WO2019140863A1 - Method of calculating relevancy, apparatus for calculating relevancy, data query apparatus, and non-transitory computer-readable storage medium - Google Patents

Method of calculating relevancy, apparatus for calculating relevancy, data query apparatus, and non-transitory computer-readable storage medium Download PDF

Info

Publication number
WO2019140863A1
WO2019140863A1 PCT/CN2018/091822 CN2018091822W WO2019140863A1 WO 2019140863 A1 WO2019140863 A1 WO 2019140863A1 CN 2018091822 W CN2018091822 W CN 2018091822W WO 2019140863 A1 WO2019140863 A1 WO 2019140863A1
Authority
WO
WIPO (PCT)
Prior art keywords
text
relevancy
similar part
dissimilar
vectors
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2018/091822
Other languages
English (en)
French (fr)
Inventor
Zhenzhong Zhang
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
BOE Technology Group Co Ltd
Original Assignee
BOE Technology Group Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by BOE Technology Group Co Ltd filed Critical BOE Technology Group Co Ltd
Priority to US16/330,545 priority Critical patent/US11281861B2/en
Priority to KR1020207014877A priority patent/KR102411921B1/ko
Priority to JP2019569437A priority patent/JP7513396B2/ja
Priority to EP18901142.2A priority patent/EP3743831A4/en
Publication of WO2019140863A1 publication Critical patent/WO2019140863A1/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3347—Query execution using vector based model
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/3332—Query translation
    • G06F16/3334—Selection or weighting of terms from queries, including natural language queries
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/338—Presentation of query results
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/10—Text processing
    • G06F40/194—Calculation of difference between files
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present invention relates to data processing technology, more particularly, to a method of calculating relevancy, an apparatus for calculating relevancy, a data query apparatus, and a non-transitory computer-readable storage medium.
  • search engines have been developed to facilitate a user to search for useful information based on a keyword entry.
  • a medical professional can search various medical literatures, books, and websites on the internet based on a search query.
  • a patient may seek health-related information on health information websites.
  • the present invention provides a method of calculating relevancy between a first text and a second text, comprising respectively mapping the first text and the second text to vectors; determining a similar part and a dissimilar part between the first text and the second text based on the vectors; and calculating the relevancy between the first text and the second text using both the similar part and the dissimilar part.
  • determining the similar part and the dissimilar part comprises determining a similar part and a dissimilar part of the first text; and determining a similar part and a dissimilar part of the second text.
  • dimension reduction is performed using one or more methods selected from a group consisting of Word2Vec, Sentence2Vec, and Doc2Vec.
  • determining the similar part and dissimilar part comprises performing a semantic content analysis based on the first text and the second text.
  • the semantic content analysis comprises performing a sematic matching process between the first text and the second text thereby obtaining a sematic overlay between the first text and the second text.
  • performing the sematic matching process comprises reconstructing a vector of a word in the first text using word vectors of the second text; and calculating a semantic overlay based on a result of the reconstructing.
  • determining the similar part and dissimilar part comprises performing a semantics decomposition on the first text and the second text.
  • the semantic overlay is calculated based on Equation (1) :
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • the similar part of the first text is calculated based on wherein S i ’represents the similar part of the first text, and A i, j is a matrix of ⁇ i, j ; and S i -S i ’represents the dissimilar part of the first text.
  • the semantic overlay is calculated based on Equation (2) :
  • T i is a d ⁇ 1 column vector of the second text
  • S j is a d ⁇ 1 column vector of the first text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • the similar part of the second text is calculated based on wherein T j ’represents the similar part of the second text, and A i, j is a matrix of ⁇ i, j ; and T j –T j ’represents the dissimilar part of the second text.
  • the relevancy between the first text and the second text is calculated using a Recurrent Neural Network using the similar part and the dissimilar part between the first text and the second text as input.
  • the input comprises a similar part and a dissimilar part of the first text; and a similar part and a dissimilar part of the second text.
  • the method further comprises training the Recurrent Neural Network using a sample data represented as (Ss, Ts, L) , wherein Ss stands for a first sample text, Ts stands for a second sample text, and L represents a sample relevancy.
  • training the Recurrent Neural Network further comprises assigning a first value of granularity to a first relevancy in a first range and a second value of granularity to a second relevancy in a second range.
  • the present invention provides an apparatus for calculating relevancy between a first text and a second text, comprising a memory; and one or more processors; wherein the memory and the one or more processors are connected with each other; the memory stores computer-executable instructions for controlling the one or more processors to respectively map the first text and the second text to vectors; determine a similar part and a dissimilar part between the first text and the second text based on the vectors; and calculate the relevancy between the first text and the second text using both the similar part and the dissimilar part.
  • the memory stores computer-executable instructions for controlling the one or more processors to determine a similar part and a dissimilar part of the first text; and determine a similar part and a dissimilar part of the second text.
  • the memory further stores computer-executable instructions for controlling the one or more processors to, subsequent to respectively mapping the first text and the second text to vectors, perform dimension reduction on the vectors, thereby representing the first text and the second text respectively using low-dimension dense vectors.
  • dimension reduction is performed using one or more methods selected from a group consisting of Word2Vec, Sentence2Vec, and Doc2Vec.
  • the memory further stores computer-executable instructions for controlling the one or more processors to perform a semantic content analysis based on the first text and the second text thereby determining the similar part and dissimilar part.
  • the memory further stores computer-executable instructions for controlling the one or more processors to perform a sematic matching process between the first text and the second text thereby obtaining a sematic overlay between the first text and the second text.
  • the memory further stores computer-executable instructions for controlling the one or more processors to reconstruct a vector of a word in the first text using word vectors of the second text, and calculate a semantic overlay based on a result of reconstruction.
  • the memory further stores computer-executable instructions for controlling the one or more processors to perform a semantics decomposition on the first text and the second text thereby determining the similar part and dissimilar part.
  • the semantic overlay is calculated based on Equation (1) :
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • the similar part of the first text is calculated based on wherein S i ’represents the similar part of the first text, and A i, j is a matrix of ⁇ i, j ; and S i -S i ’represents the dissimilar part of the first text.
