CN111639165A - Intelligent question-answer optimization method based on natural language processing and deep learning - Google Patents
Intelligent question-answer optimization method based on natural language processing and deep learning Download PDFInfo
- Publication number
- CN111639165A CN111639165A CN202010364914.1A CN202010364914A CN111639165A CN 111639165 A CN111639165 A CN 111639165A CN 202010364914 A CN202010364914 A CN 202010364914A CN 111639165 A CN111639165 A CN 111639165A
- Authority
- CN
- China
- Prior art keywords
- similarity
- word
- calculating
- semantic
- node
- 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.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an intelligent question-answer optimization method based on natural language processing and deep learning, which is high in accuracy. The invention relates to an intelligent question-answer optimization method based on natural language processing and deep learning, which comprises the following steps: (10) chinese word segmentation and word vector conversion: capturing semantic information of a specific target field, and converting word characters in natural language into N-dimensional matrix vectors which can be understood by a computer; (20) and (3) natural language processing: giving weight to the relation between the ontology semantic concepts, calculating semantic distance by using the weight, and describing semantic similarity based on the semantic distance between the concepts; (30) language disambiguation: and eliminating language ambiguity through threshold comparison.
Description
Technical Field
The invention belongs to the field of computer artificial intelligence, in particular to an intelligent question-answer optimization method based on natural language processing and deep learning
Background
The intelligent question answering is an information retrieval system which receives a question described by a user in a natural language form and searches accurate answers capable of answering the question from a large number of heterogeneous data sources by using technologies such as data processing, query expansion and the like. Different from a traditional search engine which is based on keyword query and returns a document link set, the intelligent question-answering system relates to the fields of knowledge representation, IR (Information Retrieval), NLR (Natural Language Processing) and the like, can effectively help a user to extract effective Information from mass Information resources, has more intelligent human-computer interaction experience, and is a new international research hotspot.
The current method for implementing intelligent question answering generally comprises the following steps: firstly, analyzing question sentences for natural language form questions input by a user, extracting syntax forms of the question sentences, and capturing key words and main sentence meanings in the question sentences through a word segmentation model; secondly, extracting a document set from a document library by using the extracted keywords and sentence meanings and using an information retrieval technology; thirdly, searching matched characteristic values from the sorted document set, sorting an answer set according to the characteristic values, and finally, scoring and sorting the answer set, extracting the best answer and returning the best answer to the user.
However, the current implementation method of intelligent question answering is usually based on the logical combination of keywords, the processing of semantic information cannot be touched by adopting an indexing and matching algorithm, and meanwhile, the problems that the search result is not related to the semantics of the question answering content and the information is redundant are brought, so that the accuracy of the question answering feedback result is influenced to a certain extent.
Disclosure of Invention
The invention aims to provide an intelligent question-answer optimization method based on natural language processing and deep learning, which is high in accuracy.
The technical solution for realizing the purpose of the invention is as follows:
an intelligent question-answer optimization method based on natural language processing and deep learning comprises the following steps:
(10) chinese word segmentation and word vector conversion: capturing semantic information of a specific target field, and converting word characters in natural language into N-dimensional matrix vectors which can be understood by a computer;
(20) and (3) natural language processing: giving weight to the relation between the ontology semantic concepts, calculating semantic distance by using the weight, and describing semantic similarity based on the semantic distance between the concepts;
(30) language disambiguation: and eliminating language ambiguity through threshold comparison.
Compared with the prior art, the invention has the following remarkable advantages:
the intelligent question answering accuracy is high: the invention optimizes the semantic matching algorithm and the feedback screening process in the intelligent question-answering step, realizes the vectorization of the question-answering semantics through word vector processing, and performs similarity calculation by utilizing semantic vectors, thereby greatly simplifying the matching problem of the semantics and the document library and simultaneously reducing the information redundancy. After the question-answer feedback is obtained, the accuracy of intelligent question-answer is improved through threshold comparison.
The invention is described in further detail below with reference to the figures and the detailed description.
Drawings
Fig. 1 is a main flow chart of the intelligent question-answering optimization method based on natural language processing and deep learning.
