Patent Application 15435070 - INTELLIGENT MATCHING SYSTEM WITH ONTOLOGY-AIDED

Title: INTELLIGENT MATCHING SYSTEM WITH ONTOLOGY-AIDED RELATION EXTRACTION

Application Information

Invention Title: INTELLIGENT MATCHING SYSTEM WITH ONTOLOGY-AIDED RELATION EXTRACTION
Application Number: 15435070
Submission Date: 2025-05-21T00:00:00.000Z
Effective Filing Date: 2017-02-16T00:00:00.000Z
Filing Date: 2017-02-16T00:00:00.000Z
National Class: 707
National Sub-Class: 739000
Examiner Employee Number: 77864
Art Unit: 2163
Tech Center: 2100

Rejection Summary

102 Rejections: 0
103 Rejections: 6

Cited Patents

The following patents were cited in the rejection:

US 9087043🔗
US 9792277🔗

Office Action Text

DETAILED ACTION

In view of the appeal brief filed on 1/21/2025, PROSECUTION IS HEREBY REOPENED. A new ground of rejection is set forth below.
To avoid abandonment of the application, appellant must exercise one of the following two options:
(1) file a reply under 37 CFR 1.111 (if this Office action is non-final) or a reply under 37 CFR 1.113 (if this Office action is final); or,
(2) initiate a new appeal by filing a notice of appeal under 37 CFR 41.31 followed by an appeal brief under 37 CFR 41.37. The previously paid notice of appeal fee and appeal brief fee can be applied to the new appeal. If, however, the appeal fees set forth in 37 CFR 41.20 have been increased since they were previously paid, then appellant must pay the difference between the increased fees and the amount previously paid.
A Supervisory Patent Examiner (SPE) has approved of reopening prosecution by signing below:

Claims 4, 6-8, 10-15 are pending in the application.

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 . 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 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.

Response to Arguments
In response to the Appeal Brief filed 4/15/2024, please see a new combination of references cited below.

Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 4, 6-8, 10-15 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Claims 4, 6-8, 10-15 fall within the statutory category of a process.
Step 2A, Prong One: the claims recite a Judicial Exception.
As per claim 15, the limitations of “A method to match properties to requirements using text-to-ontology mapping for automatic extraction of structured information from free-form text provided by a user comprising the steps of: defining a knowledge graph and an ontology associated with said knowledge graph based on a plurality of candidate data sets; generating a taxonomy of named entities contained in said knowledge graph correlated to relations of said named entities and meta-data associated with said named entities; defining a contextual ontology from said candidate data sets, wherein said contextual ontology facilitates a structured representation of parameters of interest in free-form text input context; generating semantically-meaningful dense word embedding vectors associating a set of language-specific text-matching attributes with nodes in said ontology associated with said knowledge graph and nodes in said contextual ontology and said knowledge graph; receiving a search query, including user provided free-form text; analyzing said user provided free-form text of said search query by performing part-of-speech tagging and dependency parsing on said user provided free-form text of said search query; performing named entity recognition and named entity linking on said user provided free-form text of said search query to associate parts of said user provided free-form text of said search query with known named entities described in said knowledge graph; performing a semantic parsing of said user provided free-form text of said search query, extracting semantic relations by traversing a dependency parsed tree and applying a classifier to detect mappings between phrases in said user provided free-form text of said search query and nodes in said contextual ontology; and scoring detected mappings against a plurality of candidate data sets”,
under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of mathematical concepts including recited “generating semantically-meaningful dense word embedding vectors associating a set of language-specific text-matching attributes with nodes in said ontology associated with said knowledge graph and nodes in said contextual ontology and said knowledge graph” where dense vectors are type of mathematical objects that normally represent data in machine learning.
Since a knowledge graph which is a graph-based representation that connects entities, e.g., concepts, places, people through relationships. Each node represents an entity and edges represent relationships between entities. Thus, properties or entities from free-form text of data sets/documents/inputs are extracted and mapped in relating to classifications/ontology.
Part-of-speech tagging assigns grammatical category to each word in a text, together with the dependency parsing on the search query to analyze the grammatical structure of the text and relationship between related words to determine semantic relations and scoring for mappings of data sets.
If a claim limitation, under its broadest reasonable interpretation, covers the performance of the limitation in the mind with the usage of pen and paper and mathematical concepts, then it falls within the "Mental Processes" and “Mathematical Concepts” groupings of abstract ideas. Accordingly, the claims recite an abstract idea.
Step 2A, Prong Two: exception is not integrated into a practical application.
The judicial exception is not integrated into a practical application because the additional elements and combination of additional elements do not impose meaningful limits on the judicial exception. In particular, the claims recite the additional elements “generating semantically-meaningful dense word embedding vectors” in claim 15 do not meaningfully limit the abstract idea, thus, not significantly more than the abstract idea itself. Thus, claim 15 is directed to abstract ideas.
Step 2B: “Inventive Concept” or “Significantly More”
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements when considered both individually and as an ordered combination do not amount to significantly more than the abstract idea. Here, said claims do not recite specific limitations (alone or when considered as an ordered combination) that were not well understood, routine, and conventional. Looking at the limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. There is no indication that the combination of elements improves the functioning of the computer or improves another technology. The claims do not amount to significantly more than the underlying abstract idea.
Claims 4,6-8, 10-14 add further limitations which are also directed to an abstract idea. The claims recite steps of processing named entity to ontology associations to resolve ambiguities and prune matches, to arrive at a context-specific structured representation of said user provided free-form text of said search query; extract named entities from said user provided free-form text of said search query; ascertain semantic relations between extracted named entities and other constituents of the user provided free-form text of said search query signifying an action (predicate) and a subject; map said semantic relations to parameters in a contextual ontology; wherein said contextual ontology comprises a graph-based representation of properties of a client entity, a vendor entity, and project requirements; tokenization of said user provided free-form text of said search query; wherein said step of part-of-speech tagging uses a trained conditional random field model; the step of shallow parsing said user provided free-form text of said search query and applying text classification; the step of normalizing phrases identified by said step of shallow parsing; assigning one or more word embedding vectors to named entities; identifying a node in said dependency parsed tree which overlaps a corresponding named entity token in said user provided free-form text of said search query which is the least distance from a root of the parsed tree corresponding to said identified named entity; traversing the dependency parsed tree upward from said node and establishing a candidate relation predicate by materializing a word embedding vector corresponding to each word encountered along a path to said root; and determine a minimum distance between any of the embedded word vectors and a word vector associated with the named entity; the step of backtracking from the root of the tree and traversing down each sub-tree of the dependency parsed tree, in breadth-first search (BFS) order; and collecting any word found which is classified as a subject above a certain pre-defined confidence score threshold.”
The “part-of-speech tagging uses a trained conditional random field model” is a type of probabilistic graphical model often used in Natural Language Processing (NLP) and computer vision tasks;
“shallow parsing said user provided free-form text of said search query and applying text classification”; “normalizing phrases identified by said step of shallow parsing” are NPL techniques of identify and labels segments of a sentence: nouns, verbs, adjectives, etc.
“one or more word embedding vectors to named entities” and “breadth-first search (BFS)” are machine learning algorithms, e.g., for search a tree data structure for a node that satisfies a given property.
Said steps can be performed using human mental analyzing and evaluation with pen and paper, and applying mathematical concepts which fall into the abstract idea of a mental process and mathematical concepts groupings of abstract ideas, similar to the independent claims. Because the additional elements do not impose meaningful limitations on the judicial exception and the additional elements are well-understood, routine and conventional functionalities in the art, the claims are directed to an abstract idea and are not patent eligible.

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 15, 6, 8 are rejected under 35 U.S.C. 103 as being unpatentable over Liang et al. (US 9087043) in view of Malon et al. (US 20140236578).
Specification, para. 5-7 discloses properties attributed to an entity, e.g., a service provider, a company etc. and requirements specified by a user, parameters specifying customer requirements or needs;
para. 26: “The contextual ontology is a graph-based representation which describes properties of interest for each defined entity”. Thus, the contextual ontology is the ontology for a specific class/entity.

As per claim 15, Liang et al. teaches
a method to match properties to requirements using text-to-ontology mapping for automatic extraction of structured information from free-form text provided by a user; comprising the steps of: defining a knowledge graph and an ontology associated with said knowledge graph based on a plurality of candidate data sets (para. 23: the service provider can analyze a document based on a knowledge base. Individual text segments (e.g., sentences, phrases, words, etc.) of the document can be tagged with classes of a model such as an ontology. Phrases matching criteria such as described herein, but not included in the knowledge base, can be associated with classes of the ontology; para. 77-78: ontology 400 includes interconnected classes or hierarchies of classes. In some implementations, classes and subclasses in the ontology are defined and arranged in a taxonomy, as represented by nested blocks in fig. 4);
generating a taxonomy of named entities contained in said knowledge graph correlated to relations of said named entities and meta-data associated with said named entities (para. 79-80: ontology is used to express relationships between the different classes, which can provide for concise expression of pertinent information included in actionable items; any particular class and its subclasses can be represented as trees or graphs of nodes in a taxonomy for that class; para. 215, 226: knowledge base can include data indicating specific entities that should be given higher or lower scoring values. For example, in a cluster 106 management context, entities such as “virtual machine,” “directory,” “server,” or “host” can be given higher scoring values, and entities such as “word processor” or “game” can be given lower scoring values);

defining a contextual ontology from said candidate data sets, wherein said contextual ontology facilitates a structured representation of parameters of interest in free-form text input context (para. 22: a document can include a sentence regarding a problem that a user is having with a software element, a sentence regarding a feature of the software element for which the user has indicated an interest (e.g., a feature the user likes), a sentence that is unrelated to the software element, and so on; para. 145: the predetermined grammar pattern for the suggestion motif defines a first class of the ontology preceding a second class of the ontology);

associating a set of language-specific text-matching attributes with nodes in said ontology associated with said knowledge graph and nodes in said contextual ontology and said knowledge graph (para. 77-80: question indicates that associated portion(s) of the user text represent Indicator questions; the mapping of the phrases to the ontology can be stored in the knowledge base, e.g., as discussed below with reference to fig. 5; any particular class and its subclasses can be represented as trees or graphs of nodes in a taxonomy for that class; para. 217, 129: determine an association between a first text segment, e.g., a word or phrase in the free-form user text, and a first individual class of a model, e.g., of ontology. The classification module determines the association, e.g., based at least in part on a characteristic pattern associated with the first text segment. Examples are discussed below with reference to figs. 6-15, updates the dictionary to include the association);

receiving a search query, including user provided free-form text; analyzing said user provided free-form text of said search query; known named entities described in said knowledge graph (para. 197: presenting the recommendation including ranked responses to a search query; para. 80-82: as illustrated by the nested blocks, any particular class and its subclasses can be represented as trees or graphs of nodes in a taxonomy for that class. The entity class can include subclasses such as Physical-entity subclass or Virtual-entity subclass. For example, a Physical Entity can be a tangible object such as an accelerometer, a gaming console. A Virtual Entity can be an intangible object such as a protocol, reference, variable, library, or method. Other examples of entities can include services, e.g., cloud services, software entities, replaceable entities, and logical entities; para. 105: the knowledge base includes at least one of an ontology, a dictionary, and a pattern set including one or more grammar pattern(s));
performing a semantic parsing of said user provided free-form text of said search query, extracting semantic relations by traversing a dependency parsed tree and applying a classifier to detect mappings between phrases in said user provided free-form text of said search query and nodes in said contextual ontology (para. 77-78: fig. 4 shows an example ontology 400 useful for representing the “semantic interpretation” of domain knowledge for a domain. The mapping of the phrases to the ontology can be stored in the knowledge base; para. 210-212: fig.10 shows parses of an example sentence. The sentence is “LINUX is not, thanks to you, crashing.” Boxes represent nodes. Parse 1002 is a dependency parse of the sentence, and arrows are labeled with the corresponding Universal Dependency relationships; para. 70: phrase-extraction module, the phrase-filtering module, the mapping module, the analysis module, the recommendation module, and the reporting module);

