DETAILED ACTION
Status of Claims
This Office action is responsive to communications filed on 2025-09-04. Claim(s) 1-20 is/are pending and are examined herein.
Claim(s) 1-20 is/are objected to.
Claim(s) 4-14 and 18-20 is/are rejected under 35 USC 112(b).
Claim(s) 1-20 is/are rejected under 35 USC 101.
Claim(s) 1-20 is/are rejected under 35 USC 103.
Notice of Pre-AIA or AIA Status
The present application, filed on or after 2013-03-16, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Regarding objections for informalities and rejections under 35 USC 112(b), the applicant’s amendments resolve some, but not all, of the concerns raised in the previous Office action. The amendments also introduce further concerns. Unresolved and newly introduced concerns are described below.
Regarding rejections under 35 USC 101, the applicant’s remarks have been fully considered.
Regarding the previously filed claims, the applicant “believes that the claim as originally drafted recited patent-eligible subject matter” [remarks, page 11] but provides no rationale in support of this belief. This unsupported statement fails to comply with 37 CFR 1.111(b) because it amounts to a general allegation that the previously filed claims may define a patentable invention without specifically pointing out any reasons therefor.
Regarding the amended claims, the applicant argues that the step of “identifying noun phrases and generating vector embeddings of the noun phrases” renders the claims eligible [remarks, pages 10-11]. The examiner respectfully disagrees. It is clear that a human mind can identify noun phrases. A human mind can also perform actions falling under the broadest reasonable interpretation of “generating vector embeddings” of noun phrases (the broadest reasonable interpretation of a vector encompasses any sequence of elements, so even, for example, an act of transcribing noun phrases as sequences of characters could be regarded as an act of “generating vector embeddings of the noun phrases” as recited by the claim). In other words, the limitation as recited is merely another mental process and does not help render the claim eligible.
The complete 101 analysis, updated in view of the applicant’s amendments, is given below.
Regarding rejections under 35 USC 103, the applicant’s remarks have been fully considered.
Regarding the previously filed claims, the applicant asserts that they are “not conceding that the subject matter encompassed by the claims prior to this Amendment is unpatentable over the art cited by the Examiner” [remarks, page 9] but provides no rationale in support of this statement of non-concession. This unsupported statement of non-concession fails to comply with 37 CFR 1.111(b) because it amounts to a general allegation that the previously filed claims define a patentable invention without specifically pointing out how the language of the previously filed claims patentably distinguishes them from the references.
Regarding the amended claims, the applicant argues that Huetle in view of Sun fails to disclose segmentation of text into sentences [remarks, pages 12-13]. The argument is moot in view of the new grounds of rejection as given below.
The applicant attempts to argues that “Huetle should be removed as a reference” because it “expressly removes segmentation by eliminating punctuation and capitalization”, thereby teaching away from sentence segmentation since “text without capitalization and punctuation cannot be segmented into sentences” [remarks, page 13]. This line of argumentation is unpersuasive for at least the following reasons:
First, the claim itself includes no limitations regarding punctuation and/or capitalization, so the fact that Huetle removes punctuation and/or capitalization is unconvincing as a point of contrast against the claimed invention. (In fact, the applicant’s specification also makes no explicit remarks regarding either punctuation or capitalization.)
Second, nothing explicitly recited in the pending claims necessitates that, when comparing the claimed invention against the disclosures of Huetle, the sentence segmentation recited by the claim must be performed after the removal of capitalization and punctuation described in Huetle. The claims recite merely a step of segmenting text into sentences without indicating anything about the stage at which the segmentation is to be performed. The original documents described in Huetle have both capitalization and punctuation (cf. [Huetle, figure 1]), which means that a person of ordinary skill in the art could apply even a standard method of sentence segmentation (using, e.g., NLTK as cited in the conclusion of this Office action) to the documents as disclosed in Huetle.
Third, in the combination as proposed in the previous Office action (and below), the keyword extractor of Huetle is anyway replaced by the keyword extractor of Sun, and Sun does not explicitly describe a step of removing punctuation and/or capitalization. In other words, the combination of Huetle and Sun may not even necessitate the removal of punctuation and/or capitalization described in Huetle.
Fourth, the examiner respectfully disagrees with the applicant’s assertion that it is impossible to perform sentence segmentation without capitalization and punctuation. A person listening to another’s speech is able to segment that speech into sentences, despite the fact speech is not marked with capitalization and punctuation. Everyone who has transcribed speech into properly capitalized and punctuated text has performed this purportedly impossible act of sentence segmentation. Even if one restricts the scope of the discussion only to writing, neither capitalization nor punctuation is actually essential for sentence segmentation. For example, Japanese orthography has no system of capitalization, and it included almost no punctuation up until the Meiji period, but practiced readers of pre-Meiji Japanese texts can nonetheless segment these texts into sentences and understand their contents. Moreover, it is not just human beings that can perform this purportedly impossible act of sentence segmentation, since general-purpose computers implementing modern methods of natural language processing can emulate the ability of human beings to perform sentence segmentation without proper capitalization and/or punctuation. The applicant is invited to consult, for example, Matusov as cited in the conclusion of this Office action.
The complete prior art rejection, updated in view of the applicant’s amendments, is given below.
Examiner’s Remarks
Claims 6, 13, and 20 recite an entity importance model H(v_e). In view of the originally filed disclosure, H(v_e) is equal to w_e, i.e., to the importance of the entity e represented by the vector embedding v_e provided as input to the model (cf. [specification, 0050; claims 6, 13, and 20 as originally filed]). In other words, the broadest reasonable interpretation, in view of the specification, of the “entity importance model H(v_e)” recited in these dependent claims encompasses at least the “neural network model” and/or the “trained importance calculator” recited in their respective parent claims. These claim elements are interpreted accordingly herein. The examiner suggests replacing “an entity importance model” with either “the neural network” or “the trained importance calculator” for consistency of nomenclature and clarity of the claimed subject matter.
Claim Objections
Claim(s) 1-20 is/are objected to because of the following informalities:
Claims 1 and 15 recites segmenting text from the specification sheet and segmenting text from the plurality of descriptive sheets into sentences [emphasis added] this should be “segmenting text from the specification sheet and into sentences and segmenting text from the plurality of descriptive sheets into sentences”). Dependent claims 2-7 and 16-20 inherit the objection.
Claims 1, 8, and 15 recites each identified technical feature in the first set of technical features represented by the vector embeddings and each identified technical feature in the second set of technical features represented by the vector embeddings [emphasis added] but this should be “each identified technical feature in the first set of technical features
Claim 8 recites a feature importance calculator. This should be “the trained importance calculator” for proper antecedent basis. Dependent claims 9-14 inherit the objection. (The applicant is also invited to consult the suggestion regarding the indefiniteness rejection of claim 8 below.)
Claim 15 recites causes the computer to perform the steps of: [emphasis added] but “the steps” lacks antecedent basis. The examiner suggests “causes the computer to perform steps comprising:” for proper antecedent basis. Dependent claims 16-20 inherit the objection.