Fig. 2 is a word vector training flow diagram.
Fig. 3 is a flowchart of the natural language processing steps of fig. 1.
FIG. 4 is a semantic similarity hierarchy model.
FIG. 5 is a flow chart of a specific implementation of semantic similarity calculation.
Detailed description of the preferred embodiments
The invention provides an intelligent question-answer optimization method based on a natural language processing technology and a deep learning technology, which is characterized in that a word segmentation module, a word vector module, a similarity calculation module and a threshold comparison module are respectively designed in a summary mode, wherein a FastText model is constructed in the word vector model, a sentence semantic similarity calculation method with multi-feature fusion is provided in the aspect of similarity calculation, an analytic hierarchy process and a semantic distance are respectively applied to the calculation of structural similarity and semantic similarity, the analytic hierarchy process is used for calculating the word shape of a sentence, the word sequence and the corresponding weight of the sentence length characteristic value, and the shortest path algorithm is used for calculating the semantic distance in a weighted ontology graph, so that the semantic similarity is described. And weighting and fusing the result similarity and the semantic similarity to calculate the sentence similarity.
As shown in fig. 1, the intelligent question-answer optimization method based on natural language processing and deep learning of the present invention includes the following steps:
(10) chinese word segmentation and word vector conversion: capturing semantic information of a specific target field, and converting word characters in natural language into N-dimensional matrix vectors which can be understood by a computer;
the step of (10) converting Chinese word segmentation and word vectors comprises:
(11) chinese word segmentation: chinese word segmentation is realized by adopting an algorithm tool HanLP;
the following contents are added to the pom of the project catalog by the word construction and the dictionary segmentation through the HMM-Bigram of the HanLP mode
Fig. 2 shows a word vector training flowchart.
(12) And (3) word vector conversion: the method adopts a training word vector model based on the Huffman tree to convert word characters in natural language into N-dimensional matrix vectors which can be understood by a computer.
The method comprises the steps of converting word characters in natural language into matrix vectors which can be understood by a computer, creating a training word vector model based on a Huffman tree, and adopting the method of implicit mapping in a shallow neural network to measure and average input vectors instead of adopting linear change. A huffman tree is created as shown in fig. 1, specifying a walk from the left sub-tree as a negative class and a walk from the right sub-tree as a positive class. The probability of registering a positive value is
Taking the probability of a negative value as
P(-)=1-P(+),
Where w1, w2, w3 … … wi respectively specify respective internal node word vectors of the Huffman tree, θ represents model parameters trained from a training data set, and the likelihood function is
And if the window size of the input layer is 2c, inputting word vectors of c words in front of the current word and c words behind the current word. For the mapping of the input layer to the hidden layer, the 2c words may be averaged. Namely, it is
Then only the argmax is passedθΠc∈C(w)And P (w | c; theta) gradient ascending training and updating the word vector and the model parameters, wherein C (w) represents the context of the current word w, and theta represents the model parameters to be trained.
(20) And (3) natural language processing: giving weight to the relation between the ontology semantic concepts, calculating semantic distance by using the weight, and describing semantic similarity based on the semantic distance between the concepts;
as shown in fig. 3, the (20) natural language processing step includes:
(21) and (3) calculating the similarity of the word shapes: the similarity of the word shapes is calculated according to the following formula,
where Com (S1, S2) is the result of the participling of sentences S1 and S2, if a certain feature item appears once in sentences S1 and S2, the minimum value of the number of occurrences is taken as the value of Com (S1, S2), Len () represents the length of the sentence;
(22) calculating word order similarity: assuming that the words appearing only Once in S1 and S2 are defined as Once (S1, S2), and S represents the number of words in the Once (S1, S2) set, the word order similarity calculation formula representing the relative positions of the keywords in the two sentences is as follows
Where AIN (S1, S2, S) is defined to represent the inverse number of words in the sentence S2, in Morsmim (S1, S2) and OrdSim (S1, S2).
(23) Calculating sentence length similarity: the sentence length similarity LenSim (S1, S2) is calculated as follows:
(24) calculating the weight ratio of the word shape, the word sequence and the sentence length: construct each level judgment matrix as follows
And calculating the corresponding eigenvector of the hierarchical matrix as a similarity weight.
As shown in fig. 4, the (24) weight comparing step includes:
(241) building a structural model: the similarity of the sentence structure is obtained by integrating the similarity of the word shape, the similarity of the word sequence and the similarity of the sentence length
StrSim(S1,S2)=
α×MorSim(S1,S2)+β×OrdSim(S1,S2)+γ×LenSim(S1,S2),
Wherein alpha, beta and gamma respectively represent weight values of word form similarity, word order similarity and sentence length similarity and satisfy alpha + beta + gamma as 1, and each weight system in the structural similarity is calculated through AHP, and a hierarchical structure model is established
(242) Calculating a similarity value weight: constructing decision matrices in each level
The maximum eigenvalue λ max of the matrix is calculated to be 3, and the corresponding eigenvector is p ═ 5, 5, 1]τDefining a consistency indexWhere λ max is the maximum eigenvalue and n is the dimension of the eigenvector. According toMatrix a calculates the available CI to be 0, and finds the average random consistency index RI to be 0.52. Calculating a consistency ratioGet CR equal to 0, according to the test coefficient standard, when CR is<0.1, the consistency of the judgment matrix is acceptable, so the eigenvector corresponding to the judgment matrix A can be used as the similarity weight.
(243) Calculating the weight: for the above matrix eigenvector P ═ 5, 5, 1]τThe weight vector obtained by normalization is W ═ 0.455, 0.455, 0.09]I.e., α -0.455, β -0.455, and γ -0.09.
As shown in fig. 5, the step of (25) calculating the semantic similarity includes:
(25) calculating semantic similarity: semantic similarity of concept C1 and concept C2 was calculated using the following formula
Where concept Ci represents the set of nodes that are the shortest path between a given node and the following node, and SemSim (C1, C2) represents the semantic distance between concepts C1 and C2.
The step of (25) calculating semantic similarity comprises
(251) Establishing a semantic similarity model: the semantic similarity is calculated as a formula
In the above formula, m, n are two directly connected concept nodes in the ontology, depth (m) represents the maximum depth of the node m, n (g) represents the total number of concept nodes, order (n) represents the sequential number of the concept node n in the brother concept node, and depth (m) is defined as follows
(252) Weight initialization and recursive computation: and (4) regarding the ontology graph as a weighted directed graph, and calculating the shortest path from the concept node to the root node by using the shortest path algorithm idea. By the formula
Initializing the weight of the concept node relation, assigning the initial node to be 0, setting other nodes to be infinite, and then using a formula
In the formula, m, n and x represent three nodes, x and n are direct-connected nodes, S is a set of all nodes in the body, and Wk(m, n) represents the weight of the path (m, n) at iteration k.
(253) Calculating a semantic distance: in the calculation process of the semantic distance, the common part in the shortest path is deleted, and the semantic similarity between the two concepts is calculated
SemDis(C1,C2)=WshrotestPath1+WshrotestPath2-2×WcomShortestPath
Wherein, C1 and C2 represent two concept nodes in the ontology graph, and shortestPathi represents the shortest path between the concept node Ci and the node. comShortestPath represents the shortest path from the first public node between C1 and C2 to the following node
(254) Calculating the shortest path: the calculation method is as follows:
wherein m, n represents two nodes directly connected in shortestPath, and k is a set of node relations.
(255) Calculating semantic similarity: the formula defining the semantic similarity of concept C1 and concept C2 is
Where concept Ci represents the set of nodes that are the shortest path between a given node and the following node, and SemSim (C1, C2) represents the semantic distance between concepts C1 and C2.
(30) Language disambiguation: and eliminating language ambiguity through threshold comparison.
The (30) language disambiguation step comprising:
(31) preprocessing a question and answer: preloading a knowledge base, selecting 100 problems from the knowledge base, sequentially putting the prepared problems into a system, and sequentially acquiring rules to acquire a threshold value
(32) And (3) threshold comparison: and after the threshold value is obtained, threshold value comparison can be carried out, whether the answer to the question is returned or not is determined through the threshold value comparison, if the similarity is greater than the threshold value, the answer to the question is returned as output, and if the similarity is less than the threshold value, the answer to the question is returned to be empty, which indicates that no matched answer is found.
Aiming at the problems of information redundancy and question-answer inaccuracy in the traditional intelligent question-answer, the invention realizes the intelligent question-answer method through word vector conversion and the improved multi-semantic similarity comparison algorithm, and improves the accuracy and precision of the intelligent question-answer.
Claims (6)
1. An intelligent question-answer optimization method based on natural language processing and deep learning is characterized by comprising the following steps:
(10) chinese word segmentation and word vector conversion: capturing semantic information of a specific target field, and converting word characters in natural language into N-dimensional matrix vectors which can be understood by a computer;
(20) and (3) natural language processing: giving weight to the relation between the ontology semantic concepts, calculating semantic distance by using the weight, and describing semantic similarity based on the semantic distance between the concepts;
(30) language disambiguation: and eliminating language ambiguity through threshold comparison.
2. The intelligent question-answer optimization method according to claim 1, wherein the (10) chinese participle and word vector conversion step comprises:
(11) chinese word segmentation: chinese word segmentation is realized by adopting an algorithm tool HanLP;
(12) and (3) word vector conversion: the method adopts a training word vector model based on the Huffman tree to convert word characters in natural language into N-dimensional matrix vectors which can be understood by a computer.
3. The intelligent question-answer optimization method according to claim 1, characterized in that said (20) natural language processing step comprises:
(21) and (3) calculating the similarity of the word shapes: the similarity of the word shapes is calculated according to the following formula,
where Com (S1, S2) is the result of the participling of sentences S1 and S2, if a certain feature item appears once in sentences S1 and S2, the minimum value of the number of occurrences is taken as the value of Com (S1, S2), Len () represents the length of the sentence;
(22) calculating word order similarity: assuming that the words appearing only Once in S1 and S2 are defined as Once (S1, S2), and S represents the number of words in the Once (S1, S2) set, the word order similarity calculation formula representing the relative positions of the keywords in the two sentences is as follows
Where AIN (S1, S2, S) is defined to represent the inverse number of words in the sentence S2, in Morsmim (S1, S2) and OrdSim (S1, S2).
(23) Calculating sentence length similarity: the sentence length similarity LenSim (S1, S2) is calculated as follows:
(24) calculating the weight ratio of the word shape, the word sequence and the sentence length: construct each level judgment matrix as follows
And calculating the corresponding eigenvector of the hierarchical matrix as a similarity weight.
(25) Calculating semantic similarity: semantic similarity of concept C1 and concept C2 was calculated using the following formula
Where concept Ci represents the set of nodes that are the shortest path between a given node and the following node, and SemSim (C1, C2) represents the semantic distance between concepts C1 and C2.
4. The intelligent question-answer optimization method according to claim 3, characterized in that said (24) weight comparison step comprises
(241) Building a structural model: the similarity of the sentence structure is obtained by integrating the similarity of the word shape, the similarity of the word sequence and the similarity of the sentence length
StrSim(S1,S2)=
α×MorSim(S1,S2)+β×OrdSim(S1,S2)+γ×LenSim(S1,S2),
Wherein alpha, beta and gamma respectively represent weight values of word form similarity, word order similarity and sentence length similarity and satisfy alpha + beta + gamma as 1, and each weight system in the structural similarity is calculated through AHP, and a hierarchical structure model is established
(242) Calculating a similarity value weight: constructing decision matrices in each level
The maximum eigenvalue λ max of the matrix is calculated to be 3, and the corresponding eigenvector is p ═ 5, 5, 1]τDefining a consistency indexWhere λ max is the maximum eigenvalue and n is the dimension of the eigenvector. Calculating CI (0) according to the matrix A, and searching average random consistencyThe index RI is 0.52. Calculating a consistency ratioAnd obtaining CR equal to 0, and judging that the consistency of the matrix is acceptable when CR is less than 0.1 according to a check coefficient standard, so that the feature vector corresponding to the judgment matrix A can be used as a similarity weight.
(243) Calculating the weight: for the above matrix eigenvector p ═ 5, 5, 1]τThe weight vector obtained by normalization is W ═ 0.455, 0.455, 0.09]I.e., α -0.455, β -0.455, and γ -0.09.
5. The intelligent question-answer optimization method according to claim 4, characterized in that said step (25) of calculating semantic similarities comprises
(251) Establishing a semantic similarity model: the semantic similarity is calculated as a formula
In the above formula, m, n are two directly connected concept nodes in the ontology, depth (m) represents the maximum depth of the node m, n (g) represents the total number of concept nodes, order (n) represents the sequential number of the concept node n in the brother concept node, and depth (m) is defined as follows
(252) Weight initialization and recursive computation: and (4) regarding the ontology graph as a weighted directed graph, and calculating the shortest path from the concept node to the root node by using the shortest path algorithm idea. By the formula
Initializing the weight of the concept node relation, assigning the initial node to be 0, setting other nodes to be infinite, and then using a formula
In the formula, m, n and x represent three nodes, x and n are direct-connected nodes, S is a set of all nodes in the body, and Wk(m, n) represents the weight of the path (m, n) at iteration k.
(253) Calculating a semantic distance: in the calculation process of the semantic distance, the common part in the shortest path is deleted, and the semantic similarity between the two concepts is calculated
SemDis(C1,C2)=WshrotestPath1+Wshrotestpath2-2×WcomShortestPath
Wherein, C1 and C2 represent two concept nodes in the ontology graph, and shortestPathi represents the shortest path between the concept node Ci and the node. comShortestPath represents the shortest path from the first public node between C1 and C2 to the following node
(254) Calculating the shortest path: the calculation method is as follows:
wherein m, n represents two nodes directly connected in shortestPath, and k is a set of node relations.
(255) Calculating semantic similarity: the formula defining the semantic similarity of concept C1 and concept C2 is
Where concept Ci represents the set of nodes that are the shortest path between a given node and the following node, and SemSim (C1, C2) represents the semantic distance between concepts C1 and C2.
6. The intelligent question-answer optimization method according to claim 5, characterized in that said (30) language disambiguation step comprises:
(31) preprocessing a question and answer: preloading a knowledge base, selecting 100 problems from the knowledge base, sequentially putting the prepared problems into a system, and sequentially acquiring rules to acquire a threshold value
(32) And (3) threshold comparison: and after the threshold value is obtained, threshold value comparison can be carried out, whether the answer to the question is returned or not is determined through the threshold value comparison, if the similarity is greater than the threshold value, the answer to the question is returned as output, and if the similarity is less than the threshold value, the answer to the question is returned to be empty, which indicates that no matched answer is found.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010364914.1A CN111639165A (en) | 2020-04-30 | 2020-04-30 | Intelligent question-answer optimization method based on natural language processing and deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010364914.1A CN111639165A (en) | 2020-04-30 | 2020-04-30 | Intelligent question-answer optimization method based on natural language processing and deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111639165A true CN111639165A (en) | 2020-09-08 |
Family
ID=72329013
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010364914.1A Withdrawn CN111639165A (en) | 2020-04-30 | 2020-04-30 | Intelligent question-answer optimization method based on natural language processing and deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111639165A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112308387A (en) * | 2020-10-20 | 2021-02-02 | 深圳思为科技有限公司 | Client intention degree evaluation method and device and cloud server |
CN113255345A (en) * | 2021-06-10 | 2021-08-13 | 腾讯科技(深圳)有限公司 | Semantic recognition method, related device and equipment |
CN113722452A (en) * | 2021-07-16 | 2021-11-30 | 上海通办信息服务有限公司 | Semantic-based quick knowledge hit method and device in question-answering system |
CN113742458A (en) * | 2021-09-18 | 2021-12-03 | 苏州大学 | Natural language instruction disambiguation method and system for mechanical arm grabbing |
CN115828930A (en) * | 2023-01-06 | 2023-03-21 | 山东建筑大学 | Distributed word vector space correction method for dynamically fusing semantic relations |
-
2020
- 2020-04-30 CN CN202010364914.1A patent/CN111639165A/en not_active Withdrawn
Non-Patent Citations (1)
Title |
---|
李兆兆: "基于语义理解的智能问答系统关键技术研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112308387A (en) * | 2020-10-20 | 2021-02-02 | 深圳思为科技有限公司 | Client intention degree evaluation method and device and cloud server |
CN112308387B (en) * | 2020-10-20 | 2024-05-14 | 深圳思为科技有限公司 | Customer intention evaluation method and device and cloud server |
CN113255345A (en) * | 2021-06-10 | 2021-08-13 | 腾讯科技(深圳)有限公司 | Semantic recognition method, related device and equipment |
CN113255345B (en) * | 2021-06-10 | 2021-10-15 | 腾讯科技(深圳)有限公司 | Semantic recognition method, related device and equipment |
CN113722452A (en) * | 2021-07-16 | 2021-11-30 | 上海通办信息服务有限公司 | Semantic-based quick knowledge hit method and device in question-answering system |
CN113722452B (en) * | 2021-07-16 | 2024-01-19 | 上海通办信息服务有限公司 | Semantic-based rapid knowledge hit method and device in question-answering system |
CN113742458A (en) * | 2021-09-18 | 2021-12-03 | 苏州大学 | Natural language instruction disambiguation method and system for mechanical arm grabbing |
CN113742458B (en) * | 2021-09-18 | 2023-04-25 | 苏州大学 | Natural language instruction disambiguation method and system oriented to mechanical arm grabbing |
CN115828930A (en) * | 2023-01-06 | 2023-03-21 | 山东建筑大学 | Distributed word vector space correction method for dynamically fusing semantic relations |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109271505B (en) | Question-answering system implementation method based on question-answer pairs | |
CN109408627B (en) | Question-answering method and system fusing convolutional neural network and cyclic neural network | |
CN108804521B (en) | Knowledge graph-based question-answering method and agricultural encyclopedia question-answering system | |
CN111639165A (en) | Intelligent question-answer optimization method based on natural language processing and deep learning | |
CN110309268B (en) | Cross-language information retrieval method based on concept graph | |
CN111522910B (en) | Intelligent semantic retrieval method based on cultural relic knowledge graph | |
CN111737496A (en) | Power equipment fault knowledge map construction method | |
CN113239700A (en) | Text semantic matching device, system, method and storage medium for improving BERT | |
CN110674252A (en) | High-precision semantic search system for judicial domain | |
CN106372187B (en) | Cross-language retrieval method for big data | |
CN109783806B (en) | Text matching method utilizing semantic parsing structure | |
CN110765755A (en) | Semantic similarity feature extraction method based on double selection gates | |
CN112163425A (en) | Text entity relation extraction method based on multi-feature information enhancement | |
CN113377897B (en) | Multi-language medical term standard standardization system and method based on deep confrontation learning | |
CN112597285B (en) | Man-machine interaction method and system based on knowledge graph | |
CN106446162A (en) | Orient field self body intelligence library article search method | |
CN114912449B (en) | Technical feature keyword extraction method and system based on code description text | |
Zhang et al. | Hierarchical scene parsing by weakly supervised learning with image descriptions | |
CN112036178A (en) | Distribution network entity related semantic search method | |
CN114332519A (en) | Image description generation method based on external triple and abstract relation | |
CN113326267A (en) | Address matching method based on inverted index and neural network algorithm | |
CN115422323A (en) | Intelligent intelligence question-answering method based on knowledge graph | |
CN114004236B (en) | Cross-language news event retrieval method integrating knowledge of event entity | |
CN114841353A (en) | Quantum language model modeling system fusing syntactic information and application thereof | |
CN112417170B (en) | Relationship linking method for incomplete knowledge graph |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20200908 |