  • the semantic overlay is calculated based on Equation (2) :
  • T i is a d ⁇ 1 column vector of the second text
  • S j is a d ⁇ 1 column vector of the first text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • the similar part of the second text is calculated based on wherein T j ’represents the similar part of the second text, and A i, j is a matrix of ⁇ i, j ; and T j –T j ’represents the dissimilar part of the second text.
  • the apparatus further comprises a Recurrent Neural Network using the similar part and the dissimilar part of the first text and the second text as input to calculate the relevancy between the first text and the second text.
  • the input comprises a similar part and a dissimilar part of the first text; and a similar part and a dissimilar part of the second text.
  • the memory further stores computer-executable instructions for controlling the one or more processors to train the Recurrent Neural Network using a sample data represented as (Ss, Ts, L) , wherein Ss stands for a first sample text, Ts stands for a second sample text, and L represents a sample relevancy.
  • the memory further stores computer-executable instructions for controlling the one or more processors to assign a first value of granularity to a first relevancy in a first range and a second value of granularity to a second relevancy in a second range.
  • the present invention provides a data query apparatus, comprising the apparatus for calculating relevancy described herein.
  • the present invention provides a data query method, comprising calculating relevancy between a first text and a second text according to the method of calculating relevancy described herein, the second text being a text selected from a plurality of potential target texts; ranking the plurality of potential target texts based on calculation of relevancy between the plurality of potential target texts and the first text; and selecting a target text out of the plurality of potential target texts based on a resulting of ranking the plurality of potential target texts.
  • the present invention provides a non-transitory computer-readable storage medium storing computer-readable instructions, the computer-readable instructions being executable by a processor to cause the processor to perform respectively map a first text and a second text to vectors; determine a similar part and a dissimilar part between the first text and the second text based on the vectors; and calculate a relevancy between a first text and a second text using both the similar part and the dissimilar part.
  • FIG. 1 is a flow chat illustrating a method of calculating relevancy in some embodiments according to the present disclosure.
  • FIG. 2 is a schematic diagram illustrating the structure of an apparatus for calculating relevancy in some embodiments according to the present disclosure.
  • FIG. 3 is a schematic diagram illustrating the structure of an apparatus for calculating relevancy in some embodiments according to the present disclosure.
  • FIG. 4 is a schematic diagram illustrating the structure of a data query apparatus in some embodiments according to the present disclosure.
  • FIG. 5 is a schematic diagram illustrating an electronic apparatus in some embodiments according to the present disclosure.
  • dissimilarity between texts In a conventional method of calculating relevancy, typically similarity between texts is considered, but not the dissimilarity between texts. In the present disclosure, it is discovered for the first time that dissimilarity between texts also provides important semantic information that facilitates accurate information matching in the process of calculating relevancy. By also considering dissimilarity between texts, a better and more accuracy result can be obtained in calculating relevancy, rendering information that is more useful in response to a user query.
  • the present disclosure provides, inter alia, a method of calculating relevancy, an apparatus for calculating relevancy, a data query apparatus, an electronic apparatus that substantially obviate one or more of the problems due to limitations and disadvantages of the related art.
  • the present disclosure provides a method of calculating relevancy.
  • the method includes respectively mapping the first text and the second text to vectors; determining a similar part and a dissimilar part between the first text and the second text based on the vectors; and calculating the relevancy between the first text and the second text using both the similar part and the dissimilar part.
  • the method includes mapping the first text into a first set of vectors, and mapping the second text into a second set of vectors.
  • the step of determining the similar part and the dissimilar part includes determining a similar part and a dissimilar part of the first text; and determining a similar part and a dissimilar part of the second text.
  • similar part refers to a part of a text having a similarity measure higher than or equal to a threshold value
  • dissimilar part refers to a part of a text having a similarity measure lower than the threshold value
  • a similar part of a first text refers to a part (e.g., a word, a sentence, a document) of a first text having an identical occurrence in the second text or having a sematic overlay between the first text and the second text
  • a similar part of a second text refers to a part (e.g., a word, a sentence, a document) of a second text having an identical occurrence in the first text or having a sematic overlay between the first text and the second text.
  • a dissimilar part of a first text refers to a part (e.g., a word, a sentence, a document) of a first text that does not have an identical occurrence in the second text and does not have a sematic overlay between the first text and the second text
  • a dissimilar part of a second text refers to a part (e.g., a word, a sentence, a document) of a second text that does not have an identical occurrence in the first text and does not have a sematic overlay between the first text and the second text.
  • the term “relevancy” refers to a measure of a degree of relevancy between two texts, e.g., how pertinent, connected, or applicable a text is to another text (e.g., a query text) .
  • values of relevancy may be expressed in granularity, and may not have a continuous relationship (e.g., having a 0, 0.5, or 1 relationship) .
  • the relevancy is calculated by a relevancy algorithm.
  • the relevancy algorithm may be trained using a sample data to calculate relevancy scores.
  • the relevancy is determined based on a plurality of criteria, e.g., considering both the similar part and the dissimilar part of the texts.
  • a user enters a query: “introduction on cold. ” Three results are found in the database.
  • Text 1 This article introduces sinusitis.
  • Text 2 This article introduces influenza, not the common cold.
  • Text 3 This article introduces influenza, not the pneumonia.
  • sim sim (., . ) used to represent the similarities between texts. When the similarities between the query text entered by the user and the three texts found in the database are calculated, it can be relatively easily determined that sim ( “Query Text” , “Text 2” ) > sim ( “Query Text” , “Text 1” ) and sim ( “Query Text” , “Text 3” ) > sim ( “Query Text” , “Text 1” ) .
  • a user enters a query asking for information on a genus A (having a plurality of species, e.g., species i and species ii) .
  • a genus A having a plurality of species, e.g., species i and species ii
  • Three results are found in the database.
  • Text 1 contains information on a closely related genus B or a species in a closely related genus B.
  • Text 2 contain information on difference between species i and species ii.
  • Text 3 contain information on difference between species i and a related genus C.
  • a function sim (., . ) used to represent the similarities between texts.
  • sim ( “Query Text” , “Text 2” ) and sim ( “Query Text” , “Text 3” ) it is relatively more difficult to compare between sim ( “Query Text” , “Text 2” ) and sim ( “Query Text” , “Text 3” ) .
