CN111339269B - Knowledge graph question-answering training and application service system capable of automatically generating templates - Google Patents
Knowledge graph question-answering training and application service system capable of automatically generating templates Download PDFInfo
- Publication number
- CN111339269B CN111339269B CN202010104143.2A CN202010104143A CN111339269B CN 111339269 B CN111339269 B CN 111339269B CN 202010104143 A CN202010104143 A CN 202010104143A CN 111339269 B CN111339269 B CN 111339269B
- Authority
- CN
- China
- Prior art keywords
- question
- query
- backbone
- module
- template
- 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.)
- Active
Links
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
- 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
- G06F16/35—Clustering; Classification
-
- 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The application discloses a knowledge graph question-answering training system for automatically generating templates, which comprises the following steps: the building module of the predicate dictionary and the category dictionary is used for building the predicate dictionary and the category dictionary respectively by using a remote supervision mode; the backbone query generation module is used for acquiring the sub-graph of the subject entity and the answer entity of each training question answer pair in the knowledge graph, and using the variable to replace the answer node in the sub-graph to form backbone query; a semantic alignment module; for aligning question phrases with backbone query semantic elements using dependency syntax analysis and shaping linear alignment techniques; the template flooding module is used for storing the dependency syntax tree, the backbone query and the corresponding relation as templates into a template library; and the sequencing model training module is used for classifying and learning every two matching templates according to the matching degree by using a machine learning two-classifier to obtain a question template sequencing model, so that the problems of high labor cost and low problem coverage rate in the prior art are solved.
Description
Technical Field
The application relates to the field of intelligent application, in particular to a knowledge graph question-answering training system for automatically generating templates and a knowledge graph question-answering application service system for automatically generating templates.
Background
The method based on the question-answering template plays an important role in knowledge graph question-answering, and performs semantic feature extraction on the user natural language question by using the modes of word segmentation, named entity recognition, predicate detection, category detection, question type classification, entity link and the like, and matches the obtained semantic features with question templates in a template library through similarity or a sorting algorithm. After the template matching is successful, the query template (usually the SPARQL query sentence) is instantiated by using the information of entities, categories and the like in the natural language question sentence, and then knowledge query is executed and a result is returned.
The knowledge graph question-answering method based on the question-answering template not only can track the whole question-answering process more clearly, but also can realize the question-answering of complex problems, but the traditional knowledge graph question-answering based on the template has the following two problems:
1. relying on manual template making requires significant labor costs.
2. It is difficult to ensure coverage of the problem.
Disclosure of Invention
The application provides a knowledge graph question-answering training and application service system capable of automatically generating templates, which solves the problems of high labor cost and low problem coverage rate in the prior art.
The application provides a knowledge graph question-answering training system for automatically generating templates, which is characterized by comprising the following steps:
the building module of the predicate dictionary and the category dictionary is used for building the predicate dictionary and the category dictionary respectively by using a remote supervision mode;
the backbone query generation module is used for acquiring the sub-graph of the subject entity and the answer entity of each training question answer pair in the knowledge graph, and using the variable to replace the answer node in the sub-graph to form the backbone query module;
the dependency syntax analysis and semantic role alignment module is used for analyzing sentences into a dependency syntax tree and describing dependency relations among all words; the semantic role alignment module is used for mapping the phrases in the question to the entities, the relations or the categories mentioned in the backbone query to form corresponding relations.
The template flooding module is used for removing the dependency tree, the backbone query and the corresponding relation as templates after the dependency tree nodes and the backbone query semantic elements which are not mapped after the semantic roles are aligned are removed according to the corresponding relation among the dependency syntax tree, the backbone query and the question elements and the backbone query elements;
and the sequencing model training module is used for performing classification learning on every two matching templates by using a machine learning two-classifier according to the matching degree to obtain a question template sequencing model.
