Zhang pg. 5 §3.1. This shows the training data used along with the ontology for it. Again, the model doesn’t care what type of embeddings it says, functionally it is all the same. The claim simply recites “processing” and “generating” with no specific functionality given to the model itself. The data enters, and the output is given. Zhang is cited simply for its teaching of the linking model and meets all functionality as claimed. It is noted that limitations from the specification are not required to be read into the claims. Therefore the references still meet the broadest reasonable interpretation of the claims.
It is also noted that while not formally cited, Costabello et al. USPAT 10,157,226 teaches these limitations entirely, “A device receives training data and an ontology for the training data, where the training data includes information associated with a subject of the ontology. The device generates a knowledge graph based on the training data and the ontology, and converts the knowledge graph into knowledge graph embeddings, where the knowledge graph embeddings include points in a k-dimensional metric space” Costabello abstract.
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.
Claim(s) 1-4, 6-11, 13-18, 20, and 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Saxena, Apoorv, Aditay Tripathi, and Partha Talukdar. "Improving multi-hop question answering over knowledge graphs using knowledge base embeddings." in view of Zhang, Wenzheng, Wenyue Hua, and Karl Stratos. "EntQA: Entity linking as question answering." further in view of Bao, Junwei, et al. "Constraint-based question answering with knowledge graph.
Regarding claims 1, 8, and 15, Saxena teaches “a computer-implemented method of generating answers to multi-hop constraint- based questions by a question-and-answer system (QAS) using knowledge graphs, comprising: accessing, by a QAS, a knowledge graph” (abstract “Knowledge Graphs (KG) are multi-relational graphs consisting of entities as nodes and relations among them as typed edges. Goal of the Question Answering over KG (KGQA) task is to answer natural language queries posed over the KG”);
“generating, at the QAS, knowledge graph embeddings based on the knowledge graph via a knowledge graph embedding model” (abstract “In a separate line of research, KG embedding methods have been proposed to reduce KG sparsity by performing missing link prediction. Such KG embedding methods, even though highly relevant, have not been explored for multi-hop KGQA so far. We fill this gap in this paper and propose EmbedKGQA. EmbedKGQA is particularly effective in performing multi-hop KGQA over sparse KGs.” i.e. generating knowledge graph embeddings and §3.3);
While Saxena generally teaches linking, Zhang more specifically teaches “further comprising: learning, at the QAS, a linking model based on training data and an ontology for the training data” (Zhang abstract “We present a new model that does not suffer from this limitation called EntQA, which stands for Entity linking as Question Answering. EntQA first proposes candidate entities with a fast retrieval module, and then scrutinizes the document to find mentions of each candidate with a powerful reader module […] and capitalizes on pretrained models for dense entity retrieval and reading comprehension”);
“processing, at the QAS, the knowledge graph embeddings via the linking model” (Zhang pg. 3
PNG
media_image1.png
410
912
media_image1.png
Greyscale
as shown, the model takes the embeddings and processes it);
“generating, at the QAS, a new knowledge graph embedding based on output of the linking model” (Zhang pg. 2
PNG
media_image2.png
148
852
media_image2.png
Greyscale
)
It would have been obvious to one having ordinary skill in the art at the time that the invention was effectively filed to combine the teachings of Saxena with that of Zhang since “We analyze EntQA and find that its retrieval performance is extremely strong (over 98 top-100 recall on the validation set of AIDA), verifying our hypothesis that finding relevant entities without knowing their mentions is easy” Zhang pg. 2. This shows that by combining the techniques, one would have a better question-answering system.
Saxena further teaches “adding the new knowledge graph embedding to the set of knowledge graph embeddings” (Saxena abstract “KG embedding methods have been proposed to reduce KG sparsity by performing missing link prediction” generating embeddings is the same functionally regardless of where the data comes from)
“receiving, at the QAS and from a user device, a natural language query question having one or more words” (abstract “Goal of the Question Answering over KG (KGQA) task is to answer natural language queries posed over the KG”);
“transforming, at the QAS, the one or more words into one or more question embeddings, each question embedding being a vector representing a corresponding word” (pg. 4 §4.3 “This module embeds the natural language question q to a fixed dimension vector eq ∈ C d .”);
While the references generally teaches constraints, Bao more explicitly teaches “identifying, at the QAS, at least a first constraint” (Bao fig. 1 shows the constraint identified in the second question)
It would have been obvious to one having ordinary skill in the art at the time that the invention was made to combine the teachings of Saxena and Zhang with that of Bao since “we propose a novel systematic KBQA approach to solve multi-constraint questions. Compared to state-of-the-art methods, our approach not only obtains comparable results on the two existing benchmark data-sets, but also achieves significant improvements on the ComplexQuestions” Bao abstract. Therefore by combining the two, the QA system is more robust to complex questions.
