Context-sensitive feature score generation

Patent metadata

Field	Value
Publication number	US-11263209-B2
Application number	US-201916395189-A
Country	US
Kind code	B2
Filing date	Apr 25, 2019
Priority date	Apr 25, 2019
Publication date	Mar 1, 2022
Grant date	Mar 1, 2022

Abstract

Document information may define words, key groups of words, and sets of context words within a document. Word feature scores for words within the document may be generated. Key group feature scores for individual key groups of words may be generated based on aggregation of word feature scores the words within the individual key groups of words and word feature scores for words within corresponding sets of context words. A document feature score for the document may be generated based on aggregation of word feature scores for words within the document. The key group feature scores and the document feature score may enable context-sensitive searching of words/word vectors in the document.

First claim

What is claimed is:

1. A system for generating context-sensitive feature scores for documents, the system comprising:
one or more physical processors configured by machine-readable instructions to:
obtain document information, the document information defining words within a document, key groups of words within the document, and sets of context words corresponding to individual ones of the key groups of words, wherein the sets of context words for the individual ones of the key groups of words are determined based on a hierarchical structure of the document and relative upper level relationships and relative lower level relationships of other words within the hierarchical structure of the document with the key groups of words;
generate word feature scores for words within individual key groups of words, for words within individual sets of context words, and for the words within the document;
generate key group feature scores for the individual key groups of words based on iterative calculation of cosine distances between the word feature scores for the words within the individual key groups of words and the word feature scores for the words within the corresponding sets of context words, the iterative calculation of the cosine distances between the word feature scores for the words within the individual key groups of words and the word feature scores for the words within the corresponding sets of context words resulting in aggregation of the cosine distances for the key group feature scores, wherein the aggregation of the cosine distances for a given key group feature score for a given key group of words represents concept of the words within the given key group of words and the words within a corresponding set of context words;
generate a document feature score for the document based on iterative calculation of cosine distances between the word feature scores for the words within the document; and
store the key group feature scores and the document feature score within one or more electronic storages, wherein storage of the key group feature scores and the document feature score enables context-sensitive searching of words that includes searching based on feature scores that take into account comprehensive context of words.

2. The system of claim 1 , wherein responsive to the given key group of words and the corresponding set of context words including a first word, a second word, and a third word, the aggregation of the cosine distances for the given key group feature score for the given key group of words includes:
calculation of an aggregated feature vector for the first word and the second word based on cosine distance calculation between a first feature vector of the first word and a second feature vector of the second word; and
calculation of an aggregated feature vector for the first word, the second word, and the third word based on cosine distance calculation between the aggregated feature vector for the first word and the second word and a third feature vector of the third word, wherein the aggregated feature vector for the first word, the second word, and the third word represents concept of combination of the first word, the second word, and the third word.

3. The system of claim 1 , wherein the document includes requirements, and individual key groups of words within the document correspond to individual requirements.

4. The system of claim 1 , wherein the hierarchical structure of the document includes, in descending level order, a document title, a section title, a section body, a bullet, and a sub-bullet.

5. The system of claim 4 , wherein the given key group of words includes words within the bullet, and the corresponding set of context words for the bullet is determined based on the relative upper level relationships of the other words within the hierarchical structure of the document with the bullet such that the corresponding set of context words for the bullet includes words within the section body, the section title, and the document title.

6. The system of claim 5 , wherein the corresponding set of context words for the bullet is determined based on the relative lower level relationships of the other words within the hierarchical structure of the document with the bullet such that the corresponding set of context words for the bullet does not include words within the sub-bullet.

7. The system of claim 6 , wherein the hierarchical structure of the document further includes a footnote, and the corresponding set of context words for the bullet is determined based on the relative lower level relationships of the other words within the hierarchical structure of the document with the bullet such that the corresponding set of context words for the bullet further includes words within the footnote.

8. The system of claim 1 , wherein the document is associated with operating system metadata, and the sets of context words corresponding to the individual ones of the key groups of words include words within at least some of the operating system metadata.

9. The system of claim 1 , wherein the word feature scores are generated based on processing of the document information through a context-sensitive document-to-vector model, the context-sensitive document-to-vector model including an attention distribution, a partial summary, and a vocabulary distribution.

10. The system of claim 9 , wherein:
the attention distribution facilitates generation of context-aware vector representation of words, generates a representation of context of the words, and assigns probabilistic weights to the words based on usage of the words;
the partial summary facilitates validation of the attention distribution and provides a check on the probabilistic weights assigned to the words; and
the vocabulary distribution facilitates combination of multiple words into a phrase by determining whether a sequence of words form a single term or separate terms.

11. A method for generating context-sensitive feature scores for documents, the method performed by a computing system including one or more processors, the method comprising:
obtaining, by the computing system, document information, the document information defining words within a document, key groups of words within the document, and sets of context words corresponding to individual ones of the key groups of words, wherein the sets of context words for the individual ones of the key groups of words are determined based on a hierarchical structure of the document and relative upper level relationships and relative lower level relationships of other words within the hierarchical structure of the document with the key groups of words;
generating, by the computing system, word feature scores for words within individual key groups of words, for words within individual sets of context words, and for the words within the document;
generating, by the computing system, key group feature scores for the individual key groups of words based on iterative calculation of cosine distances between the word feature scores for the words within the individual key groups of words and the word feature scores for the words within the corresponding sets of context words, the iterative calculation of the cosine distances between the word feature scores for the words within the individual key groups of words and the word feature scores for the words within the corresponding sets of context words resulting in aggregation of the cosine distances for the key group feature scores, wherein the aggregation of the cosine distances for a given key group feature score for a given key group of words represents concept of the words within the given key group of words and the words within a correspondi

Assignees

• Chevron Usa Inc

Inventors

• Bowden Jr Larry A
• Montero Esteban

Classifications

• G06F16/3347 Primary

  using vector based model · CPC title

• G06F16/24573

  using data annotations, e.g. user-defined metadata · CPC title

• G06F16/24575

  using context · CPC title

• G06F16/81

  Indexing, e.g. XML tags; Data structures therefor; Storage structures · CPC title

• G06F16/3346

  using probabilistic model · CPC title