Appropriate correction is required.
Claim Rejections - 35 USC 112(b)
The following is a quotation of 35 USC 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 USC 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(s) 4-14, and 18-20 is/are rejected under 35 USC 112(b) or 35 USC 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 USC 112, the applicant), regards as the invention.
Claims 4, 11, and 18 are indefinite for at least the following reasons:
In order for a formula appearing in a claim to be well-defined, every variable appearing therein must be clearly defined in the claim. However, the variables E_q and E_c are undefined in the claim. These undefined variables are interpreted in view of the specification as referring, respectively, to the identified features in the specification sheet and to the identified features in one of the plurality of descriptive sheets.
They recite v_e denotes the vector embedding for each feature/entity e, and w_e is the importance for each identified technical feature/entity, e. For proper antecedent basis and consistency of nomenclature, the examiner suggests “v_e denotes the vector embedding for each identified technical feature e, and w_e is the importance for each identified technical feature e”.
Dependent claims 5-7, 12-14, and 19-20 inherit these rejections.
Claims 7 and 14 are indefinite because they recite a loss function L(t) = max(0, (1 - s_{i,p}) - (1 - s_{i,q} + α) but this formula is insufficiently defined since none of the variables t, i, p, q, and α are defined in the claim.
Claim 8 is rendered indefinite by amendments made to the claim (due in part to the fact that the language used in claim 8 is not parallel to the language used in independent claims 1 and 15). For example, claim 8 recites the calculating including identifying noun phrases and generating vector embeddings of the noun phrases but the identification of noun phrases and generation of vector embeddings as described in the originally filed disclosure (and in independent claims 1 and 15) appear to be part of the functionality of the feature classifier, not of the importance calculator as presently recited in claim 8. Similarly, claim 8 recites the trained feature classifier identifies the technical features in the first set of technical features but the trained feature classifier as described by the originally filed disclosure (and in independent claims 1 and 15) appears to identify technical features in text (e.g., the specification sheet), not in a set of technical features as presently recited in claim 8. MPEP 2173.03(b) indicates that a claim is “indefinite when a conflict or inconsistency between the claimed subject matter and the specification disclosure renders the scope of the claim uncertain” and the aforementioned points are precisely such points of conflict or inconsistency. Moreover, claim 8 was also amended to recite text data including a specification sheet including a plurality of technical features including a first set of technical features, and a plurality of descriptive sheets each including a plurality of technical features including a second set of technical features but this clause is rendered ungrammatical/unclear due to the recitation of both “pluralit[ies]” of technical features and “set[s]” of technical features. Dependent claims 9-14 inherit the rejections. To avoid these issues (and further issues such as these), the examiner suggests amending claim 8 to be a system claim which otherwise mirrors the functional language used in independent claims 1 and 15. For the purpose of compact prosecution, the indefinite elements of the claim are interpreted as encompassing at least the interpretation suggested by the corresponding elements of independent claims 1 and 15.
Claim Rejections - 35 USC 101
35 USC 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.
Claim(s) 1-20 is/are rejected under 35 USC 101 because the claimed invention(s) is/are directed to abstract ideas without significantly more.
Claim 1
Step 1. The claim and its dependents 2-7 fall under the statutory category of methods. An analysis of step 2 for each of these claims follows.
Step 2A Prong 1. The claim recites the following abstract ideas:
identify technical features by identifying noun phrases and generating vector embeddings of the noun phrases; (This recites a mental process that can be performed in the human mind or by a human using pen and paper. A human mind can identify features and/or noun phrases and generate vectors representing noun phrases. See MPEP 2106.04(a)(2)(III).)
calculate an importance value for each identified technical feature represented by the vector embeddings; (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
segmenting text from the specification sheet and segment text from the plurality of descriptive sheets into sentences; (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
identifying a first set of technical features in the specification sheet and a second set of technical features in the plurality of descriptive sheets (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
calculating an importance for each identified technical feature in the first set of technical features represented by the vector embeddings and each identified technical feature in the second set of technical features represented by the vector embeddings (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
and calculating a matching score between the specification sheet and each of the plurality of descriptive sheets based on the importance of each identified technical feature in the first set of technical features and each identified technical feature in the second set of technical features. (This recites a mathematical concept (cf. [specification, 0059]) and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
A method of detail matching, comprising: training a feature classifier to [identify…] training a neural network model for a trained importance calculator to [calculate…] (These are generically recited training steps. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
receiving a specification sheet including a plurality of technical features; receiving a plurality of descriptive sheets each including a plurality of technical features; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
[identifying…] using the trained feature classifier; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[calculating…] using the trained feature importance calculator; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
A method of detail matching, comprising: training a feature classifier to [identify…] training a neural network model for a trained importance calculator to [calculate…] (These are generically recited training steps. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
receiving a specification sheet including a plurality of technical features; receiving a plurality of descriptive sheets each including a plurality of technical features; (This insignificant extra-solution activity is well-understood, routine, conventional as it is mere data transfer. See MPEP 2106.05(d)(II), “Receiving or transmitting data over a network” and/or “Storing and retrieving information in memory”.)
[identifying…] using the trained feature classifier; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[calculating…] using the trained feature importance calculator; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claim 2
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
[The method of claim 1, wherein] the trained importance calculator is trained using triplets of the specification sheet and the plurality of descriptive sheets. (This merely indicates the training data that is used for a generically recited training step. In other words, this recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
[The method of claim 1, wherein] the trained importance calculator is trained using triplets of the specification sheet and the plurality of descriptive sheets. (This merely indicates the training data that is used for a generically recited training step. In other words, this recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Claim 3
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The method of claim 2, further comprising] generating the vector embeddings for each identified technical feature (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
using a trained Bidirectional Encoder Representations from Transformers (BERT) model. (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
using a trained Bidirectional Encoder Representations from Transformers (BERT) model. (BERT is well-understood, routine, conventional. For example, Kui XUE et al. (Fine-tuning BERT for Joint Entity and Relation Extraction in Chinese Medical Text, published 2019; hereafter “Xue”) discusses the “well-known BERT language model” [Xue, abstract]. Other references which support the conclusion that BERT is well-understood, routine, conventional can be found in the conclusion of a previous Office action.)