  • Text 2 emphasizes more on the difference between species i and species ii, e.g., the unique characteristics of species i, which are not commonly shared by the species ii.
  • Text 3 emphasizes the difference between species i and the related genus C. Because species i and the related genus C have much less commonly shared characteristics, Text 3 may contain more information about the common characteristics of the genus A. Thus, it follows that Text 3 is more relevant to the Query Text as compared to Text 2.
  • a user enters a query asking for information on a “cancer pain” (having various types such as cancer nerve pain and cancer bone pain) .
  • a “cancer pain” having various types such as cancer nerve pain and cancer bone pain
  • Three results are found in the database.
  • Text 1 contains information on migraine.
  • Text 2 contain information on difference between cancer nerve pain and cancer bone pain.
  • Text 3 contain information on how to tell the difference between cancer bone pain and arthritis pain.
  • a function sim (., . ) used to represent the similarities between texts.
  • sim ( “Query Text” , “Text 2” ) and sim ( “Query Text” , “Text 3” ) it is relatively more difficult to compare between sim ( “Query Text” , “Text 2” ) and sim ( “Query Text” , “Text 3” ) .
  • Text 2 emphasizes more on the difference between cancer nerve pain and cancer bone pain, e.g., the unique characteristics of cancer nerve pain, which are not commonly shared by cancer bone pain.
  • Text 3 emphasizes the difference cancer bone pain and arthritis pain. Because cancer bone pain and arthritis pain are caused by two completely different pathological conditions, Text 3 may contain more information about the common characteristics of cancer pain. Thus, it follows that Text 3 is more relevant to the Query Text as compared to Text 2.
  • FIG. 1 is a flow chat illustrating a method of calculating relevancy in some embodiments according to the present disclosure.
  • a method of calculating relevancy in some embodiments includes at least obtaining a first text S and obtaining a second text T; respectively mapping the first text S and the second text T to vectors; determining a similar part and a dissimilar part between the first text S and the second text T; and calculating the relevancy between the first text S and the second text T using the similar part and the dissimilar part.
  • the step of determining the similar part and the dissimilar part includes determining a similar part and a dissimilar part of the first text; and determining a similar part and a dissimilar part of the second text.
  • the first text S or the second text T is a sentence including a plurality of words.
  • the first text S or the second text T is a paragraph including a plurality of sentences.
  • the first text S or the second text T is a document including a plurality of paragraphs.
  • the first text S or the second text T includes a huge number of documents.
  • the first text S and the second text T is obtained from a single data source. In some embodiments, the first text S and the second text T is obtained from a plurality of data sources.
  • the source of the first text S and the second text T includes texts stored in or outputted by the local devices (e.g., a cellphone, a pad, and a computer) .
  • the source of the first text and the second text includes texts transmitted through internet.
  • the first text S and the second text T are respectively mapped to separate vectors.
  • the first text S is mapped to S1, S2, ..., Sn.
  • the second text T is mapped to T1, T2, ..., Tm. Therefore, S1-Sn is an array of words forming the first text S.
  • S1-Sn is a vector corresponding to the first text S.
  • S “symptom of cold”
  • S1 “symptom”
  • S2 “of”
  • S3 “cold”
  • T1-Tn is an array of words forming the second text T
  • T1-Tn is a vector corresponding to the second text T.
  • T “symptom of influenza”
  • T1 “symptom”
  • T2 “of”
  • T3 “influenza. ”
  • VSM Vector Space Model
  • PLSA Probabilistic Latent Semantic Analysis
  • LDA Latent Dirichlet Allocation
  • the method further includes performing dimension reduction on the vectors, thereby representing the text using low-dimension dense vectors.
  • dimension reduction is performed by Principal component analysis (PCA) . If a multivariate data set is visualized as a set of coordinates in a high-dimensional data space (one axis per variable) , PCA can supply the user with a lower-dimensional picture, a projection of this object when viewed from its most informative viewpoint. This is done by using only the first few principal components so that the dimensionality of the transformed data is reduced. Therefore, PCA can linearly transform multiple variances and obtain a few principal variances.
  • dimension reduction is performed using methods such as Word2Vec, Sentence2Vec, and Doc2Vec.
  • a word embedding process is used to perform dimension reduction.
  • the word embedding process may be conducted using methods such as Word2Vec.
  • the vector S1-Sn of the first text S and the vector T1-Tm of the second text T are respectively mapped to two sets of low-dimension vectors.
  • each word of a text corresponds to a d ⁇ 1 column vector.
  • the first text S corresponds to a d ⁇ n matrix (i.e. matrix P)
  • the second text T corresponds to a d ⁇ m matrix (i.e. matrix Q) .
  • the similar part and dissimilar part between the first text and the second text are determined.
  • the similar part and dissimilar part are determined using a textual encoding method such as Simhash, Shingling, and Minhash.
  • Simhash is used to binary encode the first text and the second text
  • a Hamming distance between the first text and the second text is determined by performing XOR operation (i.e. execute XOR operation on the simhash result of those two texts) .
  • the two texts are determined to be dissimilar (e.g., the dissimilar part) ; if the number of result “1” is less than or equal to three, the two texts are determined to be similar (e.g., the similar part) .
  • the step of determining a similar part and a dissimilar part between the first text S and the second text T includes performing a semantic content analysis based on the first text S and the second text T.
  • the semantic content analysis includes performing a sematic matching process between the first text and the second text thereby obtaining a sematic overlay between the first text S and the second text T.
  • the word “cold” in the Query Text can semantically overlay the word “influenza” in Text 3, but can hardly overlay the word “sinusitis” in Text 1. Therefore, sim ( “Query Text” , “Text 3” ) > sim ( “Query Text” , “Text 1” ) .
  • the step of determining the semantic overlay between the first text S and the second text T includes reconstructing a vector of a word in the first text S (e.g., a word Si) using word vectors of the second text T; and calculating the semantic overlay based on a result of the reconstruction.
  • Equation (1) is used to calculate the semantic overlay of the words of the first text S over the second text T:
  • each of S i and T j is a d ⁇ 1 column vector; each of S i and T j is a word vector obtained from the word embedding process; ⁇ i, j is parameter. ⁇ > 0, and ⁇ is a positive real number.
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number.
  • the parameter ⁇ i, j can be solved by the following method:
  • a convex function f i is provided to satisfy To obtain a minimum value for f i , ⁇ i, j should meet the condition of
  • an n ⁇ m matrix A is provided, with an element in i-th row and j-th column to be ⁇ i, j .
  • the equation can be represented as
  • I is a m ⁇ m unit matrix
  • Equation (3) is used to calculate the semantic overlay of the words of the second text T over the first text S:
  • each of T i and S j is a d ⁇ 1 column vector; each of T i and S j is a word vector obtained from the word embedding process; ⁇ i, j is parameter. ⁇ > 0, and ⁇ is a positive real number.
  • T i is a d ⁇ 1 column vector of the first text