and scoring detected mappings against a plurality of candidate data sets (para. 243-245: representation(s) can be determined of located phrase(s) and candidate phrase(s). An individual representation can be associated with a located phrase or with both a located phrase and a candidate phrase. The corrected text segment 608 can be selected from among the plurality of stored candidate text segments based at least in part on the representation(s), e.g., the respective distances. For example, the candidate text segment corresponding to the smallest of the distances can be selected as the corrected text segment; para. 216, 221-226: scoring values can be determined based on particular instances in document(s) of selected classes);
by performing part-of-speech tagging (para. 123: perform part-of-speech processing to select phrases remaining after filtering that match predetermined (or otherwise selected) part-of-speech patterns; para. 128: the characteristic pattern for a text segment can include a tag sentence, as described below, a sequence of part-of-speech tags, or other forms described herein);
and dependency parsing on said user provided free-form text of said search query; performing named entity recognition and named entity linking on said user provided free-form text of said search query to associate parts of said user provided free-form text of said search query with known named entities (para. 23: phrases matching criteria such as described herein, but not included in the knowledge base, can be associated with classes of the ontology. Examples of such phrases can include technical phrases, e.g., domain-specific phrases, product names, or component names; para. 172: the characteristic pattern can include a dependency tree expressing relationships between words, e.g., in the Universal Dependencies set; para. 175, 210-212: fig. 10 shows parses of an example sentence. The sentence is “LINUX is not, thanks to you, crashing.” Boxes represent nodes. Parse is a dependency parse of the sentence, and arrows are labeled with the corresponding Universal Dependency relationships. The nodes are “LINUX”, “is”, “not”, “thanks”, “to”, “you”, and “crashing”; the dependency parse correctly identifies that the words “is,” “not,” and “crashing” are related, notwithstanding the interspersed “thanks to you”).
Liang et al. does not explicitly disclose dense word embedding vectors.
Malon discloses
generating semantically-meaningful dense word embedding vectors (para. 21-23: feature representations are dense vectors in a continuous feature space; for the terminal nodes, they are the word vectors in a neural probabilistic language model; para. 53: the present system extends the language model vectors with a random vector associated to each distinct word. The random vectors are fixed for all the words in the original language model, but a new one is generated the first time any unknown word is read. For known words, the original dimensions give useful syntactic and semantic information.)
Malon also teaches
by performing part-of-speech tagging and dependency parsing on said user provided free-form text of said search query; performing named entity recognition and named entity linking on said user provided free-form text of said search query to associate parts of said user provided free-form text of said search query with known named entities (para. 21: extract answers to arbitrary natural language questions from supporting sentences; para. 44: through the ranking and tagging tasks, this model learned embeddings of each word in a 50-dimensional space. Besides this learned representations, we encode capitalization and SENNA's predictions of named entity and part of speech tags with random vectors associated to each possible tag, as shown in fig. 1; para. 55-56: convolving over token sequences has achieved state-of-the-art performance in part-of-speech tagging, named entity recognition, and chunking, and competitive performance in semantic role labeling and parsing, using one basic architecture. Classifying tokens to answer questions involves not only information from nearby tokens, but long range syntactic dependencies).
Thus, 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 system of Liang et al. to include the teachings of Malon, for the advantage of parsing and analyzing user inputs in order to better recognize words and related words in natural language that focus on entities/concepts and provide relevant results to user queries and for the advantage of identifying semantically related words that helps provide the most relevant search results to the users.

As per claim 6, Liang et al. teaches
extract entities from said user provided free-from input text of said search query (para. 197: presenting the recommendation including ranked responses to a search query; para. 80-82: as illustrated by the nested blocks, any particular class and its subclasses can be represented as trees or graphs of nodes in a taxonomy for that class. The entity class can include subclasses such as Physical-entity subclass or Virtual-entity subclass. For example, a Physical Entity can be a tangible object such as an accelerometer, a gaming console. A Virtual Entity can be an intangible object such as a protocol, reference, variable, library, or method. Other examples of entities can include services, e.g., cloud services, software entities, replaceable entities, and logical entities; para. 226: knowledge base can include data indicating specific entities that should be given higher or lower scoring values. For example, in a cluster management context, entities such as “virtual machine,” “directory,” “server,” or “host” can be given higher scoring values, and entities such as “word processor” or “game” can be given lower scoring values);
ascertain semantic relations between extracted named entities and other constituents of the user provided free-form input text of said search query signifying an action (predicate) and a subject; map said semantic relations to parameters in a contextual ontology (para. 77-79: an ontology representing the "semantic interpretation" of domain knowledge for a domain. Ontology includes interconnected classes or hierarchies of classes; individual word(s) or phrase(s) appearing in document(s) 108 can be mapped to the classes of an ontology. The mapping of the phrases to the ontology can be stored in the knowledge base; para. 114: identify a subset of the phrases that have relatively significant meaning, e.g., that may contribute to understanding the actionable item or other motif in the document; fig. 4: action 402, entity 410; fig. 9);
Liang et al. does not explicitly teach named entities.
Malon et al. teaches
extract named entities from said user provided free-from input text of said search query (para. 21: extract answers to arbitrary natural language questions from supporting sentences; para. 44: through the ranking and tagging tasks, this model learned embeddings of each word in a 50-dimensional space. Besides this learned representations, we encode capitalization and SENNA's predictions of named entity and part of speech tags with random vectors associated to each possible tag; para. 55-56: convolving over token sequences has achieved state-of-the-art performance in part-of-speech tagging, named entity recognition, and chunking, and competitive performance in semantic role labeling and parsing).
Thus, 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 system of Liang et al. to include the teachings of Malon, for the advantage of parsing and analyzing user inputs in order to better recognize words and related words in natural language that focus on entities/concepts and provide relevant results to user queries and for the advantage of identifying semantically related words that helps provide the most relevant search results to the users.

As per claim 8, Liang et al. teaches
the step of tokenization of said user provided free-form text of said search query (para. 111: text in the documents can be tokenized into sentences. The phrase-extraction module can use a compression algorithm to operate on the tokenized
documents; para. 158).

Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Liang et al. (US 9087043) in view of Malon et al. (US 20140236578) and further in view of Erpenbach (US 20180113867).
As per claim 4, Malon et al. teaches
processing named entity to ontology associations, to arrive at a context-specific structured representation of said user provided free-form input text of said search query (para. 21: extract answers to arbitrary natural language questions from supporting sentences; para. 44: through the ranking and tagging tasks, this model learned embeddings of each word in a 50-dimensional space. Besides this learned representations, we encode capitalization and SENNA's predictions of named entity and part of speech tags with random vectors associated to each possible tag; para. 55-56: convolving over token sequences has achieved state-of-the-art performance in part-of-speech tagging, named entity recognition, and chunking, and competitive performance in semantic role labeling and parsing).
Liang and Malon do not teach to resolve ambiguities and prune matches.
Erpenbach teaches said limitation at para. 33-35: passage search results require more detailed analysis of the passage text to identify candidate answers. As one example of such analysis, named entity detection may be used to extract candidate answers from the passage; applies lightweight (less resource intensive) scoring techniques to a larger set of initial candidates to prune them down to a smaller set of candidates before the more intensive scoring is undertaken; para. 44: provide techniques by which a user can override the QA processing to fill in missing data, correct errors, resolve ambiguities, etc.
Thus, 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 system of Liang and Malon et al. to include the resolve ambiguities and prune matches of Erpenbach, for the advantage of clarifying questions and/or answers and providing the most relevant search results to the users.

Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Liang et al. (US 9087043) in view of Malon et al. (US 20140236578) and further in view of Pye et al. (US 20130110573).
As per claim 7, Liang et al. teaches
wherein said contextual ontology comprises a graph-based representation of properties of a client entity, a vendor entity, and project requirements (para. 53-54, 85, 23: individual text segments (e.g., sentences, phrases, words, etc.) of the document can be tagged with classes of a model such as an ontology. Phrases matching criteria such as described herein, but not included in the knowledge base, can be associated with classes of the ontology. Examples of such phrases can include technical phrases, e.g., domain-specific phrases, product names, or component names; para. 94: properties of entities; para. 106, 234: the selected keyword is or includes a product name or other entity name; para. 26: consumer and restaurant or another business service).
Liang and Malon et al. do not explicitly teach project requirements.
Pye et al. teaches said limitations at para. 40: historical component requirements could include project firm requirements, project client requirements, project jurisdiction requirements, or other types of requirements that can define a context of the project; para. 9, 52: project requirements.
Thus, 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 system of Liang and Malon to include project requirements, as taught by Pye, for the advantage of providing an efficient search program that allow sufficient parsing, searching and matching features between entities, thus better identify and display to users the most relevant search results.

Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Liang et al. (US 9087043) in view of Malon et al. (US 20140236578) and further in view of Dahlmeier et al. (US 20130325442).
As per claim 10, Malon teaches at para. 24: in fig. 1 words are entered into an original language model database which are fed to an n-dimensional vector. The same word is provided to a randomizer that generates an m-dimensional vector. The result is an (n+m) dimensional vector that includes the original part and the random part
Liang and Malon do not teach uses a trained conditional random field model.
Dahlmeier et al. teaches
part-of-speech tagging uses a trained conditional random field model (para. 89-92: FIG. 7 is an example tagging of a training sentence for the factorial conditional random fields (CRF). A sentence may be divided into words and each word tagged with a word-layer tag and a sentence-layer tag; para. 115: the system relies on shallow syntactic and lexical features derive from a chunker, including the words before, in and after the NP/noun phrase, the head word and POS tags).
Thus, 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 system of Liang and Malon to include a trained conditional random field model, as taught by Dahlmeier, for the advantage of providing an efficient search program that allow sufficient parsing, analyzing, and searching common features between entities, thus better identify and display to users the most relevant search results.

Claims 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Liang et al. (US 9087043) in view of Malon et al. (US 20140236578) and further in view of Dahlmeier et al. (US 20130325442) and Srinivasan (US 9792277).
As per claims 11-12, Liang et al. teaches
parsing said user provided free-form text of said search query and applying text classification (para. 25: analyze text in the form in which the text is presented (e.g., analyze a text segment, such as a text segment, based on surrounding text segments, such as surrounding words or phrases). This mapping can be performed in a hierarchical manner, e.g., by mapping portions of a parse tree to respective classes of a model; para. 197: presenting the recommendation including ranked responses to a search query; para. 80-82: as illustrated by the nested blocks, any particular class and its subclasses can be represented as trees or graphs of nodes in a taxonomy for that class; figs. 6, 11);
normalizing phrases identified by said step of parsing (para. 227: block 1124 can include computing a similarity metric, e.g., the dot product of two of the respective feature vectors (e.g., as they are, or after normalization)).
Malon teaches shallow parsing/chunking at para. 55.
Liang and Malon et al. do not explicitly teach shallow parsing.
Dahlmeier et al. teaches
shallow parsing (para. 114-115: Examples of feature extraction for article errors include “DeFelice”, “Han', and “Lee’. DeFelice The system for article errors uses a CCG parser to extract a rich set of Syntactic and semantic features, including part of speech (POS) tags, hypernyms from WordNet, and named entities. Han - The system relies on shallow syntactic and lexical features derived from a chunker, including the words before, in, and after the NP, the head word, and POS tags. Lee The system uses a constituency parser. The features include POS tags, Surrounding words, the head word, and hypernyms from WordNet; para. 78: Z(x) is a normalization factor to ensure a well-formed probability distribution).
Thus, 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 system of Liang, Malon and the shallow parsing of Dahlmeier where the usage of shallow parsing allows the identifying/extracting the most important phrases and thus, making it faster and more efficient than full parsing and for the advantage of providing an efficient search program that is efficient and computationally lightweight, ideal for processing large corpus of text thus provides sufficient parsing for data analyzing and searching common features between entities/concepts.
Even if Liang, Malon and Dahlmeier do not explicitly teach normalizing phrases identified by said step of parsing,
Srinivasan teaches
parsing said user provided free-form text of said search query and applying text classification; normalizing phrases identified by said step of parsing (figs. 3, 12: document/text classifier; fig. 6: format normalization, structure normalization, and sentence detection module; col. 6:53-67; col. 10:45-54; col. 14:24-64: fig 10 illustrates an exemplary embodiment of a block diagram for the linguistic analysis layer according to one or more embodiments of the invention. The linguistic analysis layer 532 can be configured to include various modules that are configured to identify clauses and phrases or concepts in the sentences and the correlation there between.…the conjunction resolution module 1004 can be configured to separate sentences with conjunctions into its constituent concepts. For example, if the sentence is "Elephants are found in Asia and Africa", the conjunction resolution module 1004 split the sentence into two different sub sentences. The first sub-sentence is "Elephants are found in Asia" and the second sub-sentence is "Elephants are found in Africa". The conjunction resolution module 1004 can process complex concepts so as to aid normalization).
Thus, 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 system of Liang, Malon, Dahlmeier to include text phrase normalization in natural language processing, as taught by Srinivasan, for the advantage of reducing the amount of different information that the computer has to process, and therefore improves efficiency.

Claims 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over Liang et al. (US 9087043) in view of Malon et al. (US 20140236578) and further in view of Govrin et al. (US 20140297268).
As per claim 13, Liang does not explicitly teach said claim.
Malon teaches
assigning one or more word embedding vectors to named entities; identifying a node in said dependency parsed tree which overlaps a corresponding named entity token in said user provided free-form text of said search query which is the least distance from a root of the parsed tree corresponding to said identified named entity (para. 7: answer free form questions using recursive neural network (RNN) includes defining feature representations at every node of a parse trees of questions and supporting sentences, when applied recursively, starting with token vectors from a neural probabilistic language model; and extracting answers to arbitrary natural language questions from supporting sentences; para. 22, 56-57: classifying tokens to answer questions involves not only information from nearby tokens, but long range syntactic dependencies. By using feature representations from our RNN and performing convolutions across siblings inside the tree, instead of token sequences in the text, we can utilize the parse tree information in a more principled way. We start at the root of the parse tree and select branches to follow, working down. At each step, the entire question is visible, via the representation at its root, and we decide whether or not to follow each branch of the support sentence);
traversing the dependency parsed tree upward from said node and establishing a candidate relation predicate by materializing a word embedding vector corresponding to each word encountered along a path to said root (para. 22-23: follow each parse tree node of a support sentence or not, by classifying its RNN embedding together with those of its siblings and the root node of the question, until reaching the tokens it selects as the answer. The positively classified nodes are followed down the tree, and any positively classified terminal nodes become the tokens in the answer. Feature representations are dense vectors in a continuous feature space; para. 33: fig. 7 shows an example of how the tree of fig. 4 is populated with features at every node using the autoencoder E with features at terminal nodes X5, X6, and X10- X15; para. 57: start at the root of the parse tree and select branches to follow, working down. At each step, the entire question is visible, via the representation at its root, and we decide whether or not to follow each branch of the support sentence);
determine a distance between any of the embedded word vectors and a word vector associated with the named entity (para. 22: follow each parse tree node
of a support sentence or not, by classifying its RNN embedding together with those of its siblings and the root node of the question, until reaching the tokens it selects as the answer; para. 31, 38: the decoder and encoder may be trained together to minimize reconstruction error, typically Euclidean distance; para. 44: through the ranking and tagging tasks, this model learned embeddings of each word in a 50-dimensional
space).
Liang and Malon do not explicitly disclose a minimum distance between any of the embedded word vectors and a word vector associated with the named entity.
Govrin et al. teaches
determine a minimum distance between any of the embedded word vectors and the word vector associated with the named entity (para. 177: Run Minimum Edit Distance (MED) algorithm--to locate pairs of words at the input and output text; para. 632-633: the process may not limit its search to the currently discussed topic branch (from the current topic to the root), but may instead visit the whole tree representing all previously discussed topics using a suitable tree searching process such as but not limited to DFS or BFS. Fig. 22 presents an example with three nested topic context nodes and two embedded entities. The process starts from the bottom topic context and makes its way to the root topic context. The process visits entities enroute. The "Topic to Topic" table and "Entity to Topic" table are used to map visited topics and entities to related topics and to create a resulting fork prediction vector).
Thus, 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 system of Liang, Malon to include a minimum distance between any of the embedded word vectors and the word vector associated with the named entity, as taught by Govrin, for the advantage of identify common features between entities and display to users the most relevant search results.

As per claim 14, Liang teaches at para. 210-211: parse is a dependency parse of the sentence, and arrows are labeled with the corresponding Universal Dependency
Relationships; para. 133: provides a confidence value or other indication of the confidence of the class determination based on the feature vector; compare the confidence value to a predetermined threshold; para. 189: If the confidence value exceeds the threshold, the first individual class can be determined to be the
second candidate class from the classifier; fig. 12: associate word or phrase with class of ontology to provide classification, determine class collection(s) for document(s) based at least in part on classification.

Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wroczynski et al. (US 20140136188) teaches at para. 19: extract information about product features from reviews you could use a constituency parser or a dependency parser but you need to write complex algorithms to search through the parse tree.
Leliwa et al. (US 20170017635) teaches at para. 21: Vectorization of words and phrases is a good example of a very successful attempt. Word embedding is a process of mapping words (or phrases in phrase embedding) from the vocabulary to vectors of real numbers. The word embedding tools take a text corpus as input, construct a vocabulary from training text data, learn vector representation of words and deliver the word vectors as output. Basically, this approach is based on the following hypothesis: words that appear in similar contexts have similar meaning. Vector representation allows to perform vector operations such as finding shortest distance between words (e.g. “France” is very close to “Spain” or “Belgium”) or arithmetic operations (e.g. “king−man+woman” is very close to “queen”). Vectorization is a relatively new and powerful approach that can automatically provide very useful knowledge to other NLP systems and therefore allow using supervised learning with much less labeled data to train accurate models. It can enrich current methods of getting actionable answers from text data in the same way as syntactic parsers enrich these methods by unveiling grammar dependencies between words and phrases.
Mirowski et al. (US 20120150532) teaches at para. 9-10: the list of predicted words can be used for a variety of different natural language applications, such as part-of-speech tagging, spelling correction, language generation, and speech recognition. Further, the continuous statistical language model can be based on a long-range dependency of a current topic according to a paragraph level and/or document level.
Mohammed (US 20140025660) teaches at para. 65: entity nodes may include, for example, nodes representing one or more entities, locations, organizations, businesses, places, things, and/or the like associated with the user; para. 107: the personal ontology graph 100 may be further populated based on inferred interests provided by a higher-level inference engine.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LINH BLACK whose telephone number is (571)272-4106. The examiner can normally be reached on 9AM-5PM EST M-F.
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, Tony Mahmoudi can be reached on 571-272-4078. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see https://ppair-my.uspto.gov/pair/PrivatePair. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/LINH BLACK/Examiner, Art Unit 2163 5/3/2025

/TONY MAHMOUDI/Supervisory Patent Examiner, Art Unit 2163