Saxena further teaches “and a topic entity in the question embeddings” (pg. 4 §4.3 “Given a question q, topic entity h ∈ E and set of answer entities A ⊆ E, it learns the question embedding”);
“identifying, at the QAS, a plurality of core relation paths in the knowledge graph embeddings, each path linking the topic entity to a different ungrounded entity” (pg. 1 right col. last ¶ “In multi-hop KGQA, the system needs to perform reasoning over multiple edges of the KG to infer the right answer. KGs are often incomplete, which creates additional challenges for KGQA systems, especially in case of multi-hop KGQA” and pg. 6 above §5.3.2 “EmbedKGQA does not limit itself to a sub-graph and utilizing the link prediction properties the KG embeddings, EmbedKGQA is able to infer the relation on missing links”);
“ranking, at the QAS, the plurality of query graphs using a CNN-based similarity scoring model” (Saxena pg. 3 ¶ above §3 “ConvE (Dettmers et al., 2018) utilizes Convolutional Neural Networks to learn a scoring function between the head entity, tail entity and relation. InteractE (Vashishth et al., 2019) improves upon ConvE by increasing feature interaction.” and left col. “Methods like (Dai et al., 2016; Dong et al., 2015; Hao et al., 2017; Lukovnikov et al., 2017; Yin et al., 2016) utilize neural networks to learn a scoring functions to rank the candidate answers” which shows scoring to rank and also Bao pg. 7 fig. 4);
Bao further teaches “associating, at the QAS, the first constraint with each of the core relation paths to generate a plurality of query graphs including at least a first query graph, each query graph being based on a combination of the question embeddings and the knowledge graph embeddings” (Bao pg. 4 “MulCG A MulCG is constructed based on a basic query graph B of a question and an ordered constraint sequence C = {C1, ..., CN } by the following operations: (1) Treat the basic query graph B of the given question as G0; (2) Iteratively add Ci to Gi−1 to generate Gi , by linking the variable vertex of Ci to a v” i.e. generating paths);
“determining, based on a top-ranked query graph and at the QAS, an answer to the query” (Bao pg. 4 into pg. 5 “For each MulCG G ∈ H(Q), a feature vector F(Q, G) is extracted and the one with the highest ranking score is selected. Finally, by executing the MulCG, we get the answers A.”); and
“presenting, via the user device and from the QAS, the answer” (presenting the answer to a user is inherent to any question answering system)
Note that independent claims 8 and 15 recite the same substantial subject matter as independent claim 1, only differing in embodiment. The differences in embodiments, a system and non-transitory computer readable medium are obvious variations of another and inherent to any computing system such as the ones of Saxena and Bao. Therefore the claims are subject to the same rejection.
Regarding claims 2, 9, and 16, the Saxena, Zhang, and Bao references have been addressed above. Bao further teaches “further comprising extending a first core relation path by linking the topic entity to an ungrounded entity, thereby generating the first query graph” (Bao pg. 4 “MulCG A MulCG is constructed based on a basic query graph B of a question and an ordered constraint sequence C = {C1, ..., CN } by the following operations: (1) Treat the basic query graph B of the given question as G0; (2) Iteratively add Ci to Gi−1 to generate Gi , by linking the variable vertex of Ci to a v”)
Regarding claims 3, 10, and 17, the Saxena, Zhang, and Bao have been addressed above. Saxena further teaches “further comprising connecting a first grounded entity of to either a lambda variable or an existential variable” (fig. 2 shows connecting to variable crime, which is ungrounded/unknown along with fig. 1)
Regarding claims 4, 11, and 18, the Saxena, Zhang, and Bao references have been addressed above. Bao further teaches “further comprising: mapping the first constraint to an aggregation function” (Bao table 1 shows mapping the constraints to an aggregation function as it shows multiple types (aggregated) of constraints); and
“attaching the mapped first constraint to either a lambda variable or an existential variable connected to the lambda variable, thereby generating the first query graph” (Bao pg. 2 ¶1 “Motivated by this issue, this work contributes to QA research in the following two aspects: (1) We propose a novel systematic KBQA approach to solve multi-constraint questions by translating a multiconstraint question (MulCQ) to a multi-constraint query graph (MulCG);” generating a graph)
Regarding claims 6, 13, and 20, the Saxena, Zhang, and Bao references have been addressed above. Saxena further teaches “wherein the set of knowledge graph embeddings includes entity embeddings and relation embeddings” (fig. 2 which shows the multiple embeddings from questions and knowledge graphs and pg. 2 right col. ¶1 “KG embedding methods learn high-dimensional embeddings for entities and relations in the KG”)
Regarding claims 7 and 14, the Saxena, Zhang, and Bao references have been addressed above. Bao further teaches “further comprising: extracting superlative linking in the question embeddings” (Bao fig. 1 shows superlative linking i.e. the highest rated answer Even Money is selected); and mapping the superlative linking to an aggregation function” (previous citations, the answer is compared to the constraint/aggregation functions to ensure it is correct)
Regarding claim 22, the Saxena, Zhang, and Bao references have been addressed. Bao further teaches “wherein associating the first constraint with each of the core relation paths comprises: applying the first constraint on partial paths during generation of the core relation paths” (Bao pg. 4 §3.2 “MulCG A MulCG is constructed based on a basic query graph B of a question and an ordered constraint sequence C = {C1,...,CN} by the following operations: (1) Treat the basic query graph B of the given question as G0; (2) Iteratively add Ci to Gi−1 to generate Gi, by linking the variable vertex of Ci to a vertex of Gi−1 with some possible path, or directly merge them as one variable vertex. (3) Output GN. Given”); and
“using the partial paths with the applied constraint to explore next path segments, thereby reducing a query search space needed to produce the answer” (previous citation)
Claim(s) 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Saxena, Zhang, and Bao, further in view of Costabello et al. US 2019/0220524.
Regarding claim 21, the Saxena, Zhang, and Bao references have been addressed above. They do not explicitly teach the remaining claim limitations. Costabello however teaches “further comprising: determining, based on the knowledge graph, a neighborhood of a particular candidate response” ([0036] “As shown in FIG. 1H, and by reference number 150, the prediction platform may determine, based on the knowledge graph and a result for a particular candidate drug (e.g., “Drug 1 has effect on SARS Type B,” and a score of 0.85), a neighborhood of the particular candidate drug (e.g., Drug 1).”);
“generating knowledge graph embeddings for the neighborhood of the particular candidate response” ([0045] “As shown in FIG. 1I, and by reference number 155, the fit/train engine, of the prediction platform, may generate knowledge graph embeddings for the neighborhood of the particular candidate drug.”);
“determining a particular neighborhood with a smallest loss of quality based on the knowledge graph embeddings” ([0047] “the prediction platform may process the revised knowledge graph embeddings (e.g., described above in connection with to FIG. 1E) and the knowledge graph embeddings for the neighborhood of the particular candidate drug (e.g., described above in connection with FIG. 1I), with a loss of quality computation, to determine a particular neighborhood (e.g., a portion of the neighborhood) with a smallest loss of quality for the particular candidate drug”);
“generating a reasoning graph based on an ontology and the particular neighborhood; and generating an explanation of the particular candidate response based on the reasoning graph” ([0055] “As shown in FIG. 1K, and by reference numbers 110 and 165, the prediction platform may process the ontology and the particular neighborhood with the smallest loss of quality, with a reasoning model, to generate a reasoning graph for the particular candidate drug.”)
It would have been obvious to one having ordinary skill in the art a the time that the invention was effectively filed to combine the teachings of Saxena, Zhang, and Bao with that of Costabello since a combination of known methods would yield predictable results. As shown in Costabello, the above methods are known in the art to give better machine learning processing/analyzing. Therefore by determining explanations for predicted links in knowledge graphs combined with the systems above, one would have a better question-answering system.
Claim(s) 23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Saxena, Zhang, and Bao further in view of Fan et al. USPAT 11,386,890.
Regarding claim 23, the Saxena, Zhang, and Bao references have been addressed above. While they generally teach the claim limitations, Fan more explicitly teaches “wherein transforming the one or more words into one or more question embeddings comprises processing the one or more words via: a word embedding layer that processes individual words to provide respective word embeddings” (Fan col. 23 ¶3 “In some embodiments, the query embedder 610 employs the encoder architecture shown in FIG. 8, including various layers of word-level embedder layer 810, bi-directional LSTM layer 815, and a sequence-to-vector pooling layer 820.”);
“a convolution layer including a convolution neural network (CNN) that learns an internal syntactic structure of sentences based on the word embeddings” (previous citation “the seq2vec layer may be implemented by mean-pooling, max-pooling, self-attention, convolution neural network (CNN) pooling, or other types of pooling techniques”);
“an encoder layer including a bi-directional long short-term memory (LSTM) encoder that captures word ordering in both a forward direction and a backwards direction” (previous citation “The query embedder 610 then may include a bi-directional LSTM 815 (or another type of neural network, recurrent neural network, etc.) that processes the word-level embeddings corresponding to the utterance text data 802.”); and
“an attention layer” (previous citation “the seq2vec layer may be implemented by mean-pooling, max-pooling, self-attention, convolution neural network (CNN) pooling, or other types of pooling techniques)
It would have been obvious to one having ordinary skill in the art at the time that the invention was effectively filed to combine the teachings of Saxena, Zhang, and Bao with that of Fan since a combination of known methods would yield predictable results. As shown in Bao the various layers and neural network aspects are known in the art and are common techniques used. These common and known techniques would operate in a known and predictable manner with combined with the teachings above, in order to generate embeddings and learn meaning from data/text.
Conclusion
THIS ACTION IS MADE FINAL. 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 KEVIN W FIGUEROA whose telephone number is (571)272-4623. The examiner can normally be reached Monday-Friday, 10AM-6PM 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, MIRANDA HUANG can be reached at (571)270-7092. 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.
KEVIN W FIGUEROA
Primary Examiner
Art Unit 2124
/Kevin W Figueroa/ Primary Examiner, Art Unit 2124
Prosecution Insights