Claim 4
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The method of claim 3, wherein] the matching scores, s_{q,c}, are calculated using s_{q,c} = sum_{e_q in E_q} w_{e_q} max_{e_c in E_c} frac{v_{e_q} v_{e_c}}{‖v_{e_q}‖ ‖v_{e_c}‖}, wherein v_e denotes the vector embedding for each feature/entity e, and w_e is the importance for each identified technical feature/entity, e. (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
Claim 5
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
[The method of claim 4, wherein] training the feature classifier utilizes a positive feature set, P, and an unlabeled feature set, U, where E = P ∪ U, where E is a whole feature set. (This merely indicates the training data that is used for a generically recited training step. In other words, this recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
[The method of claim 4, wherein] training the feature classifier utilizes a positive feature set, P, and an unlabeled feature set, U, where E = P ∪ U, where E is a whole feature set. (This merely indicates the training data that is used for a generically recited training step. In other words, this recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Claim 6
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
[The method of claim 4, wherein] matched documents are utilized to train an entity importance model H(v_e). (This amounts to an indication that the training data includes “matched documents”. In other words, this recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
[The method of claim 4, wherein] matched documents are utilized to train an entity importance model H(v_e). (This amounts to an indication that the training data includes “matched documents”. In other words, this recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Claim 7
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
based on a loss function, L(t) = max(0, (1 - s_{i,p}) - (1 - s_{i,q} + α). (This recites a mathematical concept. See MPEP 2106.04(a)(2)(I).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
[The method of claim 6, wherein] parameters of entity importance model H(v_e) are tuned (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
[The method of claim 6, wherein] parameters of entity importance model H(v_e) are tuned (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claim 8
Step 1. The claim and its dependents 9-14 fall under the statutory category of machines.
Step 2A Prong 1. The claim recites the following abstract ideas:
identify technical features; (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
calculating an importance value for each identified technical feature the calculating includes identifying noun phrases and generating vector embeddings of the noun phrases; (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
segments text from the specification sheet and text from the plurality of descriptive sheets into sentences (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
identifies the technical features in the first set of technical features and the technical features in the second set of technical features; (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
calculate an importance for each identified technical feature in the first set of technical features represented by the vector embeddings and reach identified technical feature in the second set of technical features represented by the vector embeddings (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
calculate a matching score between the specification sheet and each of the plurality of descriptive sheets based on the calculated importance of each identified technical feature in the first set of technical features and each identified technical feature in the second set of technical features, (This recites a mathematical concept (cf. [specification, 0059]) and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
A computer system for detail matching, comprising: one or more processors; a computer memory in electronic communication with the one or more processors, and a display screen in electronic communication with the computer memory and the one or more processors; wherein the computer memory includes: (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
a feature classifier trained to [identify technical features;] a neural network model configured as a trained importance calculator for [calculating an importance value for each identified technical feature;] (These are generically recited training steps. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
text data including a specification sheet including a plurality of technical features including a first set of technical features, and a plurality of descriptive sheets each including a plurality of technical features including a second set of technical features, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
wherein a text segmentor [segments text…] and the trained feature classifier [identifies the technical features…] (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
a feature importance calculator [to calculate an importance…] using the trained feature importance calculator; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and a feature matching system to [calculate a matching score…] (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
wherein a closest matching product is presented to a user on the display screen. (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
A computer system for detail matching, comprising: one or more processors; a computer memory in electronic communication with the one or more processors, and a display screen in electronic communication with the computer memory and the one or more processors; wherein the computer memory includes: (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
a feature classifier trained to [identify technical features;] a neural network model configured as a trained importance calculator for [calculating an importance value for each identified technical feature;] (These are generically recited training steps. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
text data including a specification sheet including a plurality of technical features including a first set of technical features, and a plurality of descriptive sheets each including a plurality of technical features including a second set of technical features, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
wherein a text segmentor [segments text…] and the trained feature classifier [identifies the technical features…] (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
a feature importance calculator [to calculate an importance…] using the trained feature importance calculator; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and a feature matching system to [calculate a matching score…] (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
wherein a closest matching product is presented to a user on the display screen. (The insignificant extra-solution activity is well-understood, routine, conventional as it is merely presenting output. See MPEP 2106.05(d)(II), “Presenting offers”.)
Claims 9-14 inherit limitations from claim 8 and recite additional limitations which are substantially similar to those recited by claims 2-7, respectively, so they are rejected by the same rationale.
Claim 15
Step 1. The claim and its dependents 16-20 fall under the statutory category of machines.
Step 2A Prong 1. The claim recites the following abstract ideas:
identify technical features by identifying noun phrases and generating vector embeddings of the noun phrases; (This recites a mental process that can be performed in the human mind or by a human using pen and paper. A human mind can identify features and/or noun phrases and generate vectors representing noun phrases. See MPEP 2106.04(a)(2)(III).)
calculate an importance value for each identified technical feature represented by the vector embeddings; (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
segmenting text from the specification sheet and segment text from the plurality of descriptive sheets into sentences; (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
identifying a first set of technical features in the specification sheet and a second set of technical features in the plurality of descriptive sheets (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
calculating an importance for each identified technical feature in the first set of technical features represented by the vector embeddings and each identified technical feature in the second set of technical features represented by the vector embeddings (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
and calculating a matching score between the specification sheet and each of the plurality of descriptive sheets based on the importance of each identified technical feature in the first set of technical features and each identified technical feature in the second set of technical features. (This recites a mathematical concept (cf. [specification, 0059]) and a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(I, III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
A non-transitory computer readable storage medium comprising a computer readable program for detail matching, wherein the computer readable program when executed on a computer causes the computer to perform the steps of: (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
training a feature classifier to [identify…] training a neural network model for a trained importance calculator to [calculate…] (These are generically recited training steps. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
receiving a specification sheet including a plurality of technical features; receiving a plurality of descriptive sheets each including a plurality of technical features; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
[identifying…] using the trained feature classifier; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[calculating…] using the trained feature importance calculator; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
A non-transitory computer readable storage medium comprising a computer readable program for detail matching, wherein the computer readable program when executed on a computer causes the computer to perform the steps of: (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
training a feature classifier to [identify…] training a neural network model for a trained importance calculator to [calculate…] (These are generically recited training steps. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
receiving a specification sheet including a plurality of technical features; receiving a plurality of descriptive sheets each including a plurality of technical features; (This insignificant extra-solution activity is well-understood, routine, conventional as it is mere data transfer. See MPEP 2106.05(d)(II), “Receiving or transmitting data over a network” and/or “Storing and retrieving information in memory”.)
[identifying…] using the trained feature classifier; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[calculating…] using the trained feature importance calculator; (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claims 16-20 inherit limitations from claim 8 and recite additional limitations which are substantially similar to those recited by claims 2-6, respectively, so they are rejected by the same rationale.
Claim Rejections - 35 USC 103
The following is a quotation of 35 USC 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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 USC 102(b)(2)(C) for any potential 35 USC 102(a)(2) prior art against the later invention.
Claim(s) 1-3, 8-10, and 15-17 is/are rejected under 35 USC 103 as being unpatentable over Juan HUETLE-FIGUEROA (Measuring semantic similarity of documents with weighted cosine and fuzzy logic, published 2020; hereafter, “Huetle”) in view of Si SUN et al. (Joint Keyphrase Chunking and Salience Ranking with BERT, published 2020; hereafter, “Sun”) and Qun GUO et al. (US20200226367A1, published 2020-07-16; hereafter, “Guo”).
Claim 1
Huetle discloses:
A method of detail matching, comprising: ([Huetle, abstract and section 5]: Huetle discloses a method of “matching of documents” [Huetle, abstract]. More specifically, it describes matching “CVs (resumes) and job descriptions” [Huetle, abstract], and discusses this matching from the perspective of either an employer or a jobseeker [Huetle, section 5 last paragraph].)
a feature classifier to identify technical features by identifying noun phrases ([Huetle, figures 1-2 and pages 2266-2267]: The method of Huetle includes “component (iv)” which performs “keyword extraction” [Huetle, pages 2266-2267 paragraph beginning “Fig. 2”]. In other words, component (iv) is the “feature classifier” of the claim, and the keywords are the “technical features” of the claim. The keywords depicted using highlights in [Huetle, figure 1] are “noun phrases” as required by the claim. See also: “keywords extraction was used to create an individual list of keywords for each document; All the documents have their own keyword list” [Huetle, section 6.2 second paragraph].)
receiving a specification sheet including a plurality of technical features; receiving a plurality of descriptive sheets each including a plurality of technical features; ([Huetle, abstract, figure 1, section 1]: As noted above, Huetle discusses job descriptions and CVs [Huetle, abstract and/or figure 1 elements 3-4]. Moreover, there is a plurality of each (6917 job descriptions and 80 CVs) [Huetle, section 3 first two paragraphs]. From the perspective of an employer, a given job description maps to the “specification sheet” of the claim and the 80 CVs to the “plurality of descriptive sheets” of the claim. From the perspective of a jobseeker, a given CV maps to the “specification sheet” of the claim and the 6917 job descriptions to the “plurality of descriptive sheets” of the claim. While either of these perspectives falls under the broadest reasonable interpretation of the claim, the remainder of the mappings herein use the perspective of the jobseeker for concreteness.)
identifying a first set of technical features in the specification sheet and a second set of technical features in the plurality of descriptive sheets using the trained feature classifier; ([Huetle, section 6.2]: Huetle indicates that “keywords extraction was used to create an individual list of keywords for each document; All the documents have their own keyword list” [Huetle, section 6.2 second paragraph]. In other words, the keywords in a given jobseeker’s CV maps to the “first set of technical features” of the claim, and the keywords in job descriptions map to the “second set of technical features” of the claim.)
calculating an importance for each identified technical feature in the first set of technical features [represented by the vector embeddings] and each identified technical feature in the second set of technical features ([Huetle, section 5]: Huetle discloses the use of tf-idf (term frequency-inverse document frequency) to assign weights to keywords in each document [Huetle, section 5]. As tf-idf is a measure of importance of a word to a particular document in a collection of documents, calculating tf-idf maps to “calculating an importance” as recited by the claim. The examiner notes that the combination with Sun below suggests the use of different measures of importance.)
and calculating a matching score between the specification sheet and each of the plurality of descriptive sheets ([Huetle, figure 1 and sections 5-6]: For a given CV [Huetle, figure 1 element 3], Huetle calculates similarity scores with each of the job descriptions [Huetle, figure 1 element 1]. In other words, each similarity score maps to the “matching score” of the claim. For more information about how similarity scores are calculated, see [Huetle, sections 5-6]; the examiner notes that the mapping for dependent claim 4 below gives an alternative way of mapping the “matching score” that fits the further limitations recited therein.) based on the importance of each identified technical feature in the first set of technical features and each identified technical feature in the second set of technical features. ([Huetle, sections 5-6]: The similarity scores of Huetle are based on tf-idf [Huetle, section 5 equation (3) and/or section 6.1], and are therefore “based on the importance of each identified technical feature” as mapped above.)
While Huetle discloses a keyword extractor as well as calculating importance values, it does not distinctly disclose training a keyword extractor or a neural network for calculating importance values, and it does not distinctly discuss vector embeddings. In other words, Huetle might not distinctly disclose:
training [a feature classifier]
and generating vector embeddings of the noun phrases;
training a neural network model for a trained importance calculator to calculate an importance value for each identified technical feature represented by the vector embeddings; … [calculating an importance for each identified technical feature in the first set of technical features] represented by the vector embeddings [and each identified technical feature in the second set of technical features] represented by the vector embeddings using the trained feature importance calculator;
segmenting text from the specification sheet and segmenting text from the plurality of descriptive sheets into sentences;
Sun is in the field of natural language processing. Moreover, Huetle in view of Sun discloses:
training [a feature classifier] ([Sun, abstract and section 2]: Sun discloses “a multitask BERT-based model for keyphrase extraction” [Sun, abstract] and a method of training this model [Sun, abstract and/or section 2 paragraph beginning “Joint Training”]. In the combination, Sun’s model is used as the keyword extractor described above.)
and generating vector embeddings of the noun phrases; ([Sun, section 2]: Sun discloses “us[ing] BERT to encode D to a sequence of vectors H = {h_1, …, h_i, …, h_n}” [Sun, section 2 paragraph beginning “Token Embedding”]. In other words, the vectors h_i corresponding to keywords map to the “vector embeddings” of the claim.)
training a neural network model for a trained importance calculator to calculate an importance value for each identified technical feature represented by the vector embeddings; … [calculating an importance for each identified technical feature in the first set of technical features] represented by the vector embeddings [and each identified technical feature in the second set of technical features] represented by the vector embeddings using the trained feature importance calculator; ([Sun, section 2]: The keyphrase extraction model disclosed by Sun is a neural network containing a “ranking network” [Sun, section 2 paragraph beginning “This is achieved”], where the ranking network generates “salience scores” [Sun, section 2 paragraph beginning “Ranking Network”]. The salience scores are based on the vector embeddings h_i [Sun, section 2 equations (1-4)]. In other words, the ranking network of Sun (or, alternatively, the entire network which contains the ranking network) maps to the “neural network model” and the “trained importance calculator” of the claim. In the combination, a salience score as in Sun is used in place of tf-idf as described in Huetle and maps to the “importance” and “importance value” of the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle with the keyphrase extractor of Sun because the latter “has advantages in predicting long keyphrases and extracting phrases that are not entities but also meaningful” [Sun, abstract], so the combination would be more effective overall.
Huetle in view of Sun might not distinctly disclose:
segmenting text from the specification sheet and segmenting text from the plurality of descriptive sheets into sentences;
Guo is in the field of machine learning. It discloses a method of keyword extraction [Guo, abstract]. Moreover, Huetle in view of Sun and Guo discloses:
segmenting text from the specification sheet and segmenting text from the plurality of descriptive sheets into sentences; ([Guo, 0134]: The keyword extraction method of Guo includes a step where text is “segmented into sentences” [Guo, 0134]. In the combination, the text from the CVs and job descriptions of Huetle is segmented into sentences as described in Guo in order to perform keyword extraction using the extractor disclosed by Sun.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle in view of Sun with aspects of the keyword extraction method of Guo because they overcome a number of “disadvantages” of “[t]ypical keyword extraction methods” [Guo, 0097] (including low accuracy and coverage rate [Guo, 0098], high calculation complexity [Guo, 0099], etc.), so the combination as described above would ensure the method is more effective and efficient overall.