Preferably, the predicate dictionary is constructed by using a remote supervision mode, including:
for the relation p in the knowledge graph, representing all the triples related to p in the knowledge graph by using C (p) = { (s, o): (s, p, o) ∈K }, wherein K represents the knowledge graph;
if the s and o entities in C (p) are detected in the same sentence of natural language description at the same time, extracting the intermediate language description r of the two entities in the sentence of text;
assuming that (s, p, o) is a triplet in the knowledge-graph, then r represents p, and the mapping (r→p) is added to the predicate dictionary L p In (a) and (b);
taking the quotient of the number of times the mapping appears and the sum of the number of times all the relations in the corpus are detected as the weight of the mapping.
Preferably, the method for constructing the category dictionary by using a remote supervision mode comprises the following steps:
for the category C in the knowledge graph, all entities of the category C in the knowledge graph are represented by C (C) = { e (e type C) ∈K };
the system searches on the corpus, if an entity or other noun phrases are detected, mapping (np-c) condition identification dictionary libraries;
taking the quotient of the number of times the mapping appears and the sum of the number of times all the relations in the corpus are detected as the weight of the mapping.
Preferably, the backbone query generation module is configured to obtain a topic entity of each training question answer pair and a sub-graph of the answer entity in the knowledge graph, replace answer nodes in the sub-graph with variables, and form a backbone query module, and includes:
for each training question-answer pair, detecting entity mention in the question sentence by using a named entity recognition technology;
detecting subject entities mentioned by the entities in the knowledge graph through entity links;
obtaining a sub-graph M of a subject entity and an answer entity in a question in a knowledge graph through a shortest path algorithm;
adding the type nodes of all answer nodes into the subgraph M;
and replacing answer nodes in the subgraph M by using variables to obtain a backbone query module in the form of SPARQL.
Preferably, the method further comprises: the dependency syntax analysis and semantic role alignment are used for obtaining the corresponding relation between the question phrase and the backbone query semantic element according to the dependency syntax tree and the shaping linear alignment, and the method comprises the following steps:
performing dependency syntax analysis on the question to obtain a question dependency syntax analysis tree;
acquiring question phrase arrangement combinations and backbone query semantic element combinations;
acquiring the phrase weight of the question by using the dictionary;
and aligning the question phrase and the backbone query semantic element according to the shaping linear alignment.
Preferably, the template flooding module is configured to remove nodes that are not mapped after semantic roles are aligned on the dependency tree of a question according to a correspondence between the dependency syntax tree, the backbone query, the question element, and the backbone query element, and store the dependency syntax tree, the backbone query, and the correspondence as templates in a template library after removing class nodes that are not mentioned in the correspondence in the backbone query, and includes:
according to the corresponding relation, removing the nodes which are not mapped on the question dependency tree after semantic roles are aligned, replacing specific vocabularies on the dependency syntax tree with semantic annotations according to the mapping provided by the corresponding relation, and retaining part-of-speech information of the vocabularies and side information of the tree;
removing class nodes which are not mentioned in the corresponding relation in the backbone query according to the corresponding relation, and replacing semantic element information with semantic annotation;
and storing the processed dependency syntax tree, backbone query and the corresponding relation as templates into a template library.
Preferably, the ranking model training module is configured to use a machine learning two-classifier to learn classification of every two matching templates according to the matching degree, and obtain a question template ranking model, and includes:
and acquiring training features, semantic character alignment features, semantic features and template features.
Data training is performed using a machine learning model.
The application also provides a knowledge graph question-answering application service system for automatically generating the template, which comprises the following steps:
the template matching module is used for generating a question dependency syntax tree for a new question according to dependency syntax analysis, and if the template in the template library is a subtree of the question dependency syntax tree, the template matching is successful;
the template instantiation unit is used for carrying out template instantiation according to a given question and a template set successfully matched;
the template ordering module is used for carrying out ordering prediction of templates by using an ordering model with the obtained question query pair as input data, and taking the query template with the highest score as the optimal template;
and the template query module is used for carrying out data query of the knowledge graph on the obtained optimal instantiation query statement and returning an answer.
Preferably, the template matching module is configured to generate a question dependency syntax tree for a new question according to dependency syntax analysis, and if a template in a template library is a subtree of the question dependency syntax tree, the template matching is successful, including:
generating a question dependency syntax tree for the new question according to the dependency syntax analysis;
if the templates in the template library are subtrees of the question dependency syntax tree under the condition of considering only part of speech and side information, the template matching is successful, and the templates are candidate templates of the question.
