CN113468311B - Knowledge graph-based complex question and answer method, device and storage medium - Google Patents
Knowledge graph-based complex question and answer method, device and storage medium Download PDFInfo
- Publication number
- CN113468311B CN113468311B CN202110819044.7A CN202110819044A CN113468311B CN 113468311 B CN113468311 B CN 113468311B CN 202110819044 A CN202110819044 A CN 202110819044A CN 113468311 B CN113468311 B CN 113468311B
- Authority
- CN
- China
- Prior art keywords
- question
- module
- matching
- information
- constraint
- 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
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a knowledge graph-based complex question and answer method, a device and a storage medium, wherein the method comprises the following steps: performing entity identification and entity linking on the input question; obtaining candidate key paths and knowledge graph information according to the subject entities of the questions; constructing a candidate critical path matching model, identifying a critical path of a question, and identifying a candidate constraint condition corresponding to the critical path according to the selected critical path; constructing a constraint condition screening model, and selecting a proper constraint condition; and generating a query graph of the question by combining the key path and the constraint condition, and searching out an answer in the knowledge graph. The invention can fully utilize the information of each aspect of the knowledge graph, extract the key semantic information of the complex question, realize the matching of the semantic information of the question and the knowledge graph, reduce the answer range, locate the question point of the question and improve the answer accuracy through automatic screening of constraint conditions.
Description
Technical Field
The invention relates to the technical field of natural language understanding, in particular to a knowledge graph-based complex question and answer method, a knowledge graph-based complex question and answer device and a storage medium.
Background
In recent years, intelligent question-answering robots are becoming one of important research directions of a plurality of large companies, such as microsoft ice, ali honey and the like, and questions and answers based on knowledge patterns are being widely paid attention to academia and industry as one of important parts, and are applied in various directions, such as financial fields, medical fields, educational fields and the like, in a manner of customer service, search engines and the like. In the actual application scene, the user is faced with the scene that the problems of the user are more complex, and the complex problems are also a technical difficulty to be solved by the current question-answering system.
In the related technology, the main stream method is divided into two main categories of semantic analysis and information extraction, wherein the semantic analysis method is to combine a knowledge graph structure, identify constraint conditions through a plurality of artificially customized rules, convert a question into a query graph and obtain an optimal answer through a ranking function; the information extraction method is that the question sentence and the related knowledge graph sub-graph are mapped into a low-dimensional space, and the answer is obtained by matching the vector similarity of the question sentence and the related knowledge graph sub-graph. CN111611806a proposes a semantic parsing method for knowledge graph questions and answers, in which different parts of questions and sentences are converted into different data nodes through a pre-built dictionary, and then converted into a query graph according to artificially defined rules. CN110111766a proposes a knowledge-graph common sense question-answering method based on phased query, in which the entity, constraint condition and question structure of a question are identified through models of different phases, and although the uniformity of the overall structure is satisfied, the semantic information of the question is not matched with each part of the query graph.
In summary, the existing methods have the following problems:
1. the existing method directly matches the question with the knowledge graph relationship, and does not use other information of the knowledge graph, so that the matching is inaccurate.
2. When the candidate query graphs are ranked by the existing method, a plurality of features are required to be defined manually, the query graphs are converted into feature vectors and then ranked by a model, and the ranking model only aims at the features of the query graphs and cannot correspond the query graphs to specific problems one by one.
3. In the constraint condition identification process, part of the method converts different constraint terms into the same feature vector, and the model cannot identify the different constraint terms.
Disclosure of Invention
The invention provides a knowledge graph-based complex question and answer method, a knowledge graph-based complex question and answer device and a storage medium, so as to solve the technical problems.
The technical scheme adopted by the invention is as follows: the method for providing the complex question and answer based on the knowledge graph comprises the following steps:
performing entity identification and entity linking on the input question;
obtaining candidate key paths and knowledge graph information according to the subject entities of the questions;
constructing a candidate key path matching model, and identifying a key path of a question;
identifying candidate constraint conditions corresponding to the selected critical paths according to the selected critical paths;
constructing a constraint condition screening model, and selecting a proper constraint condition;
and generating a query graph of the question by combining the key path and the constraint condition, and searching out an answer in the knowledge graph.
As a preferred mode of the complex question and answer method based on the knowledge graph, the method for carrying out entity identification and entity linking on the input question comprises the following steps:
the entity expression in the question is identified through the entity identification model, and is linked with the entity in the knowledge graph to obtain an entity set; the entity recognition model comprises a bi-directional recurrent neural network or a conditional random field.
As a preferred mode of the complex question and answer method based on the knowledge graph, the method for obtaining candidate key paths and knowledge graph information according to the subject entity of the question comprises the following steps:
sequentially selecting one entity in the entity set as a subject entity, and other entities as constraint entities in constraint conditions; and according to the subject entity, searching out candidate critical paths and information related to the critical paths in the knowledge graph.
As a preferred mode of the complex question and answer method based on the knowledge graph, the searched candidate key paths are all paths within two hops from the subject entity, and the related information comprises entity category information and answer category information.
As a preferred mode of the complex question and answer method based on the knowledge graph, the method for constructing the candidate key path matching model and identifying the key path of the question comprises the following steps:
constructing a first word vector module, a semantic matching module, an information matching module and a comprehensive matching module;
the constructing a first word vector module includes: performing fixed-length word vector processing on an input text;
the construction of the semantic matching module comprises the following steps: semantic information of the question and the candidate key path is respectively identified through a neural network, and similarity between the question and the candidate key path is calculated to obtain semantic matching scores;
the construction information matching module comprises: introducing an attention mechanism, obtaining the matching degree between the question and different candidate key paths by calculating the attention between different parts of the question and different information of the candidate key paths, and calculating the similarity between the whole question and the candidate key path information;
the building of the comprehensive matching module comprises the following steps: combining the similarity obtained by the semantic matching module and the information matching module, comprehensively calculating the matching degree of the question sentence and the whole candidate key path, calculating the loss function of the whole model, and continuously iterating the model parameters by adopting a dynamic learning rate and an optimized gradient descent method;
and training a candidate key path matching model, and obtaining the key path of the question according to the trained candidate key path matching model.
As a preferred mode of the complex question and answer method based on the knowledge graph, the method for constructing the constraint condition screening model and selecting the proper constraint condition comprises the following steps:
constructing a second word vector module and a constraint condition screening module;
the constructing the second word vector module includes: performing fixed-length word vector processing on an input text;
the construction constraint condition screening module comprises: extracting features of the question and the constraint terms, calculating the matching degree of the question and each constraint condition to obtain the most suitable constraint term under the current question, updating the question according to the most suitable constraint term, matching the updated question with the constraint conditions, and repeating the operation until the constraint condition screening model is finished;
and training a constraint condition screening model, and obtaining the constraint condition of the question according to the trained constraint condition screening model.
The invention also discloses a knowledge graph-based complex question and answer device, which comprises:
the entity link module is used for carrying out entity identification and entity link on the input question;
the candidate key path and information extraction module is used for obtaining candidate key paths and knowledge graph information according to the subject entities of the questions;
the key path matching model module is used for constructing a candidate key path matching model and identifying a key path of a question;
the constraint item screening module is used for identifying candidate constraint conditions corresponding to the selected critical path according to the selected critical path; selecting proper constraint conditions according to the constraint condition screening model;
and the answer retrieval module is used for generating a query graph of the question according to the key path and the constraint condition and retrieving the answer in the knowledge graph.
As a preferred mode of the knowledge-graph-based complex question-answering device, the keyword path matching model module further includes:
the first word vector module is used for performing fixed-length word vector processing on the input text;
the semantic matching module is used for respectively identifying semantic information of the question and the candidate key path through the neural network, and calculating the similarity between the question and the candidate key path to obtain semantic matching scores;
the information matching module is used for introducing an attention mechanism, obtaining the matching degree between the question and different candidate key paths by calculating the attention between different parts of the question and different information of the candidate key paths, and calculating the similarity between the whole question and the candidate key path information;
and the comprehensive matching module is used for combining the similarity obtained by the semantic matching module and the information matching module, comprehensively calculating the matching degree of the question sentence and the whole candidate key path, calculating the loss function of the whole model, and continuously iterating the model parameters by adopting a dynamic learning rate and an optimized gradient descent method.
As a preferred mode of the knowledge-graph-based complex question and answer device, the constraint item screening module further includes:
the second word vector module is used for performing fixed-length word vector processing on the input text;
and the constraint condition screening module is used for extracting characteristics of the question and the constraint terms, calculating the matching degree of the question and each constraint condition, obtaining the most suitable constraint term under the current question, updating the question according to the most suitable constraint term, matching the updated question with the constraint conditions, and repeating the operation until the constraint condition screening model is ended.
The invention also discloses a storage medium, wherein the storage medium is stored with a computer program, and the computer program is used for realizing the complex question and answer method based on the knowledge graph.
The beneficial effects of the invention are as follows: the invention can fully utilize the information of each aspect of the knowledge graph, extract the key semantic information of the complex question, realize the matching of the semantic information of the question and the knowledge graph, reduce the answer range, locate the question point of the question and improve the answer accuracy through automatic screening of constraint conditions.
Drawings
Fig. 1 is a schematic flow chart of a knowledge graph-based complex question and answer method disclosed by the invention.
Fig. 2 is a block diagram of a knowledge graph-based complex question and answer implementation device.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings, but embodiments of the present invention are not limited thereto.
It should be understood that the various steps recited in the method embodiments of the present disclosure may be performed in a different order and/or performed in parallel. Furthermore, method embodiments may include additional steps and/or omit performing the illustrated steps. The scope of the present disclosure is not limited in this respect.
Example 1:
referring to fig. 1, the invention discloses a knowledge-graph-based complex question and answer method, which is mainly applied to the field of knowledge-graph questions and answers of complex questions, and replies to a mixed question by identifying a subject entity, matching a key path and screening a proper constraint condition, and comprises the following specific implementation steps:
step s1: and carrying out entity identification and entity linking on the input question.
Further said step s1 comprises:
step s11: the entity expression in the question is identified through the entity identification model, and is linked with the entity in the knowledge graph to obtain an entity set; the entity recognition model comprises a Bi-directional cyclic neural network (Bi-LSTM), a Conditional Random Field (CRF), and the like.
Step s2: and obtaining candidate key paths and knowledge graph information according to the subject entity of the question.
Further said step s2 comprises:
step s21: sequentially selecting one entity in the entity set as a subject entity, and other entities as constraint entities in constraint conditions; and according to the subject entity, searching out candidate critical paths and information related to the critical paths in the knowledge graph. The retrieved candidate critical path takes all paths within two hops from the subject entity as an example, and the related information comprises information such as entity category, answer category and the like.
Step s3: and constructing a candidate key path matching model, and identifying the key path of the question.
Further said step s3 comprises:
step s31: the candidate key path matching model is constructed and comprises a word vector module, a semantic matching module, an information matching module and a comprehensive matching module, and the specific contents are as follows:
the first word vector module: the fixed-length word vector processing is carried out on the input text, and the processing method comprises the steps of randomly initializing word vectors or adopting pre-trained word vectors. The pre-trained word vectors comprise word2vec word vectors, glove word vectors and large pre-trained models such as BERT, robert, GPT are added into the models to obtain word vectors.
Semantic matching module: semantic information of the question and the candidate key path is respectively identified through a neural network, and similarity between the question and the candidate key path is calculated to obtain semantic matching scores. The neural network comprises a Convolutional Neural Network (CNN), a deep semantic model (DSSM), a convolutional deep semantic model (CDSSM) and the like, and the similarity measure mode comprises distance measure and the like.
And the information matching module is used for: and introducing an attention mechanism, obtaining the matching degree between the question and different candidate critical paths by calculating the attention between different parts of the question and different information of the candidate critical paths, and calculating the similarity between the whole question and the candidate critical path information. The attention calculation method includes dot product method and the like.
And (3) a comprehensive matching module: and combining the similarity and other information obtained by the semantic matching module and the information matching module, comprehensively calculating the matching degree of the question sentence and the whole candidate key path, calculating the loss function of the whole model, and continuously iterating the model parameters by adopting a dynamic learning rate and an optimized gradient descent method. Other information includes information such as entity scores. The comprehensive calculation method comprises a neural network and the like.
Step s32: and obtaining the key path of the question according to the trained candidate key path matching model.
Step s4: and identifying candidate constraint conditions corresponding to the selected critical path according to the selected critical path.
Step s5: and constructing a constraint condition screening model, and selecting a proper constraint condition.
Further said step s5 comprises:
step s51: and constructing a constraint condition screening model, wherein the constraint condition screening model comprises a word vector module and a constraint condition screening module, and the specific contents are as follows.
The second word vector module: the fixed-length word vector processing is carried out on the input text, and the processing method comprises the steps of randomly initializing word vectors or adopting pre-trained word vectors. The pre-trained word vectors comprise word2vec word vectors, glove word vectors and large pre-trained models such as BERT, robert, GPT are added into the models to obtain word vectors.
Constraint condition screening module: extracting features of the question and the constraint terms, calculating the matching degree of the question and each constraint condition, obtaining the most suitable constraint term under the current question, updating the question according to the most suitable constraint term, matching the updated question with the constraint conditions, and repeating the operation until the constraint condition screening model is finished. The feature extraction mode comprises a Bi-directional cyclic neural network (Bi-RNN), a Bi-directional long-short-term memory network (Bi-LSTM), a Bi-directional gate cyclic unit network (Bi-GRU) and the like. The question updating mode comprises vector superposition, vector decrementing and the like.
Step s52: and obtaining the constraint condition of the question according to the trained constraint condition screening model.
Step s6: and generating a query graph of the question by combining the key path and the constraint condition, and searching out an answer in the knowledge graph.
Further said step s6 comprises:
step s61: and obtaining a query graph of the question according to the key path and constraint conditions obtained by the model. The query graph is implemented in a logical expression.
Step s62: and searching in the knowledge graph according to the query graph of the question sentence to obtain a final answer.
Example 2
Referring to fig. 2, the invention also discloses a knowledge graph-based complex question and answer device, which comprises:
and the entity link module is used for carrying out entity identification and entity link on the input question.
And the candidate key path and information extraction module is used for obtaining candidate key paths and knowledge graph information according to the subject entity of the question.
And the key path matching model module is used for constructing a candidate key path matching model and identifying the key path of the question.
The constraint item screening module is used for identifying candidate constraint conditions corresponding to the selected critical path according to the selected critical path; selecting proper constraint conditions according to the constraint condition screening model;
and the answer retrieval module is used for generating a query graph of the question according to the key path and the constraint condition and retrieving the answer in the knowledge graph.
Preferably, the critical path matching model module further comprises:
and the first word vector module is used for performing fixed-length word vector processing on the input text.
The semantic matching module is used for respectively identifying semantic information of the question and the candidate key path through the neural network, and calculating the similarity between the question and the candidate key path to obtain semantic matching scores.
The information matching module is used for introducing an attention mechanism, obtaining the matching degree between the question and different candidate key paths by calculating the attention between different parts of the question and different information of the candidate key paths, and calculating the similarity between the whole question and the candidate key path information.
And the comprehensive matching module is used for combining the similarity obtained by the semantic matching module and the information matching module, comprehensively calculating the matching degree of the question sentence and the whole candidate key path, calculating the loss function of the whole model, and continuously iterating the model parameters by adopting a dynamic learning rate and an optimized gradient descent method.
Preferably, the constraint item screening module further comprises:
the second word vector module is used for performing fixed-length word vector processing on the input text;
and the constraint condition screening module is used for extracting characteristics of the question and the constraint terms, calculating the matching degree of the question and each constraint condition, obtaining the most suitable constraint term under the current question, updating the question according to the most suitable constraint term, matching the updated question with the constraint conditions, and repeating the operation until the constraint condition screening model is ended.
Example 3
The invention also discloses a storage medium, and the storage medium is stored with a computer program, and the computer program is used for realizing the complex question and answer method based on the knowledge graph in the embodiment 1.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be appreciated by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (6)
1. A knowledge graph-based complex question and answer method is characterized by comprising the following steps:
performing entity identification and entity linking on the input question;
obtaining candidate key paths and knowledge graph information according to the subject entities of the questions;
constructing a candidate key path matching model, and identifying a key path of a question;
identifying candidate constraint conditions corresponding to the selected critical paths according to the selected critical paths;
constructing a constraint condition screening model, and selecting a proper constraint condition;
generating a query graph of the question by combining the key path and the constraint condition, and searching an answer in the knowledge graph;
the method for constructing the candidate key path matching model and identifying the key path of the question comprises the following steps:
constructing a first word vector module, a semantic matching module, an information matching module and a comprehensive matching module;
the constructing a first word vector module includes: performing fixed-length word vector processing on an input text;
the construction of the semantic matching module comprises the following steps: semantic information of the question and the candidate key path is respectively identified through a neural network, and similarity between the question and the candidate key path is calculated to obtain semantic matching scores;
the construction information matching module comprises: introducing an attention mechanism, obtaining the matching degree between the question and different candidate key paths by calculating the attention between different parts of the question and different information of the candidate key paths, and calculating the similarity between the whole question and the candidate key path information;
the building of the comprehensive matching module comprises the following steps: combining the similarity obtained by the semantic matching module and the information matching module, comprehensively calculating the matching degree of the question sentence and the whole candidate key path, calculating the loss function of the whole model, and continuously iterating the model parameters by adopting a dynamic learning rate and an optimized gradient descent method;
training a candidate key path matching model, and obtaining a key path of the question according to the trained candidate key path matching model;
the method for constructing the constraint condition screening model and selecting the proper constraint condition comprises the following steps:
constructing a second word vector module and a constraint condition screening module;
the constructing the second word vector module includes: performing fixed-length word vector processing on an input text;
the construction constraint condition screening module comprises: extracting features of the question and the constraint terms, calculating the matching degree of the question and each constraint condition to obtain the most suitable constraint term under the current question, updating the question according to the most suitable constraint term, matching the updated question with the constraint conditions, and repeating the operation until the constraint condition screening model is finished;
and training a constraint condition screening model, and obtaining the constraint condition of the question according to the trained constraint condition screening model.
2. The knowledge-based complex question-answering method according to claim 1, wherein the method for entity recognition and entity linking of input questions comprises:
and recognizing entity expression in the question through the entity recognition model, and linking with the entity in the knowledge graph to obtain an entity set.
3. The knowledge-based complex question and answer method according to claim 2, wherein the method for obtaining candidate key paths and knowledge-based information according to the subject entities of questions comprises:
sequentially selecting one entity in the entity set as a subject entity, and other entities as constraint entities in constraint conditions; and according to the subject entity, searching out candidate critical paths and information related to the critical paths in the knowledge graph.
4. The knowledge-based complex question-answering method according to claim 3, wherein the retrieved candidate key paths are all paths within two hops from the subject entity, and the related information includes entity category information and answer category information.
5. A knowledge graph-based complex question and answer device is characterized by comprising:
the entity link module is used for carrying out entity identification and entity link on the input question;
the candidate key path and information extraction module is used for obtaining candidate key paths and knowledge graph information according to the subject entities of the questions;
the key path matching model module is used for constructing a candidate key path matching model and identifying a key path of a question;
the constraint item screening module is used for identifying candidate constraint conditions corresponding to the selected critical path according to the selected critical path; selecting proper constraint conditions according to the constraint condition screening model;
the answer retrieval module is used for generating a query graph of the question according to the key path and the constraint condition and retrieving an answer from the knowledge graph;
the critical path matching model module further comprises:
the first word vector module is used for performing fixed-length word vector processing on the input text;
the semantic matching module is used for respectively identifying semantic information of the question and the candidate key path through the neural network, and calculating the similarity between the question and the candidate key path to obtain semantic matching scores;
the information matching module is used for introducing an attention mechanism, obtaining the matching degree between the question and different candidate key paths by calculating the attention between different parts of the question and different information of the candidate key paths, and calculating the similarity between the whole question and the candidate key path information;
the comprehensive matching module is used for combining the similarity obtained by the semantic matching module and the information matching module, comprehensively calculating the matching degree of the question sentence and the whole candidate key path, calculating the loss function of the whole model, and continuously iterating the model parameters by adopting a dynamic learning rate and an optimized gradient descent method;
the constraint item screening module further includes:
the second word vector module is used for performing fixed-length word vector processing on the input text;
and the constraint condition screening module is used for extracting characteristics of the question and the constraint terms, calculating the matching degree of the question and each constraint condition, obtaining the most suitable constraint term under the current question, updating the question according to the most suitable constraint term, matching the updated question with the constraint conditions, and repeating the operation until the constraint condition screening model is ended.
6. A storage medium having stored thereon a computer program for implementing the knowledge-graph-based complex question-answering method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110819044.7A CN113468311B (en) | 2021-07-20 | 2021-07-20 | Knowledge graph-based complex question and answer method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110819044.7A CN113468311B (en) | 2021-07-20 | 2021-07-20 | Knowledge graph-based complex question and answer method, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113468311A CN113468311A (en) | 2021-10-01 |
| CN113468311B true CN113468311B (en) | 2023-09-19 |
Family
ID=77881553
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110819044.7A Active CN113468311B (en) | 2021-07-20 | 2021-07-20 | Knowledge graph-based complex question and answer method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113468311B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115544106B (en) * | 2022-12-01 | 2023-02-28 | 云南电网有限责任公司信息中心 | Internal event retrieval method and system for call center platform and computer equipment |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776684A (en) * | 2018-05-25 | 2018-11-09 | 华东师范大学 | Optimization method, device, medium, equipment and the system of side right weight in knowledge mapping |
| CN109271506A (en) * | 2018-11-29 | 2019-01-25 | 武汉大学 | A kind of construction method of the field of power communication knowledge mapping question answering system based on deep learning |
| CN109684448A (en) * | 2018-12-17 | 2019-04-26 | 北京北大软件工程股份有限公司 | A kind of intelligent answer method |
| CN110232113A (en) * | 2019-04-12 | 2019-09-13 | 中国科学院计算技术研究所 | A kind of method and system improving the accuracy of knowledge base question and answer |
| CN110275962A (en) * | 2019-06-25 | 2019-09-24 | 北京百度网讯科技有限公司 | Method and apparatus for output information |
| CN110837550A (en) * | 2019-11-11 | 2020-02-25 | 中山大学 | Knowledge graph-based question and answer method and device, electronic equipment and storage medium |
| CN111143539A (en) * | 2019-12-31 | 2020-05-12 | 重庆和贯科技有限公司 | Knowledge graph-based question-answering method in teaching field |
| CN112199478A (en) * | 2020-09-11 | 2021-01-08 | 北京三快在线科技有限公司 | Automatic question answering method, device, electronic equipment and computer readable storage medium |
| CN112328766A (en) * | 2020-11-10 | 2021-02-05 | 四川长虹电器股份有限公司 | Knowledge graph question-answering method and device based on path search |
| CN112579797A (en) * | 2021-02-20 | 2021-03-30 | 支付宝(杭州)信息技术有限公司 | Service processing method and device for knowledge graph |
| CN112989005A (en) * | 2021-04-16 | 2021-06-18 | 重庆中国三峡博物馆 | Knowledge graph common sense question-answering method and system based on staged query |
| CN112989004A (en) * | 2021-04-09 | 2021-06-18 | 苏州爱语认知智能科技有限公司 | Query graph ordering method and system for knowledge graph question answering |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2001067225A2 (en) * | 2000-03-06 | 2001-09-13 | Kanisa Inc. | A system and method for providing an intelligent multi-step dialog with a user |
-
2021
- 2021-07-20 CN CN202110819044.7A patent/CN113468311B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776684A (en) * | 2018-05-25 | 2018-11-09 | 华东师范大学 | Optimization method, device, medium, equipment and the system of side right weight in knowledge mapping |
| CN109271506A (en) * | 2018-11-29 | 2019-01-25 | 武汉大学 | A kind of construction method of the field of power communication knowledge mapping question answering system based on deep learning |
| CN109684448A (en) * | 2018-12-17 | 2019-04-26 | 北京北大软件工程股份有限公司 | A kind of intelligent answer method |
| CN110232113A (en) * | 2019-04-12 | 2019-09-13 | 中国科学院计算技术研究所 | A kind of method and system improving the accuracy of knowledge base question and answer |
| CN110275962A (en) * | 2019-06-25 | 2019-09-24 | 北京百度网讯科技有限公司 | Method and apparatus for output information |
| CN110837550A (en) * | 2019-11-11 | 2020-02-25 | 中山大学 | Knowledge graph-based question and answer method and device, electronic equipment and storage medium |
| CN111143539A (en) * | 2019-12-31 | 2020-05-12 | 重庆和贯科技有限公司 | Knowledge graph-based question-answering method in teaching field |
| CN112199478A (en) * | 2020-09-11 | 2021-01-08 | 北京三快在线科技有限公司 | Automatic question answering method, device, electronic equipment and computer readable storage medium |
| CN112328766A (en) * | 2020-11-10 | 2021-02-05 | 四川长虹电器股份有限公司 | Knowledge graph question-answering method and device based on path search |
| CN112579797A (en) * | 2021-02-20 | 2021-03-30 | 支付宝(杭州)信息技术有限公司 | Service processing method and device for knowledge graph |
| CN112989004A (en) * | 2021-04-09 | 2021-06-18 | 苏州爱语认知智能科技有限公司 | Query graph ordering method and system for knowledge graph question answering |
| CN112989005A (en) * | 2021-04-16 | 2021-06-18 | 重庆中国三峡博物馆 | Knowledge graph common sense question-answering method and system based on staged query |
Non-Patent Citations (1)
| Title |
|---|
| "基于知识的自动问答与问题生成的研究";鲍军威;《中国博士学位论文全文数据库 信息科技辑》(第01期);I138-149 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113468311A (en) | 2021-10-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11501182B2 (en) | Method and apparatus for generating model | |
| CN117033608B (en) | Knowledge graph generation type question-answering method and system based on large language model | |
| CN111444320B (en) | Text retrieval method and device, computer equipment and storage medium | |
| CN108932342A (en) | A kind of method of semantic matches, the learning method of model and server | |
| CN113239700A (en) | Text semantic matching device, system, method and storage medium for improving BERT | |
| CN110727779A (en) | Question-answering method and system based on multi-model fusion | |
| JP2022534377A (en) | Text classification method, apparatus and computer readable storage medium | |
| CN109508459B (en) | Method for extracting theme and key information from news | |
| WO2021139229A1 (en) | Text rhetorical sentence generation method, apparatus and device, and readable storage medium | |
| CN110765277B (en) | Knowledge-graph-based mobile terminal online equipment fault diagnosis method | |
| CN112232087B (en) | Specific aspect emotion analysis method of multi-granularity attention model based on Transformer | |
| CN112328766B (en) | Knowledge graph question-answering method and device based on path search | |
| CN112328800A (en) | System and method for automatically generating programming specification question answers | |
| CN108536735B (en) | Multi-mode vocabulary representation method and system based on multi-channel self-encoder | |
| CN115599899B (en) | Intelligent question-answering method, system, equipment and medium based on aircraft knowledge graph | |
| CN112926308B (en) | Method, device, equipment, storage medium and program product for matching text | |
| CN111611452A (en) | Method, system, device and storage medium for ambiguity recognition of search text | |
| CN111666374A (en) | Method for integrating additional knowledge information into deep language model | |
| CN113468311B (en) | Knowledge graph-based complex question and answer method, device and storage medium | |
| CN111881264B (en) | Method and electronic equipment for searching long text in question-answering task in open field | |
| CN111581365B (en) | Predicate extraction method | |
| CN116775846A (en) | Domain knowledge question and answer method, system, equipment and medium | |
| CN116562280A (en) | Literature analysis system and method based on general information extraction | |
| CN110851584A (en) | Accurate recommendation system and method for legal provision | |
| Karpagam et al. | Deep learning approaches for answer selection in question answering system for conversation agents |
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 |