  • S j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number.
  • the semantics decomposition can be used to obtain the similar part and the dissimilar part between the first text S and the second text T (e.g., the similar part and dissimilar part of the first text S when comparing the first text S to the second text T, and the similar part and dissimilar part of the second text T when comparing the second text T to the first text S.
  • the matrix A (see equation (2) ) is based on how every word in the first text S is represented or overlaid in the second text T, wherein the i-th row of the matrix A shows how word S i is represented or overlaid in text T.
  • S i represents the similar part of the first text S, e.g., a part of word S i of the first text S that can be overlaid by the second text T.
  • S i -S i ’ represents the dissimilar part of the first text S.
  • the first text S is represented by two parts, similar part (represented by a d ⁇ n matrix, refer to the similar part S′ + in FIG.
  • the text can be represented by a d ⁇ 2n matrix instead of a d ⁇ n matrix.
  • calculating the similar part and dissimilar part between the second text T and the first text S includes adopting the equation wherein A i, j is the matrix of ⁇ i, j , T j ’represents the similar part of the second text T, and T j –T j ’represents the dissimilar part of the second text T.
  • the second text T includes two parts, a similar part (represented by a d ⁇ m matrix, refer to the similar part T′ + in FIG. 1) and a dissimilar part (represented by a d ⁇ m matrix, refer to the dissimilar part T′ - in FIG. 2) .
  • the relevancy between the first text S and the second text T can be calculated.
  • calculating the relevancy includes determining a Jaccard similarity through textual encoding.
  • calculating the relevancy is performed by textual semantics calculation.
  • calculating the relevancy is performed by machine learning.
  • the relevancy between the first text and the second text can be calculated using a Recurrent Neural Network (RNN) .
  • RNN Recurrent Neural Network
  • using RNN includes using sample data to train RNN.
  • the sample data is represented as (Ss, Ts, L) , wherein Ss and Ts represent sample texts (e.g., Ss represent a first sample text, and Ts represent a second sample text) , and L represents relevancy.
  • Ss and Ts are the originally inputted first text S and second text T.
  • Ss and Ts are sample texts different from the first text S and the second text T.
  • the granularity of the relevancy can be assigned.
  • the granularity is 2, L ⁇ ⁇ 0, 1 ⁇ , wherein 0 represents irrelevant, and 1 represents relevant. In another example, the granularity is 3, L ⁇ ⁇ 0, 0.5, 1 ⁇ , wherein 0 represents irrelevant, 0.5 represents somewhat relevant, 1 represents very relevant.
  • the sample data is inputted in the RNN.
  • a composite relevancy based on the similar part of the first text S, the similar part of the second text T, the dissimilar part of the first text S, and the dissimilar part of the second text T is then calculated.
  • the granularity is set to 2
  • a composite relevancy in a first range is defined to have a granularity of 1
  • a composite relevancy in a second range different from the first range is defined to have a granularity of 0.
  • a composite relevancy in a first range is defined to have a granularity of 1
  • a composite relevancy in a second range is defined to have a granularity of 0.5
  • a composite relevancy in a third range is defined to have a granularity of 0, the first range, the second range, and the third range being different from each other.
  • the RNN can be used to calculate the relevancy based on input information.
  • the present method In the present method of calculating the relevancy, the effects of both the similar part and the dissimilar part between two texts are considered. Consequently, the present method enables a relevancy calculation with a finer granularity classification, thereby enhancing the accuracy of textual matching, and rendering highly accurate search or query results.
  • FIG. 2 and FIG. 3 are schematic diagrams illustrating the structure of an apparatus for calculating relevancy in some embodiments according to the present disclosure.
  • the apparatus for calculating relevancy has a data acquirer 200 configured to obtain a first text and a second text; a mapping engine 210 configured to respectively map the first text and the second text to vectors; an relevancy calculator 220 configured to determine a similar part and a dissimilar part between the first text and the second text based on the vectors, and calculate the relevancy between the first text and the second text based on the similar part and the dissimilar part.
  • the relevancy calculator 220 configured to determine a similar part and a dissimilar part of the first text, and determining a similar part and a dissimilar part of the second text.
  • the mapping engine 210 includes a dimension reduction processor 211 configured to reduce the dimension of vectors respectively corresponding to the first text and the second text.
  • the relevancy calculator 220 includes a semantics matching processor 221 configured to perform semantic matching between the first text and the second text; a semantics decomposition processor 222 configured to determine the similar part and the dissimilar part between the first text and the second text; a calculator 223 configured to calculate the relevancy between the first text and the second text basing on the similar part and the dissimilar part.
  • the semantics decomposition processor 222 configured to determine a similar part and a dissimilar part of the first text, and determining a similar part and a dissimilar part of the second text.
  • the semantics matching processor 221 is configured to determine semantic overlay between the first text and the second text by reconstructing the first text’s vector using the second text’s vector, e.g., by reconstructing a vector of a word in the first text using word vectors of the second text.
  • the semantics matching processor 221 is configured to use Equation (1) , to calculate the semantic overlay of the words of the first text S over the second text T, wherein, S i is a column vector (e.g., a d ⁇ 1 column vector) of the first text; T j is the column vector (e.g., a d ⁇ 1 column vector) of the second text; ⁇ i, j is a parameter for semantic overly; ⁇ > 0, and ⁇ is a positive real number.
  • Equation (1) Equation (1) , to calculate the semantic overlay of the words of the first text S over the second text T, wherein, S i is a column vector (e.g., a d ⁇ 1 column vector) of the first text; T j is the column vector (e.g., a d ⁇ 1 column vector) of the second text; ⁇ i, j is a parameter for semantic overly; ⁇ > 0, and ⁇ is a positive real number.
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • the semantics matching processor 221 is configured to use Equation (3) , to calculate the semantic overlay of the words of the second text T over the first text S, wherein, T i is a column vector (e.g., a d ⁇ 1 column vector) of the second text; S j is the column vector (e.g., a d ⁇ 1 column vector) of the first text; ⁇ i, j is a parameter for semantic overly; ⁇ > 0, and ⁇ is a positive real number.
  • T i is a d ⁇ 1 column vector of the first text
  • S j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • the semantics decomposition processor 222 is configured to obtain the similar part and the dissimilar part of the first text based on the equation of wherein A i, j is the matrix of ⁇ i, j ; S i ’stands for the similar part of the first text; and S i -S i ’stands for the dissimilar part of the second text.
  • the semantics decomposition processor 222 is configured to obtain the similar part and the dissimilar part of the second text based on the equation of wherein A i, j is the matrix of ⁇ i, j ; T j ’stands for the similar part of the first text; and T j –T j ’stands for the dissimilar part of the second text.
  • the apparatus for calculating relevancy includes a memory; and one or more processor.
  • the memory and the at least one processor are connected with each other.
  • the memory stores computer-executable instructions for controlling the one or more processors to respectively map the first text and the second text to vectors; determine a similar part and a dissimilar part between the first text and the second text based on the vectors; and calculate the relevancy between the first text and the second text using both the similar part and the dissimilar part.
  • the memory further stores computer-executable instructions for controlling the one or more processors to, subsequent to respectively mapping the first text and the second text to vectors, perform dimension reduction on the vectors, thereby representing the text using low-dimension dense vectors.
  • dimension reduction is performed using one or more methods selected from a group consisting of Word2Vec, Sentence2Vec, and Doc2Vec.
  • the memory stores computer-executable instructions for controlling the one or more processors to determine a similar part and a dissimilar part of the first text; and determine a similar part and a dissimilar part of the second text.
  • the memory further stores computer-executable instructions for controlling the one or more processors to perform a semantic content analysis based on the first text and the second text thereby determining the similar part and dissimilar part.
  • the memory further stores computer-executable instructions for controlling the one or more processors to perform a sematic matching process between the first text and the second text thereby obtaining a sematic overlay between the first text and the second text.
  • the memory further stores computer-executable instructions for controlling the one or more processors to reconstruct a vector of a word in the first text S using word vectors of the second text, and calculate the semantic overlay based on a result of reconstruction.
  • the semantic overlay of the words of the first text S over the second text T is calculated based on Equation (1) :
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation; ⁇ > 0
  • ⁇ is a positive real number.
  • the semantic overlay of the words of the second text T over the first text S is calculated based on Equation (3) ,
  • T i is a column vector (e.g., a d ⁇ 1 column vector) of the second text
  • S j is the column vector (e.g., a d ⁇ 1 column vector) of the first text
  • ⁇ i, j is a parameter for semantic overly
  • ⁇ > 0, and ⁇ is a positive real number.
  • T i is a d ⁇ 1 column vector of the first text
  • S j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation; ⁇ > 0; and ⁇ is a positive real number.
  • the similar part of the first text is calculated based on S i ’represents the similar part of the first text, and A i, j is a matrix of ⁇ i, j .
  • S i -S i ’ represents the dissimilar part of the first text.
  • the similar part of the second text is calculated based on represents the similar part of the second text, and A i, j is a matrix of ⁇ i, j .
  • T j –T j ’ represents the dissimilar part of the second text.
  • the memory further stores computer-executable instructions for controlling the one or more processors to perform a semantics decomposition on the first text and the second text thereby determining the similar part and dissimilar part.
  • the apparatus further includes a Recurrent Neural Network using the similar part and the dissimilar part of the first text and the second text as input to calculate the relevancy between the first text and the second text.
  • the memory further stores computer-executable instructions for controlling the one or more processors to train the Recurrent Neural Network using a sample data represented as (Ss, Ts, L) , wherein Ss stands for a first sample text, Ts stands for a second sample text, and L represents a sample relevancy.
  • Ss and Ts are the originally inputted first text S and second text T.
  • Ss and Ts are sample texts different from the first text S and the second text T.
  • the memory further stores computer-executable instructions for controlling the one or more processors to assign a first value of granularity to a first relevancy in a first range and a second value of granularity to a second relevancy in a second range.
  • the present disclosure also provides a non-transitory computer-readable storage medium storing computer-readable instructions.
  • the computer-readable instructions are executable by a processor to cause the processor to perform one or more steps of the method described herein.
  • the computer-readable instructions are executable by a processor to cause the processor to respectively map the first text and the second text to vectors; determine a similar part and a dissimilar part between the first text and the second text based on the vectors; and calculate the relevancy between the first text and the second text using both the similar part and the dissimilar part.
  • the computer-readable instructions are executable by a processor to further cause the processor to, subsequent to respectively mapping the first text and the second text to vectors, perform dimension reduction on the vectors, thereby representing the text using low-dimension dense vectors.
  • dimension reduction is performed using one or more methods selected from a group consisting of Word2Vec, Sentence2Vec, and Doc2Vec.
  • the computer-readable instructions are executable by a processor to cause the processor to determine a similar part and a dissimilar part of the first text; and determine a similar part and a dissimilar part of the second text.
  • the computer-readable instructions are executable by a processor to cause the processor to perform a semantic content analysis based on the first text and the second text thereby determining the similar part and dissimilar part.
  • the computer-readable instructions are executable by a processor to cause the processor to perform a sematic matching process between the first text and the second text thereby obtaining a sematic overlay between the first text and the second text.
  • the computer-readable instructions are executable by a processor to cause the processor to reconstruct a vector of a word in the first text S using word vectors of the second text, and calculate the semantic overlay based on a result of reconstruction.
  • the semantic overlay of the words of the first text S over the second text T is calculated based on Equation (1) :
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation
  • ⁇ is a positive real number
  • S i is a d ⁇ 1 column vector of the first text
  • T j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation; ⁇ > 0
  • ⁇ is a positive real number.
  • the semantic overlay of the words of the second text T over the first text S is calculated based on Equation (3) :
  • T i is a column vector (e.g., a d ⁇ 1 column vector) of the second text
  • S j is the column vector (e.g., a d ⁇ 1 column vector) of the first text
  • ⁇ i, j is a parameter for semantic overly
  • ⁇ > 0, and ⁇ is a positive real number.
  • T i is a d ⁇ 1 column vector of the first text
  • S j is a d ⁇ 1 column vector of the second text
  • ⁇ i, j is parameter for semantic overlay calculation; ⁇ > 0; and ⁇ is a positive real number.
  • the similar part of the first text is calculated based on S i ’represents the similar part of the first text, and A i, j is a matrix of ⁇ i, j .
  • S i -S i ’ represents the dissimilar part of the first text.
  • the similar part of the second text is calculated based on T j ’represents the similar part of the second text, and A i, j is a matrix of ⁇ i, j .
  • T j –T j ’ represents the dissimilar part of the second text.
  • the computer-readable instructions are executable by a processor to cause the processor to perform a semantics decomposition on the first text and the second text thereby determining the similar part and dissimilar part.
  • the present disclosure also provides a data query apparatus.
  • the data query apparatus includes the apparatus for calculating relevancy as described herein.
  • FIG. 4 is a schematic diagram illustrating the structure of a data query apparatus in some embodiments according to the present disclosure.
  • the data query apparatus in some embodiments includes a human-computer interaction interface 401.
  • the human-computer interaction interface 401 is a medical question answering platform, which can receive patients’questions and display the corresponding answers.
  • the human-computer interaction interface 401 is a medical literature searching system, which can receive the content the user is looking for and display the corresponding literature.
  • the data query apparatus further includes a content matching processor 402 and a database 403.
  • the content matching processor 402 includes the relevancy calculator 220 described herein.
  • the content matching processor 402 collects the potential target contents from the database 403.
  • the content matching processor 402 adopts the relevancy calculating methods described herein to calculate the relevancy between the potential target contents and the content entered by the user.
  • the content matching processor 402 selects the content the user looks for from the potential target contents based on a ranking of relevancy.
  • the present disclosure also provides a data query method.
  • the data query method includes calculating relevancy between a first text and a second text according to the method of calculating relevancy described herein.
  • the second text is a text selected from a plurality of potential target texts.
  • the data query method further includes ranking the plurality of potential target texts based on calculation of relevancy between the plurality of potential target texts and the first text; and selecting a target text out of the plurality of potential target texts based on a resulting of ranking the plurality of potential target texts.
  • the method further includes outputting the target text as a result of data query to a user.
  • FIG. 5 is a schematic diagram illustrating the electronic apparatus in some embodiments according to the present disclosure.
  • the electronic apparatus 500 includes a processor 502.
  • the processor 502 is configured to run computer instructions to perform at least one step of the relevancy calculation described herein.
  • the electronic apparatus 500 includes at least a memory 501 connected to at least one the processor 502.
  • the memory 501 is configured to store computer instructions.
  • the electronic apparatus 500 is a local end apparatus, which means the electronic apparatus 500 can calculate the relevancy on the user end.
  • the electronic apparatus 500 is a local and remote interaction apparatus, i.e. the electronic apparatus 500 obtains at least a first text and a second text at the user end apparatus (i.e. local end) , and the remote internet processor (i.e. remote end) receives the first text and the second text and perform the similarity calculation.
  • the electronic apparatus 500 can include a plurality of user end apparatus and a plurality of remote internet processor connecting to the user end apparatus.
  • the plurality of the user end apparatus upload information containing at least a first text and a second text to the remote internet processor.
  • the information uploaded is the query text entered or stored by the user.
  • the remote internet processor collects the information uploaded by every user end apparatus.
  • the remote internet processor then performs the relevancy calculation.
  • the memory 501 can be formed by any type of volatile storage devices or non-volatile storage devices or the combination of both the volatile and non-volatile storage devices.
  • the types of volatile storage devices or non-volatile storage devices include Static Random-Access Memory (SRAM) , Electrically Erasable Programmable Read-only Memory (EEPROM) , Erasable Programmable Read-only Memory (EPROM) , Programmable Read-only Memory (PROM) , Read-only Memory (ROM) , magnetic memory, flash memory, disk or CD.
  • the processor 502 is a logical operation apparatus having data processing capabilities and/or program execution capabilities, such as Central Processing Unit (CPU) , Field-Programmable Gate Array (FPGA) , Microcontroller Unit (MCU) , Digital Signal Processor (DSP) , Application-Specific Integrated Circuit (ASIC) , graphics processing unit (GPU) .
  • CPU Central Processing Unit
  • FPGA Field-Programmable Gate Array
  • MCU Microcontroller Unit
  • DSP Digital Signal Processor
  • ASIC Application-Specific Integrated Circuit
  • GPU graphics processing unit
  • one or a plurality of the processor can be configured to execute the relevancy calculation concurrently with a parallel processor.
  • one or a plurality of the processor can be configured to execute part of the relevancy calculation.
  • other processors can be configured to execute the remaining part of the relevancy calculation.
  • the computer instructions include at least one processor’s operation defined by the instruction set corresponding to the processor.
  • the computer instruction can be logically contained and represented by at least one computer programs.
  • the electronic apparatus 500 can be connected to input apparatus such as a user interface, a keyboard.
  • the electronic apparatus 500 can be connected to output apparatus such as a speaker.
  • the electronic apparatus can be connected to the apparatus performing the interaction between user and the electronic apparatus 500, such as a display device.
  • the electronic apparatus 500 can be connected to various input apparatus, output apparatus or other interaction apparatus through networks including wireless network, wired network, and/or any combination of wireless and wired network.
  • the network includes a local area network, the internet, a telecommunication network, an internet of thins based on the internet and/or a telecommunications network., and any combination of the above-mentioned network.
  • the wired network can use different materials to communicate, including twisted pair, coaxial able, and fiber.
  • the wireless network can also use different materials to communicate, including a 3G/4G/5G mobile communication network, Bluetooth, Zigbee, and Wi-Fi.
  • the present disclosure also provides a computer readable storage medium, which is configured to store computer instructions. When the computer instructions are executed, at least one of the step of the relevancy calculation is performed.
  • the term “the invention” , “the present invention” or the like does not necessarily limit the claim scope to a specific embodiment, and the reference to exemplary embodiments of the invention does not imply a limitation on the invention, and no such limitation is to be inferred.
  • the invention is limited only by the spirit and scope of the appended claims.
  • these claims may refer to use “first” , “second” , etc. following with noun or element.
  • Such terms should be understood as a nomenclature and should not be construed as giving the limitation on the number of the elements modified by such nomenclature unless specific number has been given. Any advantages and benefits described may not apply to all embodiments of the invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
PCT/CN2018/091822 2018-01-22 2018-06-19 Method of calculating relevancy, apparatus for calculating relevancy, data query apparatus, and non-transitory computer-readable storage medium Ceased WO2019140863A1 (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
US16/330,545 US11281861B2 (en) 2018-01-22 2018-06-19 Method of calculating relevancy, apparatus for calculating relevancy, data query apparatus, and non-transitory computer-readable storage medium
KR1020207014877A KR102411921B1 (ko) 2018-01-22 2018-06-19 관련성을 계산하는 방법, 관련성을 계산하기 위한 장치, 데이터 쿼리 장치, 및 비-일시적 컴퓨터-판독가능 저장 매체
JP2019569437A JP7513396B2 (ja) 2018-01-22 2018-06-19 関連性を算出する方法、関連性を算出する装置、データクエリ装置及び非一時的なコンピュータ可読記録媒体
EP18901142.2A EP3743831A4 (en) 2018-01-22 2018-06-19 PROCESS FOR CALCULATION OF RELEVANCE, DEVICE FOR CALCULATION OF RELEVANCE, DEVICE FOR DATA REQUEST AND NON-TRANSITORY COMPUTER-READABLE STORAGE MEDIUM

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810060942.7A CN108170684B (zh) 2018-01-22 2018-01-22 文本相似度计算方法及系统、数据查询系统和计算机产品
CN201810060942.7 2018-01-22

Publications (1)

Publication Number Publication Date
WO2019140863A1 true WO2019140863A1 (en) 2019-07-25

Family

ID=62515240

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/091822 Ceased WO2019140863A1 (en) 2018-01-22 2018-06-19 Method of calculating relevancy, apparatus for calculating relevancy, data query apparatus, and non-transitory computer-readable storage medium

Country Status (6)

Country Link
US (1) US11281861B2 (cg-RX-API-DMAC7.html)
EP (1) EP3743831A4 (cg-RX-API-DMAC7.html)
JP (1) JP7513396B2 (cg-RX-API-DMAC7.html)
KR (1) KR102411921B1 (cg-RX-API-DMAC7.html)
CN (1) CN108170684B (cg-RX-API-DMAC7.html)
WO (1) WO2019140863A1 (cg-RX-API-DMAC7.html)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022269510A1 (en) * 2021-06-23 2022-12-29 Servicenow Canada Inc. Method and system for interactive searching based on semantic similarity of semantic representations of text objects

Families Citing this family (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108170684B (zh) * 2018-01-22 2020-06-05 京东方科技集团股份有限公司 文本相似度计算方法及系统、数据查询系统和计算机产品
CN110688119A (zh) * 2018-07-04 2020-01-14 阿里巴巴集团控股有限公司 业务功能和单元的生成、实现方法和服务器、设备、介质
CN109145085B (zh) * 2018-07-18 2020-11-27 北京市农林科学院 语义相似度的计算方法及系统
CN110852056B (zh) * 2018-07-25 2024-09-24 中兴通讯股份有限公司 一种获取文本相似度的方法、装置、设备及可读存储介质
CN110852057A (zh) * 2018-07-27 2020-02-28 北京京东尚科信息技术有限公司 一种计算文本相似度的方法和装置
CN111382246B (zh) * 2018-12-29 2023-03-14 深圳市优必选科技有限公司 文本的匹配方法、匹配装置、终端及计算机可读存储介质
CN110147429B (zh) * 2019-04-15 2023-08-15 平安科技(深圳)有限公司 文本比较方法、装置、计算机设备和存储介质
CN110265104B (zh) * 2019-05-08 2022-04-29 长沙市中心医院 诊断报告符合度检测方法、装置、计算机设备和存储介质
CN110176315B (zh) * 2019-06-05 2022-06-28 京东方科技集团股份有限公司 医疗问答方法及系统、电子设备、计算机可读介质
CN110750977B (zh) * 2019-10-23 2023-06-02 支付宝(杭州)信息技术有限公司 一种文本相似度计算方法及系统
CN110969006B (zh) * 2019-12-02 2023-03-21 支付宝(杭州)信息技术有限公司 一种文本排序模型的训练方法和系统
CN111159343A (zh) * 2019-12-26 2020-05-15 上海科技发展有限公司 基于文本嵌入的文本相似性搜索方法、装置、设备和介质
CN111160028B (zh) * 2019-12-31 2023-05-16 东软集团股份有限公司 判断两个文本语义相似度的方法、装置、存储介质及设备
WO2021191938A1 (ja) * 2020-03-23 2021-09-30 日本電信電話株式会社 要約生成装置、要約生成方法及びプログラム
DE102020109953A1 (de) * 2020-04-09 2021-10-14 Contexon GmbH Verfahren und Computersystem zur Bestimmung der Relevanz eines Textes
CN112819893A (zh) * 2021-02-08 2021-05-18 北京航空航天大学 三维语义地图的构建方法和装置
CN114973219B (zh) * 2021-02-25 2025-12-12 阿里巴巴集团控股有限公司 一种文字内容提取方法和装置
CN113761868B (zh) * 2021-04-20 2023-06-09 腾讯科技(深圳)有限公司 文本处理方法、装置、电子设备及可读存储介质
CN114168837A (zh) * 2021-11-18 2022-03-11 深圳市梦网科技发展有限公司 一种Chatbot的搜索方法、设备及存储介质
CN114461789B (zh) * 2022-02-10 2026-05-22 北京京东尚科信息技术有限公司 关联性预测模型训练方法和装置、摘要生成方法和装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080114750A1 (en) 2006-11-14 2008-05-15 Microsoft Corporation Retrieval and ranking of items utilizing similarity
US20140249799A1 (en) * 2013-03-04 2014-09-04 Microsoft Corporation Relational similarity measurement
CN107133202A (zh) * 2017-06-01 2017-09-05 北京百度网讯科技有限公司 基于人工智能的文本校验方法和装置
CN107220220A (zh) * 2016-03-22 2017-09-29 索尼公司 用于文本处理的电子设备和方法
CN107491547A (zh) * 2017-08-28 2017-12-19 北京百度网讯科技有限公司 基于人工智能的搜索方法和装置
CN108170684A (zh) * 2018-01-22 2018-06-15 京东方科技集团股份有限公司 文本相似度计算方法及系统、数据查询系统和计算机产品

Family Cites Families (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6658626B1 (en) * 1998-07-31 2003-12-02 The Regents Of The University Of California User interface for displaying document comparison information
US6990628B1 (en) * 1999-06-14 2006-01-24 Yahoo! Inc. Method and apparatus for measuring similarity among electronic documents
US6560620B1 (en) * 1999-08-03 2003-05-06 Aplix Research, Inc. Hierarchical document comparison system and method
US6766316B2 (en) * 2001-01-18 2004-07-20 Science Applications International Corporation Method and system of ranking and clustering for document indexing and retrieval
US20040093564A1 (en) * 2002-11-07 2004-05-13 International Business Machines Corporation Method and apparatus for visualizing changes in data
JP4391157B2 (ja) 2003-08-15 2009-12-24 株式会社リコー 文書処理装置、文書処理方法、文書処理プログラム、および記録媒体
CA2601768C (en) * 2005-03-18 2016-08-23 Wink Technologies, Inc. Search engine that applies feedback from users to improve search results
US9892111B2 (en) * 2006-10-10 2018-02-13 Abbyy Production Llc Method and device to estimate similarity between documents having multiple segments
US9189482B2 (en) * 2012-10-10 2015-11-17 Abbyy Infopoisk Llc Similar document search
US8082233B2 (en) * 2007-03-29 2011-12-20 Microsoft Corporation Comparing data sets through identification of matching blocks
US8122032B2 (en) * 2007-07-20 2012-02-21 Google Inc. Identifying and linking similar passages in a digital text corpus
US7945525B2 (en) * 2007-11-09 2011-05-17 International Business Machines Corporation Methods for obtaining improved text similarity measures which replace similar characters with a string pattern representation by using a semantic data tree
US20090240498A1 (en) * 2008-03-19 2009-09-24 Microsoft Corporation Similiarity measures for short segments of text
JP5458880B2 (ja) * 2009-03-02 2014-04-02 富士通株式会社 文書検査装置、コンピュータ読み取り可能な記録媒体および文書検査方法
WO2011029474A1 (en) * 2009-09-09 2011-03-17 Universität Bremen Document comparison
JP2013149061A (ja) * 2012-01-19 2013-08-01 Nec Corp 文書類似性評価システム、文書類似性評価方法およびコンピュータ・プログラム
US9720979B2 (en) * 2012-06-22 2017-08-01 Krishna Kishore Dhara Method and system of identifying relevant content snippets that include additional information
US20150212993A1 (en) * 2014-01-30 2015-07-30 International Business Machines Corporation Impact Coverage
US9477654B2 (en) * 2014-04-01 2016-10-25 Microsoft Corporation Convolutional latent semantic models and their applications
US20160196342A1 (en) * 2015-01-06 2016-07-07 Inha-Industry Partnership Plagiarism Document Detection System Based on Synonym Dictionary and Automatic Reference Citation Mark Attaching System
JP2016197289A (ja) * 2015-04-02 2016-11-24 日本電信電話株式会社 パラメタ学習装置、類似度算出装置、方法、及びプログラム
JP2017010249A (ja) * 2015-06-22 2017-01-12 日本電信電話株式会社 パラメタ学習装置、文類似度算出装置、方法、及びプログラム
JP6555810B2 (ja) * 2015-07-15 2019-08-07 日本放送協会 類似度算出装置、類似検索装置、および類似度算出プログラム
CN105138647A (zh) * 2015-08-26 2015-12-09 陕西师范大学 一种基于Simhash算法的旅游网络社区划分方法
US9836671B2 (en) 2015-08-28 2017-12-05 Microsoft Technology Licensing, Llc Discovery of semantic similarities between images and text
CN106485507B (zh) * 2015-09-01 2019-10-18 阿里巴巴集团控股有限公司 一种软件推广作弊的检测方法、装置及系统
CN105893410A (zh) 2015-11-18 2016-08-24 乐视网信息技术(北京)股份有限公司 一种关键词提取方法和装置
US10102191B2 (en) * 2016-06-16 2018-10-16 Adobe Systems Incorporated Propagation of changes in master content to variant content
US10769522B2 (en) * 2017-02-17 2020-09-08 Wipro Limited Method and system for determining classification of text
CN107436864B (zh) * 2017-08-04 2021-03-02 识因智能科技(北京)有限公司 一种基于Word2Vec的中文问答语义相似度计算方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080114750A1 (en) 2006-11-14 2008-05-15 Microsoft Corporation Retrieval and ranking of items utilizing similarity
US20140249799A1 (en) * 2013-03-04 2014-09-04 Microsoft Corporation Relational similarity measurement
CN107220220A (zh) * 2016-03-22 2017-09-29 索尼公司 用于文本处理的电子设备和方法
CN107133202A (zh) * 2017-06-01 2017-09-05 北京百度网讯科技有限公司 基于人工智能的文本校验方法和装置
CN107491547A (zh) * 2017-08-28 2017-12-19 北京百度网讯科技有限公司 基于人工智能的搜索方法和装置
CN108170684A (zh) * 2018-01-22 2018-06-15 京东方科技集团股份有限公司 文本相似度计算方法及系统、数据查询系统和计算机产品

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3743831A4

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022269510A1 (en) * 2021-06-23 2022-12-29 Servicenow Canada Inc. Method and system for interactive searching based on semantic similarity of semantic representations of text objects
US12259912B2 (en) 2021-06-23 2025-03-25 Servicenow Canada Inc. Method and system for interactive searching based on semantic similarity of semantic representations of text objects

Also Published As

Publication number Publication date
CN108170684A (zh) 2018-06-15
EP3743831A4 (en) 2021-09-29
JP2021510858A (ja) 2021-04-30
KR20200078576A (ko) 2020-07-01
KR102411921B1 (ko) 2022-06-23
CN108170684B (zh) 2020-06-05
EP3743831A1 (en) 2020-12-02
US20210334464A1 (en) 2021-10-28
US11281861B2 (en) 2022-03-22
JP7513396B2 (ja) 2024-07-09

Similar Documents

Publication Publication Date Title
US11281861B2 (en) Method of calculating relevancy, apparatus for calculating relevancy, data query apparatus, and non-transitory computer-readable storage medium
CN114547267B (zh) 智能问答模型的生成方法、装置、计算设备和存储介质
CN112119388A (zh) 训练图像嵌入模型和文本嵌入模型
US11023503B2 (en) Suggesting text in an electronic document
CN112074828A (zh) 训练图像嵌入模型和文本嵌入模型
US8612369B2 (en) System and methods for finding hidden topics of documents and preference ranking documents
CN120873195A (zh) 用于基于图谱的动态信息检索及合成的系统和方法
US20200045122A1 (en) Method and apparatus for pushing information
WO2019052261A1 (zh) 用于问答服务的方法、问答服务系统以及存储介质
CN117271818B (zh) 视觉问答方法、系统、电子设备及存储介质
US11113476B2 (en) Machine learning based intent resolution for a digital assistant
US20190095788A1 (en) Supervised explicit semantic analysis
CN111539197A (zh) 文本匹配方法和装置以及计算机系统和可读存储介质
CN107368515A (zh) 应用程序页面推荐方法及系统
CN110390052A (zh) 搜索推荐方法、ctr预估模型的训练方法、装置及设备
CN106156023B (zh) 语义匹配的方法、装置和系统
CN112132075A (zh) 图文内容处理方法及介质
US20220058464A1 (en) Information processing apparatus and non-transitory computer readable medium
JP7257553B2 (ja) 化粧品製品推奨のための方法及び装置
CN106407316A (zh) 基于主题模型的软件问答推荐方法和装置
CN117009621A (zh) 信息搜索方法、装置、电子设备、存储介质及程序产品
US12536449B1 (en) Self-supervised retriever optimization via attention-derived feedback in retrieval augmented generation systems
CN110287396B (zh) 文本匹配方法及装置
CN117149967A (zh) 应答生成方法、装置、服务器和计算机可读存储介质
Liang et al. DeepLTRS: A deep latent recommender system based on user ratings and reviews

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18901142

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2019569437

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20207014877

Country of ref document: KR

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018901142

Country of ref document: EP

Effective date: 20200824

WWW Wipo information: withdrawn in national office

Ref document number: 2018901142

Country of ref document: EP