The application also provides a knowledge graph question-answering system for automatically generating the template, which comprises a knowledge graph question-answering training system for automatically generating the template and a knowledge graph question-answering application service system for automatically generating the template.
The application provides a knowledge graph question-answering training and application service system for automatically generating templates, which is used for automatically learning a question template and a query template through a simple question-answering pair related to a specific knowledge graph based on dependency syntax analysis, and simultaneously utilizing the combination characteristic of dependency syntax analysis results and the template learned from simple questions to answer complex questions, thereby capturing complete complex questions without any specific template and solving the problems of high labor cost and low question coverage rate in the prior art.
Drawings
Fig. 1 is a schematic diagram of a knowledge graph question-answering training and application service system for automatically generating templates.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may be embodied in many other forms than those herein described, and those skilled in the art will readily appreciate that the present application may be similarly embodied without departing from the spirit or essential characteristics thereof, and therefore the present application is not limited to the specific embodiments disclosed below.
In the architecture schematic diagram of the knowledge graph question-answering training and application service system provided in fig. 1, which is automatically generated by a template, the training system includes: the system comprises a predicate dictionary and category dictionary construction module, a backbone query generation module, a dependency syntax analysis and semantic role alignment module, a template bloom module and a sequencing model training module.
And the predicate dictionary and category dictionary building module is used for building the predicate dictionary and the category dictionary respectively by using a remote supervision mode. Because of the diversified natural language representations of the relations or the categories in the knowledge graph, all natural language description forms of the relations or the categories cannot be collected in a manual enumeration mode. The system uses a remote supervision mode to respectively construct a predicate dictionary (relationship dictionary) and a category dictionary.
Predicate dictionary L p : for the relation p in the knowledge graph, all three related to p in the knowledge graph are represented by C (p) = { (s, o): (s, p, o) ∈K }A tuple, wherein K represents a knowledge graph; the system searches on the corpus, and if two entities s and o in C (p) are detected in the same sentence natural language description at the same time, extracting two entity intermediate language descriptions r in the sentence text; assuming that (s, p, o) is a triplet in the knowledge-graph, then r represents p, and the mapping (r→p) is added to the predicate dictionary L p In (a) and (b); since the assumption of remote supervision is not always true, the system takes as the weight of the map the quotient of the number of occurrences of the map and the sum of the number of times all relationships in the corpus are detected.
Category dictionary L c : for the category C in the knowledge graph, all entities of the category C in the knowledge graph are represented by C (C) = { e (e type C) ∈K }; the system searches on the corpus, and if entities or other noun phrases, such as ' e ' and other np ' are detected (where np represents a part-of-speech phrase), the condition dictionary library L is mapped (np→c) c The method comprises the steps of carrying out a first treatment on the surface of the The calculation method of each entry weight is identical to that of predicate dictionary L p 。
And the backbone query generation module is used for acquiring the sub-graphs of the subject entity and the answer entity of each training question answer pair in the knowledge graph, and using the variables to replace answer nodes in the sub-graphs to form the backbone query module. For each training question-answer pair, firstly, detecting the entity mention in the question sentence by using Named Entity Recognition (NER) and other technologies, and then detecting the topic entity mentioned about the entity in the knowledge graph through entity link. And obtaining the sub-graph M of the subject entity and the answer entity in the question in the knowledge graph through a shortest path algorithm. Since the types of answer entities play a key role in knowledge-graph questions and answers, the system adds the type nodes of all answer nodes to the subgraph M. The answer nodes in sub-graph M are replaced by variables to obtain backbone queries in SPARQL form.