Claim 2
Huetle in view of Sun and Guo discloses the elements of the parent claim(s). It also discloses:
[The method of claim 1, wherein] the trained importance calculator is trained using triplets of the specification sheet and the plurality of descriptive sheets. ([Sun, section 2]: Sun discloses the use of a “positive set P_+” and a “negative set P_−” associated to each document D, where P_+ is the set of keywords in D and P_- is the set of all phrases in D which are not keywords [Sun, section 2 paragraph beginning “To obtain ranking labels”]. In other words, a triple (D, P_+, P_-) falls under the broadest reasonable interpretation of one of the “triplets” of the claim.)
The same motivation to combine applies.
Claim 3
Huetle in view of Sun and Guo discloses the elements of the parent claim(s). It also discloses:
[The method of claim 2, further comprising] generating vector embeddings for each identified technical feature using a trained Bidirectional Encoder Representations from Transformers (BERT) model. ([Sun, section 2]: Sun discloses “us[ing] BERT to encode D to a sequence of vectors H = {h_1, …, h_i, …, h_n}” [Sun, section 2 paragraph beginning “Token Embedding”]. In other words, the vectors h_i corresponding to keywords map to the “vector embeddings” of the claim.)
The same motivation to combine applies.
Claim 8
Huetle discloses:
A computer system for detail matching, comprising: one or more processors; a computer memory in electronic communication with the one or more processors, and a display screen in electronic communication with the computer memory and the one or more processors; wherein the computer memory includes: ([Huetle, abstract, figure 1, and sections 3 and 5]: Huetle discloses a method of “matching of documents” [Huetle, abstract]. More specifically, it describes matching “CVs (resumes) and job descriptions” [Huetle, abstract], and discusses this matching from the perspective of either an employer or a jobseeker [Huetle, section 5 last paragraph]. The method is implemented as a “web application” [Huetle, section 3 paragraph beginning “The new gold standard”], a screenshot of which is depicted in [Huetle, figure 1]. Any computer on which this web application is executed is the “computer system for detail matching” of the claim, its processor(s), memory, and display mapping, respectively, to the “one or more processors”, the “computer memory”, and the “display screen” of the claim.)
a trained feature classifier [trained] to identify technical features; … identifying noun phrases ([Huetle, figures 1-2 and pages 2266-2267]: The method of Huetle includes “component (iv)” which performs “keyword extraction” [Huetle, pages 2266-2267 paragraph beginning “Fig. 2”]. In other words, component (iv) is the “feature classifier” of the claim, and the keywords are the “technical features” of the claim. The keywords depicted using highlights in [Huetle, figure 1] are “noun phrases” as required by the claim. See also: “keywords extraction was used to create an individual list of keywords for each document; All the documents have their own keyword list” [Huetle, section 6.2 second paragraph].)
text data including a specification sheet including a plurality of technical features including a first set of technical features, and a plurality of descriptive sheets each including a plurality of technical features including a second set of technical features, ([Huetle, abstract, figure 1, section 1]: As noted above, Huetle discusses job descriptions and CVs [Huetle, abstract and/or figure 1 elements 3-4]. Moreover, there is a plurality of each (6917 job descriptions and 80 CVs) [Huetle, section 3 first two paragraphs]. From the perspective of an employer, a given job description maps to the “specification sheet” of the claim and the 80 CVs to the “plurality of descriptive sheets” of the claim. From the perspective of a jobseeker, a given CV maps to the “specification sheet” of the claim and the 6917 job descriptions to the “plurality of descriptive sheets” of the claim. While either of these perspectives falls under the broadest reasonable interpretation of the claim, the remainder of the mappings herein use the perspective of the jobseeker for concreteness. Any dataset containing the “specification sheet” and the “plurality of descriptive sheet” (for example, the “raw documents dataset” [Huetle, element 1 figure i)]) maps to the “text data” of the claim.)
and the trained feature classifier identifies the technical features in the first set of technical features and the technical features in the second set of technical features; ([Huetle, section 6.2]: Huetle indicates that “keywords extraction was used to create an individual list of keywords for each document; All the documents have their own keyword list” [Huetle, section 6.2 second paragraph]. In other words, the keywords in a given jobseeker’s CV maps to the “first set of technical features” of the claim, and the keywords in job descriptions map to the “second set of technical features” of the claim.)
calculate an importance for each identified technical feature in the first set of technical features [represented by the vector embeddings] and each identified technical feature in the second set of technical features ([Huetle, section 5]: Huetle discloses the use of tf-idf (term frequency-inverse document frequency) to assign weights to keywords in each document [Huetle, section 5]. As tf-idf is a measure of importance of a word to a particular document in a collection of documents, calculating tf-idf maps to “calculating an importance” as recited by the claim. The examiner notes that the combination with Sun below suggests the use of different measures of importance.)
and a feature matching system to calculate a matching score between the specification and each of the plurality of descriptive sheets ([Huetle, figure 1 and sections 5-6]: For a given CV [Huetle, figure 1 element 3], Huetle calculates similarity scores with each of the job descriptions [Huetle, figure 1 element 1]. In other words, each similarity score maps to the “matching score” of the claim. For more information about how similarity scores are calculated, see [Huetle, sections 5-6]; the examiner notes that the mapping for dependent claim 4 below gives an alternative way of mapping the “matching score” that fits the further limitations recited therein.) based on the calculated importance of each identified technical feature in the first set of technical features and each identified technical feature in the second set of technical features, ([Huetle, sections 5-6]: The similarity scores of Huetle are based on tf-idf [Huetle, section 5 equation (3) and/or section 6.1], and are therefore “based on the importance of each identified technical feature” as mapped above.)
wherein a closest matching product is presented to a user on the display screen. ([Huetle, figure 1]: For a given CV, the web application of Huetle displays a list of the most similar (i.e., “closest matching”) job descriptions [Huetle, figure 1 elements 1-2].)
While Huetle discloses a keyword extractor as well as calculating importance values, it does not distinctly disclose training a keyword extractor or a neural network for calculating importance values, and it does not distinctly discuss vector embeddings. In other words, Huetle might not distinctly disclose:
[a trained feature classifier] trained [to identify technical features]
and generating vector embeddings of the noun phrases;
a neural network model configured as a trained importance calculator for calculating an importance value for each identified technical feature the calculating includes … a feature importance calculator [to calculate an importance for each identified technical feature in the first set of technical features] represented by the vector embeddings [and each identified technical feature in the second set of technical features] represented by the vector embeddings using the trained feature importance calculator;
wherein a text segmentor segments text from the specification sheet and text from the plurality of descriptive sheets into sentences
Sun is in the field of natural language processing. Moreover, Huetle in view of Sun discloses:
[a trained feature classifier] trained [to identify technical features] ([Sun, abstract and section 2]: Sun discloses “a multitask BERT-based model for keyphrase extraction” [Sun, abstract] and a method of training this model [Sun, abstract and/or section 2 paragraph beginning “Joint Training”]. In the combination, Sun’s model is used as the keyword extractor described above.)
and generating vector embeddings of the noun phrases; ([Sun, section 2]: Sun discloses “us[ing] BERT to encode D to a sequence of vectors H = {h_1, …, h_i, …, h_n}” [Sun, section 2 paragraph beginning “Token Embedding”]. In other words, the vectors h_i corresponding to keywords map to the “vector embeddings” of the claim.)
a neural network model configured as a trained importance calculator for calculating an importance value for each identified technical feature the calculating includes … a feature importance calculator [to calculate an importance for each identified technical feature in the first set of technical features] represented by the vector embeddings [and each identified technical feature in the second set of technical features] represented by the vector embeddings using the trained feature importance calculator; ([Sun, section 2]: The keyphrase extraction model disclosed by Sun is a neural network containing a “ranking network” [Sun, section 2 paragraph beginning “This is achieved”], where the ranking network generates “salience scores” [Sun, section 2 paragraph beginning “Ranking Network”]. The salience scores are based on the vector embeddings h_i [Sun, section 2 equations (1-4)]. In other words, the ranking network of Sun (or, alternatively, the entire network which contains the ranking network) maps to the “neural network model” and the “trained importance calculator” of the claim. In the combination, a salience score as in Sun is used in place of tf-idf as described in Huetle and maps to the “importance” and “importance value” of the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle with the keyphrase extractor of Sun because the latter “has advantages in predicting long keyphrases and extracting phrases that are not entities but also meaningful” [Sun, abstract], so the combination would be more effective overall.
Huetle in view of Sun might not distinctly disclose:
wherein a text segmentor segments text from the specification sheet and text from the plurality of descriptive sheets into sentences
Guo is in the field of machine learning. It discloses a method of keyword extraction [Guo, abstract]. Moreover, Huetle in view of Sun and Guo discloses:
wherein a text segmentor segments text from the specification sheet and text from the plurality of descriptive sheets into sentences ([Guo, 0134]: The keyword extraction method of Guo includes a step where text is “segmented into sentences” [Guo, 0134]. In the combination, the text from the CVs and job descriptions of Huetle is segmented into sentences as described in Guo in order to perform keyword extraction using the extractor disclosed by Sun.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle in view of Sun with aspects of the keyword extraction method of Guo because they overcome a number of “disadvantages” of “[t]ypical keyword extraction methods” [Guo, 0097] such as low accuracy and coverage rate [Guo, 0098], high calculation complexity [Guo, 0099], etc., so the combination as described above would ensure the method is more effective and efficient overall.
Claims 9-10 inherit limitations from claim 8 and recite additional limitations which are substantially similar to those recited by claims 2-3, respectively, so they are rejected by the same rationale.
Claim 15
Huetle discloses:
A non-transitory computer readable storage medium comprising a computer readable program for detail matching, wherein the computer readable program when executed on a computer causes the computer to perform the steps of: ([Huetle, abstract, figure 1, and sections 3 and 5]: Huetle discloses a method of “matching of documents” [Huetle, abstract]. More specifically, it describes matching “CVs (resumes) and job descriptions” [Huetle, abstract], and discusses this matching from the perspective of either an employer or a jobseeker [Huetle, section 5 last paragraph]. The method is implemented as a “web application” [Huetle, section 3 paragraph beginning “The new gold standard”], a screenshot of which is depicted in [Huetle, figure 1]. The web application is the “computer readable program for detail matching” of the claim, and any computer or server on which the web application is stored provides the “non-transitory computer readable storage medium” of the claim.)
a feature classifier to identify technical features by identifying noun phrases ([Huetle, figures 1-2 and pages 2266-2267]: The method of Huetle includes “component (iv)” which performs “keyword extraction” [Huetle, pages 2266-2267 paragraph beginning “Fig. 2”]. In other words, component (iv) is the “feature classifier” of the claim, and the keywords are the “technical features” of the claim. The keywords depicted using highlights in [Huetle, figure 1] are “noun phrases” as required by the claim. See also: “keywords extraction was used to create an individual list of keywords for each document; All the documents have their own keyword list” [Huetle, section 6.2 second paragraph].)
receiving a specification sheet including a plurality of technical features; receiving a plurality of descriptive sheets each including a plurality of technical features; ([Huetle, abstract, figure 1, section 1]: As noted above, Huetle discusses job descriptions and CVs [Huetle, abstract and/or figure 1 elements 3-4]. Moreover, there is a plurality of each (6917 job descriptions and 80 CVs) [Huetle, section 3 first two paragraphs]. From the perspective of an employer, a given job description maps to the “specification sheet” of the claim and the 80 CVs to the “plurality of descriptive sheets” of the claim. From the perspective of a jobseeker, a given CV maps to the “specification sheet” of the claim and the 6917 job descriptions to the “plurality of descriptive sheets” of the claim. While either of these perspectives falls under the broadest reasonable interpretation of the claim, the remainder of the mappings herein use the perspective of the jobseeker for concreteness.)
identifying a first set of technical features in the specification sheet and a second set of technical features in the plurality of descriptive sheets using the trained feature classifier; ([Huetle, section 6.2]: Huetle indicates that “keywords extraction was used to create an individual list of keywords for each document; All the documents have their own keyword list” [Huetle, section 6.2 second paragraph]. In other words, the keywords in a given jobseeker’s CV maps to the “first set of technical features” of the claim, and the keywords in job descriptions map to the “second set of technical features” of the claim.)
calculating an importance for each identified technical feature in the first set of technical features [represented by the vector embeddings] and each identified technical feature in the second set of technical features ([Huetle, section 5]: Huetle discloses the use of tf-idf (term frequency-inverse document frequency) to assign weights to keywords in each document [Huetle, section 5]. As tf-idf is a measure of importance of a word to a particular document in a collection of documents, calculating tf-idf maps to “calculating an importance” as recited by the claim. The examiner notes that the combination with Sun below suggests the use of different measures of importance.)
and calculating a matching score between the specification sheet and each of the plurality of descriptive sheets ([Huetle, figure 1 and sections 5-6]: For a given CV [Huetle, figure 1 element 3], Huetle calculates similarity scores with each of the job descriptions [Huetle, figure 1 element 1]. In other words, each similarity score maps to the “matching score” of the claim. For more information about how similarity scores are calculated, see [Huetle, sections 5-6]; the examiner notes that the mapping for dependent claim 4 below gives an alternative way of mapping the “matching score” that fits the further limitations recited therein.) based on the importance of each identified technical feature in the first set of technical features and each identified technical feature in the second set of technical features. ([Huetle, sections 5-6]: The similarity scores of Huetle are based on tf-idf [Huetle, section 5 equation (3) and/or section 6.1], and are therefore “based on the importance of each identified technical feature” as mapped above.)
While Huetle discloses a keyword extractor as well as calculating importance values, it does not distinctly disclose training a keyword extractor or a neural network for calculating importance values, and it does not distinctly discuss vector embeddings. In other words, Huetle might not distinctly disclose:
training [a feature classifier]
and generating vector embeddings of the noun phrases;
training a neural network model for a trained importance calculator to calculate an importance value for each identified technical feature represented by the vector embeddings; … [calculating an importance for each identified technical feature in the first set of technical features] represented by the vector embeddings [and each identified technical feature in the second set of technical features] represented by the vector embeddings using the trained feature importance calculator;
segmenting text from the specification sheet and segmenting text from the plurality of descriptive sheets into sentences;
Sun is in the field of natural language processing. Moreover, Huetle in view of Sun discloses:
training [a feature classifier] ([Sun, abstract and section 2]: Sun discloses “a multitask BERT-based model for keyphrase extraction” [Sun, abstract] and a method of training this model [Sun, abstract and/or section 2 paragraph beginning “Joint Training”]. In the combination, Sun’s model is used as the keyword extractor described above.)
and generating vector embeddings of the noun phrases; ([Sun, section 2]: Sun discloses “us[ing] BERT to encode D to a sequence of vectors H = {h_1, …, h_i, …, h_n}” [Sun, section 2 paragraph beginning “Token Embedding”]. In other words, the vectors h_i corresponding to keywords map to the “vector embeddings” of the claim.)
training a neural network model for a trained importance calculator to calculate an importance value for each identified technical feature represented by the vector embeddings; … [calculating an importance for each identified technical feature in the first set of technical features] represented by the vector embeddings [and each identified technical feature in the second set of technical features] represented by the vector embeddings using the trained feature importance calculator; ([Sun, section 2]: The keyphrase extraction model disclosed by Sun is a neural network containing a “ranking network” [Sun, section 2 paragraph beginning “This is achieved”], where the ranking network generates “salience scores” [Sun, section 2 paragraph beginning “Ranking Network”]. The salience scores are based on the vector embeddings h_i [Sun, section 2 equations (1-4)]. In other words, the ranking network of Sun (or, alternatively, the entire network which contains the ranking network) maps to the “neural network model” and the “trained importance calculator” of the claim. In the combination, a salience score as in Sun is used in place of tf-idf as described in Huetle and maps to the “importance” and “importance value” of the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle with the keyphrase extractor of Sun because the latter “has advantages in predicting long keyphrases and extracting phrases that are not entities but also meaningful” [Sun, abstract], so the combination would be more effective overall.
Huetle in view of Sun might not distinctly disclose:
segmenting text from the specification sheet and segmenting text from the plurality of descriptive sheets into sentences;
Guo is in the field of machine learning. It discloses a method of keyword extraction [Guo, abstract]. Moreover, Huetle in view of Sun and Guo discloses:
segmenting text from the specification sheet and segmenting text from the plurality of descriptive sheets into sentences; ([Guo, 0134]: The keyword extraction method of Guo includes a step where text is “segmented into sentences” [Guo, 0134]. In the combination, the text from the CVs and job descriptions of Huetle is segmented into sentences as described in Guo in order to perform keyword extraction using the extractor disclosed by Sun.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle in view of Sun with aspects of the keyword extraction method of Guo because they overcome a number of “disadvantages” of “[t]ypical keyword extraction methods” [Guo, 0097] such as low accuracy and coverage rate [Guo, 0098], high calculation complexity [Guo, 0099], etc., so the combination as described above would ensure the method is more effective and efficient overall.
Claims 16-17 inherit limitations from claim 8 and recite additional limitations which are substantially similar to those recited by claims 2-3, respectively, so they are rejected by the same rationale.
Claim(s) 4, 6-7, 11, 13-14, 18 and 20 is/are rejected under 35 USC 103 as being unpatentable over Huetle in view of Sun and Guo, further in view of Tianyi ZHANG et al. (BERTScore: Evaluating Text Generation with BERT, published 2020; hereafter “Zhang”).
Claim 4
Huetle in view of Sun and Guo discloses the elements of the parent claim(s). It does not distinctly disclose:
[The method of claim 3, wherein] the matching scores, s_{q,c}, are calculated using s_{q,c} = sum_{e_q in E_q} w_{e_q} max_{e_c in E_c} frac{v_{e_q} · v_{e_c}}{‖v_{e_q}‖ ‖v_{e_c}‖}, wherein v_e denotes the vector embedding for each feature/entity e, and w_e is the importance for each identified technical feature/entity, e.
Zhang is in the field of natural language processing. Moreover, Zhang discloses:
[The method of claim 3, wherein] the matching scores, s_{q,c}, are calculated using s_{q,c} = sum_{e_q in E_q} w_{e_q} max_{e_c in E_c} frac{v_{e_q} · v_{e_c}}{‖v_{e_q}‖ ‖v_{e_c}‖}, wherein v_e denotes the vector embedding for each feature/entity e, and w_e is the importance for each identified technical feature/entity, e. ([Zhang, sections 2-3]: Zhang works in the setting of computing a matching score f(x, hat{x}) between two sets of tokens x = ⟨x_1, …, x_k⟩ and hat{x} = ⟨hat{x}_1, …, hat{x}_l⟩ [Zhang, section 2 first paragraph]. In the combination, ⟨x_1, …, x_k⟩ represents the set of keywords in “specification sheet” as mapped above, and ⟨hat{x}_1, …, hat{x}_l⟩ the set of keywords in one of the “plurality of descriptive sheets” as mapped above. Zhang describes several strategies for computing f(x, hat{x}), with the method described in [Zhang, section 3] mapping to s_{q, c} as described in the claim. To elaborate, ⟨x_1, …, x_k⟩ and ⟨hat{x}_1, …, hat{x}_l⟩ are first converted to vectors ⟨x_1, …, x_k⟩ and ⟨hat{x}_1, …, hat{x}_l⟩ using BERT [Zhang, section 3 paragraph beginning “We experiment”], where the vectors are “pre-normalized” in the sense that their norm is 1 [Zhang, section 3 paragraph beginning “Similarity Measure”]. In other words, the sets of vectors ⟨x_1, …, x_k⟩ and ⟨hat{x}_1, …, hat{x}_l⟩ map to E_q and E_c of the claim, respectively, which means that the x_i are the vectors e_q in E_q of the claim and the hat{x}_j are the vectors e_c in E_c of the claim. Since the vectors in Zhang are pre-normalized, the cosine similarity x_i^T hat{x}_j maps to the quantity frac{v_{e_q} · v_{e_c}}{‖v_{e_q}‖ ‖v_{e_c}‖} appearing in the formula [Zhang, section 3 paragraph beginning “Similarity Measure”]. Zhang discloses computing the maximum of these cosine similarities over tokens x_j in hat{x}, and then taking a weighted sum over x_i in x where each maximum is weighted by an “importance” [Zhang, section 3 paragraph beginning “Importance Weighting”]. Zhang measures importance using idf, but in the combination, the importance is taken to be the salience score of Sun as described under the parent claim. With these mappings, the formula appearing in the numerator of [Zhang, section 3 paragraph beginning “Importance Weighting” second displayed equation] is precisely the same as the formula for s_{q,c} appearing in the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle in view of Sun and Guo with the matching score of Zhang because it “correlates better with human judgments and provides stronger model selection performance than existing metrics” [Zhang, abstract], so the combination would be more effective overall.
Claim 6
Huetle in view of Sun, Guo, and Zhang discloses the elements of the parent claim(s). It also discloses:
[The method of claim 4, wherein] matched documents are utilized to train an entity importance model H(v_e). ([Sun, sections 2-3]: As noted under parent claims, the ranking network of Sun is trained using documents D that are associated/matched with a positive set P_+ of keyphrases that occur in D and a negative set P_- of phrases occurring in D that are not keyphrases [Sun, section 2]. The ranking network is the “trained importance calculator” (and/or the “entity importance model”) of the claim and the documents D that are used for training fall under the broadest reasonable interpretation of the “matched documents” of the claim. For more details about the documents used for training, see [Sun, section 3 paragraph beginning “Dataset”].)
The same motivation to combine applies.
Claim 7
Huetle in view of Sun, Guo, and Zhang discloses the elements of the parent claim(s). It also discloses:
[The method of claim 6, wherein] parameters of the entity importance model H(v_e) are tuned based on a loss function, L(t) = max(0, (1 - s_{i,p}) - (1 - s_{i,q}) + α). ([Sun, section 2]: Sun discloses the use of the ranking loss L_Rank computed as a sum of terms of the form max(0, 1 - f^*(p_+, D) + f^*(p_-, D)) where p_+ and p_- are drawn from P_+ and P_-, respectively [Sun, section 2 equation (9)], and f^* is the salience score [Sun, section 2 equation (5)]. In other words, letting f^*(p_+, D) correspond to s_{i,p} in the claim and f^*(p_-, D) to s_{i,q}, the summand max(0, 1 - f^*(p_+, D) + f^*(p_-, D)) in Sun maps to the L(t) of the claim with α = 1. The examiner notes that, just as the loss function L_Rank in Sun is a sum of the terms max(0, 1 - f^*(p_+, D) + f^*(p_-, D)) , so too is the applicant’s loss function a sum of the terms L(t) [specification, 0057].)
The same motivation to combine applies.
Claims 11 and 13-14 inherit limitations from claim 8 and recite additional limitations which are substantially similar to those recited by claim 4 and 6-7, respectively, so they are rejected by the same rationale.
Claims 18 and 20 inherit limitations from claim 15 and recite additional limitations which are substantially similar to those recited by claim 4 and 6, respectively, so they are rejected by the same rationale.
Claim(s) 5, 12, and 19 is/are rejected under 35 USC 103 as being unpatentable over Huetle in view of Sun, Guo, and Zhang, further in view of Lucas STERCKX et al. (Supervised Keyphrase Extraction as Positive Unlabeled Learning, published 2016; hereafter “Sterckx”).
Claim 5
Huetle in view of Sun, Guo, and Zhang discloses the elements of the parent claim(s). It also discloses:
[The method of claim 4, wherein] training the feature classifier utilizes a positive feature set, P, ([Sun, section 2]: As noted under the parent claim, Sun discloses the use of a “positive set P_+” which maps to the “positive feature set, P” of the claim.)
While Sun discloses the set P_- of phrases in D which are not labeled as keyphrases, these are treated in Sun as a negative feature set rather than an unlabeled feature set. While this is merely a matter of perspective and the claim itself does not include any positively recited steps that clearly distinguish the training step of the claimed invention from the training method of Sun, it may nonetheless be argued that Huetle in view of Sun, Guo, and Zhang does not distinctly disclose:
and an unlabeled feature set, U, where E = P ∪ U, where E is a whole feature set.
Sterckx is in the field of natural language processing and discloses a method of training keyphrase extractors [Sterckx, abstract]. Moreover, Huetle in view of Sun, Guo, Zhang, and Sterckx discloses:
and an unlabeled feature set, U, where E = P ∪ U, where E is a whole feature set. ([Sterckx, abstract and section 3]: Sterckx discloses a method of “Positive Unlabeled Learning” for keyphrase extraction [Sterckx, abstract], in which “an incomplete set of positive examples is available [during training] as well as a set of unlabeled examples, of which some are positive and others negative” [Sterckx, section 3 first paragraph]. The incomplete set of positive examples of Sterckx corresponds to the set P_+ from Sun and the set P of the claim as mapped above. In the combination, the negative set P_- of Sun is regarded instead as an unlabeled as described in Sterck and maps to the “unlabeled feature set, U” of the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the document matching method of Huetle in view of Sun, Guo, and Zhang with the use of positive unlabeled learning for keyphrase extraction as in Sterkcx because “unlabeled keyphrase candidates are not reliable as negative examples by default” [Sterckx, section 3 first paragraph] and treating them as unlabeled examples (instead of as negative examples) “leads to higher average F1 scores and better rankings of keyphrases” [Sterkcx, abstract].
Claims 12 and 19 inherit limitations from claim 8 and 15, respectively, and recite additional limitations which are substantially similar to those recited by claim 5, so they are rejected by the same rationale.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
The NLTK 2.0 Documentation (backdated to 2012-05-12; hereafter, “NLTK”) is a part of the API for the well-known Python natural language processing package NLTK. It includes a “Punkt Sentence Tokenizer” which “divides a text into a list of sentences” [NLTK, first paragraph under the section titled “Punkt Module”].
Evgeny MATUSOV et al. (Automatic Sentence Segmentation and Punctuation Prediction for Spoken Language Translation, published 2006; hereafter, “Matusov”) discloses, as its title suggests, an automated method of sentence segmentation. The input to this sentence segmentation method is the output of an automated speech recognition (ASR) system. It is indicated explicitly that the “raw output of an ASR system… is a long sequence of words” [Matusov, section 4 paragraph beginning “Figure 1”]; the ASR system “neither perform[s] a proper segmentation of the output into sentences or sentence-like units (SUs), nor predict[s] punctuation marks. Usually, only acoustic segmentation into utterances is performed” [Matusov, section 1 first paragraph]. In other words, the sentence segmentation methods disclosed in Matusov call into question the applicant’s unsupported assertion that a “person of ordinary skill in the art would recognize that text without capitalization and punctuation cannot be segmented into sentences” [remarks, page 13]. Matusov performs sentence segmentation on precisely such texts.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Shishir AGRAWAL whose telephone number is +1 703-756-1183. The examiner can normally be reached Monday through Friday, 08:00-16:00 Pacific Time.
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, Alexey SHMATOV can be reached on +1 571-270-3428. The fax phone number for the organization where this application or proceeding is assigned is +1 571-273-8300.
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 +1 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call +1 800-786-9199 (IN USA OR CANADA) or +1 571-272-1000.
/S.A./Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights