CN115905677A - Intelligent search system for medical field - Google Patents
Intelligent search system for medical field Download PDFInfo
- Publication number
- CN115905677A CN115905677A CN202211125136.6A CN202211125136A CN115905677A CN 115905677 A CN115905677 A CN 115905677A CN 202211125136 A CN202211125136 A CN 202211125136A CN 115905677 A CN115905677 A CN 115905677A
- Authority
- CN
- China
- Prior art keywords
- search
- data
- natural language
- module
- medical
- 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.)
- Pending
Links
Images
Classifications
-
- 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses an intelligent search system for the medical field, which consists of the following modules: the system comprises a data management module, a data processing module, a natural language processing service module, a natural language understanding module, an information retrieval module and a result display module. The invention realizes the search function based on natural language understanding by applying and researching Solr and NLU models and processing medical data by using a natural language processing technology, and uses the clear intention characteristics of the question sentences of patients, which is the reason for selecting the medical data. The NLU model data are trained through a large number of medical data labels, so that the NLU model data have excellent intention classification characteristics in different departments and different types of diseases.
Description
Technical Field
The invention relates to an intelligent search system for medical field data based on natural language understanding, and belongs to the technical field of computers.
Background
In daily life, abundant information resources bring different levels of influence to the life of people, realize symbiotic integration from the social level, subvert the traditional thinking mode, lay the foundation for harmony of human beings and nature, and digitalization becomes important force for constructing modern society. On a personal level, the change of the life style, the working style and the learning style brings convenience and rapidness to people. But the information resources on the network are numerous, which makes it a difficult problem to retrieve the available information from the rich and diverse data information. The traditional search engine can directly help the user to acquire available information and resources required by the user through query, but the accuracy, speed and relevance of information retrieval cannot well meet the requirements of the user. According to Smartinstight estimates, there are currently 50 hundred million searches per day worldwide, with 35 hundred million searches from Google accounting for 70% of the global search volume, which equates to processing 4 million searches per second.
The working principle of the traditional search engine mainly comprises the steps of network crawling, library capturing and building, webpage processing, information retrieval and result sequencing and displaying. The Google search engine, one of the best search engines on the WWW, was originally a prototype system implemented by the university of stanford, the research institute sergeybun and lawrence page. The architecture of Google is similar to a conventional search engine, and the biggest difference from the conventional search engine is that the web pages are subjected to an authority-based ranking process, so that the most important web pages appear at the top of the results.
Google calculates the PageRank value of the webpage through a PageRank algorithm, so that the position of the webpage in the result set is determined to appear, and the higher the PageRank value is, the higher the position of the webpage in the result is. Although the workflow can basically solve the problem of the search requirement of the user, the workflow has the problems of large information amount of search results, incapability of accurately understanding the intention of the user by returned contents and the like. These problems present significant difficulties for conventional search engines. However, the Google search engine released at present uses artificial intelligence technologies such as knowledge maps and natural language understanding to optimize search results, so that the search engine is more humanized.
The user wants to use the search service more conveniently and quickly, and obtains the search answer wanted by the user more accurately, and the problems are solved by the intelligent search. Intelligent search includes intellectual search services based on knowledge-graphs, intellectual search services based on natural language processing, etc., which have added to the application of the field of artificial intelligence to make traditional search engines more intelligent and "human-friendly". So that the intelligent search of natural language processing with Artificial Intelligence (AI) becomes an important research direction which is currently popular. The method is characterized in that related technologies such as natural language processing and natural language understanding are applied to a search engine, the intention of user search can be understood more accurately, the natural language can be processed into an expression mode which can be understood by a machine into a vector through means such as deep learning and machine learning, the expression mode is used for training a model, intention keywords and entity keywords are extracted from natural language query sentences, the keywords are added into a natural language processing module, interaction between the natural language and the search engine is achieved, and the user can experience more accurate, timely and efficient search experience.
Disclosure of Invention
The invention aims to provide a solution for intelligent search based on natural language understanding, and an intelligent perception search system based on natural language understanding is designed and researched aiming at question and answer data of relevant consultation in the medical field.
The technical scheme adopted by the invention is that an intelligent search system facing the medical field is composed of the following modules: the system comprises a data management module, a data processing module, a natural language processing service module, a natural language understanding module, an information retrieval module and a result display module. The data management module and the data processing module input processing into the information retrieval module, the information retrieval module and the natural language processing service module are mutually interacted, the information retrieval module and the natural language understanding module are mutually interacted, and the information retrieval module is connected with the result display module.
The invention describes the whole system architecture through the following steps, which is convenient for understanding the connection between the flow and the modules of the whole system.
Step 1: managing data;
firstly, a Solr collection is created through a Solr data management module and used for storing medical data, the local medical data is a csv format file, and data format conversion is carried out by reading the csv file data and used for subsequent index data work.
After the collection is created, the content fields in the collection are configured, because the fields existing in the medical data need to be predefined during the subsequent data indexing, only all the fields existing well are predefined, and the medical data of the required response fields are successfully imported into the collection of the Solr.
Step 2: processing data;
local medical data is stored in csv format, and before the data is indexed into Solr's collection, data needs to be cleaned and responded to filtering processes, including short text filtering, repeated text filtering and invalid text filtering.
Because these data are web crawl data, repeated text and redundant invalid text exist, and the search effect and the evaluation of the data are influenced if the data are not processed before being indexed. For example, after data is crawled and stored, the data is classified and classified into attribute fields of responses, and different processing technologies are used for realizing corresponding processing requirements according to patient question sentences, doctor answers and affiliated departments.
And 3, step 3: retrieving information;
the interaction of information retrieval is that a user builds a searchable interface with Solr data and can support two search modes, including keyword-based general search and natural language understanding-based search.
The module provides an inputtable search interface for a user, after the user inputs a query sentence in natural language, the user can select whether to start an intention recognition module stage by configuring Pipeline, if the module is not started, the module is processed by natural language processing, the query sentence is segmented by Chinese segmentation and the like, related search of full database data is directly carried out after keywords are obtained, and a search result is obtained, so that the effect is poor, the search requirement of the user can be basically met, and the real search intention of the user cannot be well understood.
If the intention recognition module is started, a search algorithm template for response needs to be configured in an intention recognition stage, and after a user inputs a natural language query sentence, an unstructured natural language search sentence is converted into structured json format data through the intention recognition module, wherein the structured json format data comprises a medical entity in the sentence, an intention keyword predicted by a model and a medical keyword. These data are the padding data in the subsequent search algorithm template. Natural language understanding-based searches are more aware of the user's true search intent and more accurate in returning search results than keyword-based general searches. Instead of only using the keywords to retrieve all data, the search efficiency and accuracy can be improved.
And 4, step 4: displaying the result;
the two search results are displayed differently, and the result display of the common search based on the keywords only comprises the search results, the search time consumption and the ranking scores of the results. The search result presentation based on natural language understanding also comprises the result generated by the intention recognition stage besides the result of the ordinary search, and can be modified and saved as the data trained by the NLU model, so that the optimization model can improve the intention recognition capability of the model.
In summary, the invention realizes the search function based on natural language understanding by applying and researching Solr and NLU models and processing the medical data by using the natural language processing technology, and the reason for selecting the medical data is that the question sentences of the patients have sharp intention characteristics. The NLU model data are trained through a large number of medical data labels, so that the NLU model data have excellent intention classification characteristics in different departments and different types of diseases.
Drawings
FIG. 1 is a diagram of a smart-aware search system and its sub-modules.
FIG. 2 is a model training profile illustration.
FIG. 3 is a flow diagram of a natural language understanding module.
FIG. 4 is a template diagram of a custom search algorithm.
Fig. 5 is a flow chart of natural language search.
FIG. 6 is a block diagram of a smart aware search system.
Detailed Description
In order to further clarify the objects, technical solutions and advantages of the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The invention develops an intelligent perception search system which can understand the search intention of a user in a humanized way by using natural language processing technologies such as word segmentation, named entity recognition, part of speech tagging and the like and combining the intention recognition technologies of a basic search engine Solr and a Rasa _ NLU adopted by the invention, a SpringBoot framework, a Python resource package and the like.
From the functional point of view, the system can be divided into the following six sub-modules: the system comprises a data management module, a data processing module, a natural language understanding module, an information retrieval module and a result display module. The organization structure is shown in figure 1.
A data management module: the medical data integration method comprises the steps of integrating medical data, wherein the medical data comprise 6 major categories, such as andrology, internal medicine, obstetrics and gynecology, oncology, pediatrics and surgery, integrating the data and indexing all the data into a collection of Solr for storage by writing a schema configuration file of the Solr. Solr's data management is mainly responsible for the creation, configuration and deletion of the collection of data sets, and the configuration, addition and deletion of the collection fields are the original data sets supporting the subsequent queries of users. The collection field includes a question field, an answer field, an intention field, a ranking score, a department field, and the like.
A data processing module: all data are cleaned, related intention keywords in different departments are defined, and real consultation problems of patients in each department are classified, and the consultation problems of the patients can mark different medical field entities, such as medical places, disease time, disease types and the like. The original data needs to be subjected to corresponding data processing before being stored in the collection of the Solr, such as short text filtering, repeated text filtering and invalid text filtering, because the data serves as the original data set to be queried and the training data set for model training, corresponding data support is provided for the natural language processing module and the natural language understanding module, and all the data to be indexed have expandability and certain high value.
The data processing module also has an important function of configuring the Pipeline for searching, configuring the stage responding in the Pipeline by creating the Pipeline name, and realizing the searching function based on natural language understanding by configuring a custom searching algorithm template required by the intention recognition algorithm searching in the system.
A natural language processing module: the natural language processing module is mainly used for processing the query sentence of the user, processing the natural language search sentence of the user by technologies such as dictionary deactivation and word segmentation, using Solr to configure Ansj segmentation for basic segmentation of the query sentence, and then further improving the accuracy of the search result by using the natural language understanding module.
A natural language understanding module: although the search behavior of the user can also be realized by using a regular or template matching mode, the problems of the reduction of the accuracy of the search result and the incapability of really understanding the intention of the user exist consistently. The natural language understanding module is mainly used for understanding the real intention of user search and extracting corresponding medical entities from user search questions for searching, and the accuracy of searching can be improved through intention keywords and medical entity keywords, and the specific configuration is as follows: the natural language understanding module can support spaCy, MITIE, skleann, tensorflow, etc. by configuring different Pipeline and realizing the backend thereof.
The Pipeline used in the experiment is MITIE + Jieba + Sklearn, and the configuration file is config _ Jieba _ MITIE _ Sklearn. Yml, as shown in FIG. 2. The configuration of sklern + MITIE is used because sklern has a fast and good intent classification characteristic, and MITIE can generate a good feature vector and a good recognition characteristic of an entity, and the Jieba participle is used because Jieba participles have a good word segmentation effect in all chinese participlers. The configuration Pipeline of MITIE + Jieba + Sklearn was used in conclusion.
The MITIE model used in Pipeline is obtained through unsupervised model training, and the MITIE model used in the experiment is generated through open source Chinese wikipedia and Baidu encyclopedia due to the fact that the training model consumes time and consumes high memory.
Medical data stored in collection in Solr and an obtained MITIE word vector model are used for training an NLU model, but the medical data need to be labeled, and the labeling can be performed manually or by using a data labeling platform.
The NLU model can then be trained by using the labeled medical data and the word vector model, which model is the model that the user subsequently intends to recognize. The model training can be performed in an off-line mode or in an on-line real-time mode. And using the trained model file to start an intention recognition service, providing the service for the system in a form of a web service interface through a flash framework based on Python, and packaging a return result.
In summary, the whole process of the natural language understanding module is shown in fig. 3, and the user performs the configuration work of the subsequent search algorithm by calling the result generated by the intention recognition interface, including the intention keyword and the medical entity.
An information retrieval module: the information retrieval module is divided into a common search and a search based on natural language understanding, and is mainly based on Solr's collection medical data.
The ordinary search, namely the matching based on key words (the Solr search engine configures a Chinese word segmentation device) and the query analyzer packaged by the Solr realize the search function.
The search based on natural language understanding is that the user query statement calls the intention keyword and the medical entity keyword generated by the intention recognition service interface, and the user-defined search algorithm is matched to form the query statement, as shown in fig. 4, the query statement is used in a Solr query parser, so that the search function based on natural language understanding is realized.
The natural language search includes the above two ways to perform the search requirement on the data, as shown in fig. 5. The specific process is as follows:
(1) Search flow based on natural language understanding: firstly, creating data management Pipeline, wherein the data management Pipeline is used for configuring different stages to operate data, configuring a natural language understanding stage, compiling a search algorithm template in the stage, judging whether the stage is added and started or not, and if the intention identification stage is started, generating three parts of contents including intention keywords, medical entities and medical keywords through a natural language understanding module. And finally, replacing the keywords in the search template with the contents for recombination, using the recombined keywords in a search engine to generate a final search result, and completing a search function based on natural language understanding.
(2) A common search process based on keywords: and if the intention recognition stage is closed and the condition that the stage is not used or not created is created, using the default search based on the keywords, using the keywords of the user search statement generated by the natural language processing module in a Solr search engine to generate a final search result and finishing the common search function based on the keywords.
And a result display module: the common search result presentation of the user includes the total number of search results, the search time, and the ranking scores of the search results. The highest ranking score is shown first and all search results are shown in descending order. Natural language understanding based searches include results presentation and modifiable functionality of a natural language understanding service interface in addition to search results contained by a common search. The result of the intention recognition comprises an intention keyword (intent), a medical entity keyword (entity) and a medical entity (value), and if the recognition result is wrong, the results can be modified and stored as intention recognition model training data, so that the accuracy of the model is further optimized, and the intention recognition capability of the model is improved.
Claims (10)
1. An intelligent search system oriented to the medical field is characterized by comprising the following modules: the system comprises a data management module, a data processing module, a natural language processing service module, a natural language understanding module, an information retrieval module and a result display module; the data management module and the data processing module input processing into the information retrieval module, the information retrieval module and the natural language processing service module are mutually interacted, the information retrieval module and the natural language understanding module are mutually interacted, and the information retrieval module is connected with the result display module.
2. The intelligent search system for medical field according to claim 1, wherein the whole system architecture is described by the following steps, which facilitate understanding of the connection between the processes and modules of the whole system;
step 1: data management of the data management module;
firstly, a Solr collection is established through a Solr data management module and used for storing medical data, the local medical data is a csv format file, and data format conversion is carried out by reading the csv file data for subsequent index data work;
after the collection is created, configuring content fields in the collection, and importing the medical data of the required response fields into the collection of the Solr;
step 2: data processing of the data processing module;
storing local medical data in a csv format, and performing cleaning and response filtering treatment on the local medical data before indexing the local medical data into a collection of Solr, wherein the filtering treatment comprises short text filtering, repeated text filtering and invalid text filtering;
and step 3: information retrieval of the information retrieval module;
the information retrieval interaction is that a user builds a searchable interface with Solr data and supports two search modes, including common search based on keywords and search based on natural language understanding;
the information retrieval module provides an inputtable search interface for a user, the user selects whether to start an intention recognition module stage by configuring Pipeline after inputting a query sentence in natural language, if the module is not started, the module is processed by natural language processing, the query sentence is segmented by Chinese segmentation, related search of data in a full database is directly carried out after keywords are obtained, and a search result is obtained;
if the intention identification module is started, a search algorithm template for response needs to be configured in an intention identification stage, after a user inputs a natural language query sentence, an unstructured natural language search sentence is converted into structured json format data through the intention identification module, wherein the structured json format data comprises a medical entity in the sentence, an intention keyword predicted by a model and a medical keyword; these intention keywords and medical keywords are the fill data in the subsequent search algorithm templates; compared with the common search based on key words, the search based on natural language understanding has the advantages that the real search intention of a user is more understood and the search result is more accurately returned;
and 4, step 4: displaying the result of the result displaying module;
the two search results are displayed differently, and the result display of the common search based on the keywords only comprises the search results, the search time consumption and the ranking scores of the results; the search results based on natural language understanding are shown to include the results generated by the intention recognition stage in addition to the results of the ordinary search, and can be modified and saved as NLU model training data for optimizing the model to improve the intention recognition capability of the model.
3. The medical field-oriented intelligent search system of claim 1, wherein the data management module: the method comprises the steps of integrating medical data, wherein the medical data comprises 6 major categories, namely andrology, internal medicine, obstetrics and gynecology, oncology, pediatrics and surgery, integrating the data and indexing all the data into a collection of Solr for storage by compiling a schema configuration file of the Solr; solr is responsible for the creation, configuration and deletion of a data set collection, and the configuration, addition and deletion of a collection field are original data sets supporting subsequent queries of users; the collection field includes a question field, an answer field, an intention field, a ranking score, a department field.
4. The medical-field-oriented intelligent search system according to claim 1, wherein the data processing module: cleaning all data, defining related intention keywords in different departments, classifying real consultation problems of patients in each department, and marking different medical field entities on the consultation problems of the patients; before being stored in the collection of Solr, the original data needs to be correspondingly processed, so that all the data to be indexed have expandability and certain high-value type.
5. The system of claim 1, wherein the data processing module further comprises an important function of configuring Pipeline for searching, configuring the responding stage in the Pipeline by creating the Pipeline name, and implementing the natural language understanding-based search function by configuring a custom search algorithm template required for searching by the intention recognition algorithm.
6. The medical-field-oriented intelligent search system according to claim 1, wherein the natural language processing module: the natural language processing module processes the query sentence of the user, processes the natural language search sentence of the user by configuring a stop dictionary and a word segmentation mode, configures Ansj word segmentation by Solr for basic segmentation of the query sentence, and then further improves the accuracy of the search result by using the natural language understanding module.
7. The medical-field-oriented intelligent search system according to claim 1, wherein the natural language understanding module is configured to understand a real intention of a user search and extract a corresponding medical entity from a user search question for search, and the accuracy of the search is improved by an intention keyword and a medical entity keyword, and the natural language understanding module is specifically configured as follows: the natural language understanding module supports spaCy, MITIE, sklern and tensorflow at the back end of the natural language understanding module by configuring different Pipeline.
8. The medical field-oriented intelligent search system according to claim 1, wherein the information retrieval module: the information retrieval module is divided into common search and search based on natural language understanding, and is used for carrying out medical data based on Solr collection;
the ordinary search, namely the query analyzer based on the matching of the key words and Solr encapsulation, realizes the search function;
the search based on natural language understanding is that the user query sentence calls the intention keyword and medical entity keyword generated by the intention recognition service interface, the self-defined search algorithm is matched to form the query sentence, and the query sentence is used in the Solr query analyzer, so that the search function based on natural language understanding is realized.
9. The medical field-oriented intelligent search system according to claim 1, wherein the natural language search includes the above two ways to perform the data search requirement, and the specific flow is as follows:
(1) Search flow based on natural language understanding: firstly, creating data management Pipeline which is used for configuring different stages to operate data, configuring a natural language understanding stage, compiling a search algorithm template in the stage, judging whether the stage is added and started or not, and if the intention identification stage is started, generating three parts of contents including intention keywords, medical entities and medical keywords through a natural language understanding module; finally, replacing the keywords in the search template for recombination, using the keywords in a search engine to generate a final search result, and completing a search function based on natural language understanding;
(2) General search procedure based on keywords: and if the intention recognition stage is closed and the condition that the stage is not used or not created is created, using the default search based on the keywords, using the keywords of the user search statement generated by the natural language processing module in a Solr search engine to generate a final search result and finishing the common search function based on the keywords.
10. The medical field-oriented intelligent search system according to claim 1, wherein the result presentation module: the common search result display of the user comprises the total number of the search results, the search time consumption and the ranking scores of the search results; the highest ranking score is displayed at the head, and all search results are displayed in a descending manner; the natural language understanding-based search includes a result presentation of a natural language understanding service interface and modifiable functions in addition to search results included in a general search; and if the recognition result is wrong, the results can be modified and stored as intention recognition model training data, the accuracy of the model is optimized, and the intention recognition capability of the model is improved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211125136.6A CN115905677A (en) | 2022-09-14 | 2022-09-14 | Intelligent search system for medical field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211125136.6A CN115905677A (en) | 2022-09-14 | 2022-09-14 | Intelligent search system for medical field |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115905677A true CN115905677A (en) | 2023-04-04 |
Family
ID=86482422
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211125136.6A Pending CN115905677A (en) | 2022-09-14 | 2022-09-14 | Intelligent search system for medical field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115905677A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116705270A (en) * | 2023-08-07 | 2023-09-05 | 北方健康医疗大数据科技有限公司 | Medical data management system, method and storage medium |
-
2022
- 2022-09-14 CN CN202211125136.6A patent/CN115905677A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116705270A (en) * | 2023-08-07 | 2023-09-05 | 北方健康医疗大数据科技有限公司 | Medical data management system, method and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109284363B (en) | Question answering method and device, electronic equipment and storage medium | |
| CN109492077B (en) | Knowledge graph-based petrochemical field question-answering method and system | |
| Joby | Expedient information retrieval system for web pages using the natural language modeling | |
| CN102087669B (en) | Intelligent search engine system based on semantic association | |
| CN109271505A (en) | A kind of question answering system implementation method based on problem answers pair | |
| CN111353030A (en) | Knowledge question and answer retrieval method and device based on travel field knowledge graph | |
| CN106776532B (en) | Knowledge question-answering method and device | |
| Quamar et al. | An ontology-based conversation system for knowledge bases | |
| KR20050036541A (en) | Semi-automatic construction method for knowledge of encyclopedia question answering system | |
| CN113590783B (en) | NLP natural language processing-based traditional Chinese medicine health preserving intelligent question-answering system | |
| US20030212663A1 (en) | Neural network feedback for enhancing text search | |
| KR102398832B1 (en) | Device, method and computer program for deriving response based on knowledge graph | |
| CN112115252B (en) | Intelligent auxiliary writing processing method and device, electronic equipment and storage medium | |
| CN111324691A (en) | Intelligent question-answering method for minority nationality field based on knowledge graph | |
| CN111061828B (en) | Digital library knowledge retrieval method and device | |
| CN113569023A (en) | Chinese medicine question-answering system and method based on knowledge graph | |
| CN112015915A (en) | Question-answering system and device based on knowledge base generated by questions | |
| CN112036178A (en) | Distribution network entity related semantic search method | |
| CN111104437A (en) | Test data unified retrieval method and system based on object model | |
| CN112507089A (en) | Intelligent question-answering engine based on knowledge graph and implementation method thereof | |
| US11487795B2 (en) | Template-based automatic software bug question and answer method | |
| CN115905677A (en) | Intelligent search system for medical field | |
| CN113010657B (en) | Answer processing method and answer recommendation method based on answer text | |
| CN114647719A (en) | Question-answering method and device based on knowledge graph | |
| CN112084312A (en) | Intelligent customer service system constructed based on 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 |