The dependency syntax analysis and semantic role alignment module is used for analyzing sentences into a dependency syntax tree and describing dependency relationships among the words, namely indicating syntactic collocation relationships among the words, wherein the collocation relationships are related to semantics. Both machine learning-based and neural network-based dependency syntax analysis are applicable to the system; the semantic role alignment module is used to map phrases in question sentences to the mentioned entities, relationships, or categories in the backbone query. The purpose of semantic role alignment is to map phrases in question sentences to the mentioned entities, relationships, or categories in the backbone query. This alignment will determine which phrases in the question can be mapped into the backbone query, which answer type constraints in the backbone query will be preserved, and the mapping between the question phrases and the backbone query semantic elements. The dependency syntax analysis and semantic role alignment are used for obtaining the corresponding relation between the question phrase and the backbone query semantic element according to the dependency syntax tree and the shaping linear alignment, and the method comprises the following steps: performing dependency syntax analysis on the question to obtain a question dependency syntax analysis tree; acquiring question phrase arrangement combinations and backbone query semantic element combinations; acquiring the phrase weight of the question by using the dictionary; and aligning the question phrase and the backbone query semantic element according to the shaping linear alignment.
First, phrase set ph= { Ph is formed by using all sub-phrases in question 1 ,ph 2 ,...,ph i Definition of all element sets S in backbone queries q . From predicate dictionary L p And category dictionary L c Obtaining ph i E Ph to s j ∈S q Is not a function of the mapping (ph i →s j ) W is used as the weight of (2) ij Indicating that the weight is obtained from the dictionary, X ij Indicating whether the mapping remains valued at 0 or 1. The goal of semantic role alignment is to maximize:
the constraints are:
wherein equation (2) represents that each backbone query semantic term is aligned by at most one phrase.
Equation (3) represents that once each phrase is aligned with an entity, any vocabulary in the phrase cannot be aligned with other backbone query elements, E j The representative element j is an entity.
Equation (4) represents that the answer type constraint is at most intelligently preserved, and S (v) represents the type element in the backbone query.
Through the formula, the phrase elements in the question and the semantic elements in the backbone query can be acquired by using an integer linear programming means to perform optimal alignment.
The template ubiquity module is used for carrying out the following processing on the components in order to enable the template to have the ubiquity capability according to the corresponding relation mt among the dependency syntax tree, the backbone query, the question phrase and the backbone query semantic elements:
A. nodes which are not mapped after semantic roles are aligned on the question dependency tree are removed, specific vocabularies on the dependency syntax tree are replaced by semantic notes (entities, relations and categories) according to the mapping provided by mt, and part-of-speech information of the vocabularies and side information of the tree are reserved.
B. And removing class nodes which are not mentioned in the corresponding relation mt in the backbone query, and replacing semantic element information with semantic notes.
C. And storing the processed dependency syntax tree, backbone query and the corresponding relation as templates into a template library.
And the sequencing model training module is used for performing classification learning on every two matching templates by using a machine learning two-classifier according to the matching degree to obtain a question template sequencing model. Training features include semantic character alignment features, semantic features, template features, and the like. Algorithms such as random forests, support vector machines, neural networks and the like are suitable for the system.
In a knowledge graph question-answering training system automatically generated by a template, after training is completed, a new question is received and is responded, so that the response to the new question is completed by an application service system based on the knowledge graph question-answering training system automatically generated by the template.
As shown in fig. 1, a knowledge-graph question-answering application service system for automatically generating a template includes: the system comprises a template matching module, a template instantiation module, a template ordering module and a template query module.
The template matching module is used for analyzing the new question according to the dependency syntax to generate a question dependency syntax tree; if the templates in the template library are subtrees of the question dependency syntax tree under the condition of considering only part of speech and side information, the template matching is successful, and the templates are candidate templates of the question.
The template instantiation module is used for carrying out template instantiation according to a given question and a template set successfully matched; the instantiation is specifically to instantiate a query sentence according to a mapping rule mt and a dictionary in a template, and replace semantic annotations (entities, relations and categories) in the query template in the template with specific semantic elements. The template ordering unit is used for carrying out ordering prediction of templates by using the ordering model by taking the obtained question query pair as input data, and taking the query template with the highest score as the optimal template;
and the template query module is used for carrying out data query of the knowledge graph on the obtained optimal instantiation query statement and returning an answer.
The knowledge graph question-answering training system and the application service system which are automatically generated by the template form the knowledge graph question-answering system which is automatically generated by the template.
The application provides a knowledge graph question-answering training and application service system for automatically generating templates, which is used for automatically learning a question template and a query template through a simple question-answering pair related to a specific knowledge graph based on dependency syntax analysis, and simultaneously utilizing the combination characteristic of dependency syntax analysis results and the template learned from simple questions to answer complex questions, thereby capturing complete complex questions without any specific template and solving the problems of high labor cost and low question coverage rate in the prior art.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present application and not for limiting the same, and although the present application has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made to the specific embodiments of the present application without departing from the spirit and scope of the present application, and all modifications and equivalents are intended to be included in the scope of the claims of the present application.
Claims (5)
1. A knowledge graph question-answering training system for automatically generating templates is characterized by comprising the following components:
the building module of predicate dictionary and category dictionary is used for building predicate dictionary and category dictionary respectively by using remote supervision mode, and comprises:
for the relation p in the knowledge graph, representing all the triples related to p in the knowledge graph by using C (p) = { (s, o): (s, p, o) ∈K }, wherein K represents the knowledge graph;
if the s and o entities in C (p) are detected in the same sentence of natural language description at the same time, extracting the intermediate language description r of the two entities in the sentence of text;
assuming that (s, p, o) is a triplet in the knowledge-graph, then r represents p, and the mapping (r→p) is added to the predicate dictionary L p In (a) and (b);
taking the quotient of the number of times of occurrence of the mapping and the sum of the number of times of detection of all the relations in the corpus as the weight of the mapping;
for each training question-answer pair, detecting entity mention in the question sentence by using a named entity recognition technology;
detecting subject entities mentioned by the entities in the knowledge graph through entity links;
obtaining a sub-graph M of a subject entity and an answer entity in a question in a knowledge graph through a shortest path algorithm;
adding the type nodes of all answer nodes into the subgraph M;
replacing answer nodes in the subgraph M by using variables to obtain a backbone query module in a SPARQL form;
the backbone query generation module is used for acquiring the sub-graph of the subject entity and the answer entity of each training question answer pair in the knowledge graph, and using the variable to replace the answer node in the sub-graph to form the backbone query module;
the dependency syntax analysis and semantic role alignment module is used for analyzing sentences into a dependency syntax tree and describing dependency relations among all words; the semantic role alignment module is used for mapping the phrases in the question to the entities, the relations or the categories mentioned in the backbone query to form corresponding relations;
the template flooding module is used for removing the dependency tree, the backbone query and the corresponding relation as templates after the dependency tree nodes and the backbone query semantic elements which are not mapped after the semantic roles are aligned are removed according to the corresponding relation among the dependency syntax tree, the backbone query and the question elements and the backbone query elements;
and the sequencing model training module is used for performing classification learning on every two matching templates by using a machine learning two-classifier according to the matching degree to obtain a question template sequencing model.
2. The system of claim 1, wherein the backbone query generation module is configured to obtain a topic entity and a sub-graph of an answer entity in a knowledge graph for each training question-answer pair, and use a variable to replace an answer node in the sub-graph to form the backbone query module, and comprises:
for each training question-answer pair, detecting entity mention in the question sentence by using a named entity recognition technology;
detecting subject entities mentioned by the entities in the knowledge graph through entity links;
obtaining a sub-graph M of a subject entity and an answer entity in a question in a knowledge graph through a shortest path algorithm;
adding the type nodes of all answer nodes into the subgraph M;
and replacing answer nodes in the subgraph M by using variables to obtain a backbone query module in the form of SPARQL.
3. The system of claim 1, further comprising: the dependency syntax analysis and semantic role alignment are used for obtaining the corresponding relation between the question phrase and the backbone query semantic element according to the dependency syntax tree and the shaping linear alignment, and the method comprises the following steps:
performing dependency syntax analysis on the question to obtain a question dependency syntax analysis tree;
acquiring question phrase arrangement combinations and backbone query semantic element combinations;
acquiring the phrase weight of the question by using the dictionary;
and aligning the question phrase and the backbone query semantic element according to the shaping linear alignment.
4. The system of claim 1, wherein the template flooding module is configured to remove nodes on the dependency tree that are not mapped after semantic role alignment according to correspondence between the dependency tree, the backbone query, the question element, and the backbone query element, and to store the dependency tree, the backbone query, and the correspondence as templates in the template library after removing class nodes that are not mentioned in the correspondence in the backbone query, comprising:
according to the corresponding relation, removing the nodes which are not mapped on the question dependency tree after semantic roles are aligned, replacing specific vocabularies on the dependency syntax tree with semantic annotations according to the mapping provided by the corresponding relation, and retaining part-of-speech information of the vocabularies and side information of the tree;
removing class nodes which are not mentioned in the corresponding relation in the backbone query according to the corresponding relation, and replacing semantic element information with semantic annotation;
and storing the processed dependency syntax tree, backbone query and the corresponding relation as templates into a template library.
5. The system of claim 1, wherein the ranking model training module is configured to perform classification learning on each two matching templates according to the matching degree by using a machine learning two-classifier, and obtain the question template ranking model, and the method comprises:
acquiring training features, semantic role alignment features, semantic features and template features; data training is performed using a machine learning model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010104143.2A CN111339269B (en) | 2020-02-20 | 2020-02-20 | Knowledge graph question-answering training and application service system capable of automatically generating templates |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010104143.2A CN111339269B (en) | 2020-02-20 | 2020-02-20 | Knowledge graph question-answering training and application service system capable of automatically generating templates |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111339269A CN111339269A (en) | 2020-06-26 |
| CN111339269B true CN111339269B (en) | 2023-09-26 |
Family
ID=71185432
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010104143.2A Active CN111339269B (en) | 2020-02-20 | 2020-02-20 | Knowledge graph question-answering training and application service system capable of automatically generating templates |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111339269B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111897944B (en) * | 2020-08-17 | 2024-03-22 | 哈尔滨工业大学 | Knowledge graph question-answering system based on semantic space sharing |
| CN112632237A (en) * | 2020-12-07 | 2021-04-09 | 厦门渊亭信息科技有限公司 | Knowledge graph-based question-answer template automatic generation method and device |
| CN112765312B (en) * | 2020-12-31 | 2022-05-10 | 湖南大学 | Knowledge graph question-answering method and system based on graph neural network embedded matching |
| CN113066358B (en) * | 2021-04-14 | 2023-01-10 | 吴光银 | Science teaching auxiliary system |
| CN112989005B (en) * | 2021-04-16 | 2022-07-12 | 重庆中国三峡博物馆 | Knowledge graph common sense question-answering method and system based on staged query |
| CN114417807B (en) * | 2022-01-24 | 2023-09-22 | 中国电子科技集团公司第五十四研究所 | Human-like language description expression method for collaboration scene of presence or absence |
| CN114579710B (en) * | 2022-03-15 | 2023-04-25 | 西南交通大学 | Method for generating problem query template of high-speed train |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908042A (en) * | 2010-08-09 | 2010-12-08 | 中国科学院自动化研究所 | Tagging method of bilingual combination semantic role |
| CN105701253A (en) * | 2016-03-04 | 2016-06-22 | 南京大学 | Chinese natural language interrogative sentence semantization knowledge base automatic question-answering method |
| CN107451230A (en) * | 2017-07-24 | 2017-12-08 | 宗晖（上海）机器人有限公司 | A kind of answering method and question answering system |
| CN108363743A (en) * | 2018-01-24 | 2018-08-03 | 清华大学深圳研究生院 | A kind of intelligence questions generation method, device and computer readable storage medium |
| CN110147436A (en) * | 2019-03-18 | 2019-08-20 | 清华大学 | A kind of mixing automatic question-answering method based on padagogical knowledge map and text |
| CN110399457A (en) * | 2019-07-01 | 2019-11-01 | 吉林大学 | A kind of intelligent answer method and system |
| CN110502621A (en) * | 2019-07-03 | 2019-11-26 | 平安科技（深圳）有限公司 | Answering method, question and answer system, computer equipment and storage medium |
| CN110532358A (en) * | 2019-07-05 | 2019-12-03 | 东南大学 | A kind of template automatic generation method towards knowledge base question and answer |
| CN110807091A (en) * | 2019-03-01 | 2020-02-18 | 王涵 | Hotel intelligent question-answer recommendation and decision support analysis method and system |
-
2020
- 2020-02-20 CN CN202010104143.2A patent/CN111339269B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908042A (en) * | 2010-08-09 | 2010-12-08 | 中国科学院自动化研究所 | Tagging method of bilingual combination semantic role |
| CN105701253A (en) * | 2016-03-04 | 2016-06-22 | 南京大学 | Chinese natural language interrogative sentence semantization knowledge base automatic question-answering method |
| CN107451230A (en) * | 2017-07-24 | 2017-12-08 | 宗晖（上海）机器人有限公司 | A kind of answering method and question answering system |
| CN108363743A (en) * | 2018-01-24 | 2018-08-03 | 清华大学深圳研究生院 | A kind of intelligence questions generation method, device and computer readable storage medium |
| CN110807091A (en) * | 2019-03-01 | 2020-02-18 | 王涵 | Hotel intelligent question-answer recommendation and decision support analysis method and system |
| CN110147436A (en) * | 2019-03-18 | 2019-08-20 | 清华大学 | A kind of mixing automatic question-answering method based on padagogical knowledge map and text |
| CN110399457A (en) * | 2019-07-01 | 2019-11-01 | 吉林大学 | A kind of intelligent answer method and system |
| CN110502621A (en) * | 2019-07-03 | 2019-11-26 | 平安科技（深圳）有限公司 | Answering method, question and answer system, computer equipment and storage medium |
| CN110532358A (en) * | 2019-07-05 | 2019-12-03 | 东南大学 | A kind of template automatic generation method towards knowledge base question and answer |
Non-Patent Citations (2)
| Title |
|---|
| John DeNero.The Complexity of Phrase Alignment Problems.《Conference Paper》.2008,第1-4页. * |
| 阮翀.基于多译文的中文转述语料库建设及转述评价方案.《中文信息学报》.2018,第67-75页）. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111339269A (en) | 2020-06-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111339269B (en) | Knowledge graph question-answering training and application service system capable of automatically generating templates | |
| Arora et al. | Character level embedding with deep convolutional neural network for text normalization of unstructured data for Twitter sentiment analysis | |
| CN111259653B (en) | Knowledge graph question-answering method, system and terminal based on entity relationship disambiguation | |
| CN110825881B (en) | Method for establishing electric power knowledge graph | |
| CN109710744B (en) | Data matching method, device, equipment and storage medium | |
| CN112650840A (en) | Intelligent medical question-answering processing method and system based on knowledge graph reasoning | |
| CN110245238B (en) | Graph embedding method and system based on rule reasoning and syntax mode | |
| CN112328800A (en) | System and method for automatically generating programming specification question answers | |
| CN113168499A (en) | Method for searching patent document | |
| CN113196277A (en) | System for retrieving natural language documents | |
| CN113764112A (en) | Online medical question and answer method | |
| CN112149427B (en) | Verb phrase implication map construction method and related equipment | |
| CN113569023A (en) | Chinese medicine question-answering system and method based on knowledge graph | |
| CN113593661A (en) | Clinical term standardization method, device, electronic equipment and storage medium | |
| CN117033571A (en) | Knowledge question-answering system construction method and system | |
| CN111651569B (en) | Knowledge base question-answering method and system in electric power field | |
| CN112632250A (en) | Question and answer method and system under multi-document scene | |
| CN113535897A (en) | Fine-grained emotion analysis method based on syntactic relation and opinion word distribution | |
| Cabrio et al. | Abstract dialectical frameworks for text exploration | |
| CN116402066A (en) | Attribute-level text emotion joint extraction method and system for multi-network feature fusion | |
| KR101333485B1 (en) | Method for constructing named entities using online encyclopedia and apparatus for performing the same | |
| CN116127099A (en) | Combined text enhanced table entity and type annotation method based on graph rolling network | |
| CN112417170B (en) | Relationship linking method for incomplete knowledge graph | |
| Silva et al. | Xte: Explainable text entailment | |
| CN113742445B (en) | Text recognition sample obtaining method and device and text recognition method and device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |