CN114328880A - Intelligent question and answer method and system for automobile field - Google Patents
Intelligent question and answer method and system for automobile field Download PDFInfo
- Publication number
- CN114328880A CN114328880A CN202210061029.5A CN202210061029A CN114328880A CN 114328880 A CN114328880 A CN 114328880A CN 202210061029 A CN202210061029 A CN 202210061029A CN 114328880 A CN114328880 A CN 114328880A
- Authority
- CN
- China
- Prior art keywords
- question
- answer
- text
- answering
- module
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an intelligent question-answering method and system used in the field of automobiles, wherein the method comprises the following steps: converting voice or text information input by a user into an input text; correcting the input text to obtain a corrected text after error correction; performing text matching on the corrected text and a pre-trained file matching model, and judging that the corrected text is a single-round question-answer question or a plurality of rounds of question-answer questions; if the question is a single round of question and answer, adopting a retrieval question and answer to return a reply; if the answer questions are multi-turn questions, verifying whether the predefined slot position has a corresponding slot value in the process of communicating with the user, if so, extracting slot value information in the correction text, contacting the context, and returning a final reply according to the obtained information; if not, the user is asked for the answer, and the slot value information is extracted according to the answer of the user until the final reply is returned. The method can fully understand the user intention, is suitable for various scenes in the automobile field, and provides high-quality service for the user.
Description
Technical Field
The invention belongs to the technical field of intelligence, and particularly relates to an intelligent question answering method and system for the field of automobiles.
Background
At the present stage, the intelligent question-answering system as a productivity tool undergoes the level conversion from the specific effect of the algorithm to cost reduction and efficiency improvement in actual combat. Nevertheless, the landing of the intelligent question-answering system also faces many problems, and how to better understand the user's problems is the most concerned of all relevant practitioners.
At present, the intelligent question-answering system is developed rapidly, and each technology is mature day by day. Chinese patent CN202110778221.1 discloses an intelligent question-answering system facing the automobile field, and specifically discloses that the system comprises a knowledge base module, a visual interaction module, an intention identification module, a graph matching module, a template matching module, a retrieval module and an end-to-end module; the knowledge base module stores a knowledge map and a corpus of the automobile field; after the user inputs the question, the input content of the user is judged, and the corresponding module of the system is called to process according to different judged user purposes to obtain the answer of the question. Dividing the user purpose into two categories of automobile field questioning and chatting, and aiming at the automobile field questioning, obtaining question answers by using a question answering method based on an automobile field knowledge map; an answer is generated for the chat using an end-to-end module based on deep learning. The patent can improve the classification precision and accurately identify the user intention. However, the whole intelligent question-answering system cannot solve complex question-answering scenes. In a real scene, a user often has a contextual relation, and needs to understand the needs of the user and guide the user to execute according to a flow, so that better service is provided for the user.
Disclosure of Invention
In view of the above-mentioned deficiencies in the prior art, the present invention aims to provide an intelligent question-answering method and system for the automotive field, which can fully understand the user's intention and is suitable for various scenes in the automotive field to provide high-quality services for the user.
The technical scheme of the invention is realized as follows:
an intelligent question answering method for the automobile field comprises the following steps:
s1: converting voice or text information input by a user into an input text;
s2: correcting the input text by adopting a pre-trained error correction model to obtain a corrected text after error correction;
s3: performing text matching on the corrected text and a pre-trained file matching model so as to judge that the corrected text is a single round of question-answering questions or multiple rounds of question-answering questions; if the corrected text is a single round of question and answer questions, the step goes to S4, and if the corrected text is a multi-round of question and answer questions, the step goes to S5;
s4: returning a reply by adopting a retrieval question and answer, wherein the retrieval question and answer are matched with standard questions in the standard corpus and the trained semantic model according to the corrected text, and returning a reply corresponding to the standard question with high similarity as a final reply;
s5: pre-defining slot position information, verifying whether a corresponding slot value exists in a slot position in the communication process with a user, if so, extracting slot value information in a correction text, contacting with a context, and returning a final reply according to the obtained information; if not, the user is asked for the answer, and the slot value information is extracted according to the answer of the user until the final reply is returned.
Further, retrieving the answer includes the steps of:
(1) firstly, matching the corrected text with a first standard problem in a standard corpus, calculating character similarity between the corrected text and the first standard problem, and returning to the first standard problem with higher similarity in the previous N numbers;
(2) matching the corrected text with a second standard problem in a pre-trained semantic model, calculating semantic similarity between the corrected text and the second standard problem in the semantic model, and returning the previous M second standard problems with higher similarity;
(3) recalling the N returned first standard problems and the M returned second standard problems to form a candidate set;
(4) and matching the corrected text with the candidate set, calculating the similarity between the corrected text and the candidate set by adopting an interactive semantic model, sequencing based on the calculation result, and finally returning a final reply according to a preset threshold value.
Further, in the retrieval answers, the threshold value of the accurate reply is 0.9, the threshold value of the recommended reply is 0.6, when the calculation result is greater than or equal to 0.9, 1 reply with the highest score is returned, when the calculation result is greater than or equal to 0.6 and less than 0.9, the recommended reply is returned, and a plurality of replies with higher scores are returned.
Further, for a single round of question-answering, returning reply by adopting atlas question-answering, wherein the atlas question-answering matches the corrected text with a pre-constructed knowledge atlas by using an image matching method, and returning final reply if matching is successful; at this time, the final reply is returned according to the accuracy rate or the business rule of the reply of the retrieval question-answer and the atlas question-answer.
Further, the atlas question-answer comprises the following steps:
(1) extracting SPO triples from a corpus in advance, and then constructing a knowledge graph;
(2) carrying out syntactic analysis and SPO extraction on the corrected text, and then generating a graph search language;
(3) and (4) carrying out searching and matching on the graph search language in the knowledge graph, thereby returning a final reply.
An intelligent question-answering system for the field of automobiles comprises a conversion module, an error correction module, an intention identification module, a retrieval question-answering module and a multi-turn question-answering module.
The conversion module is used for converting the voice or text information input by the user into an input text.
The error correction module is used for correcting the error of the input text according to the pre-trained error correction model so as to obtain a corrected text after error correction.
The intention recognition module is used for judging whether the corrected text is a single-turn question-answer question or a multi-turn question-answer question, transmitting the single-turn question-answer question to the search question-answer module, and transmitting the multi-turn question-answer question to the multi-turn question-answer module.
The retrieval question-answering module is used for returning a reply corresponding to a standard question with higher corrected text character similarity or semantic similarity.
The multi-turn question-answering module is used for verifying whether a corresponding slot value exists in a predefined slot position in the communication process with a user, if so, extracting slot value information in the correction text, contacting with the context, and returning a final reply according to the obtained information; if not, the user is asked for the answer, and the slot value information is extracted according to the answer of the user until the final reply is returned.
Furthermore, the intelligent question-answering system also comprises a map question-answering module and an arbitration module.
The map question-answering module is internally provided with a pre-constructed knowledge map and used for receiving single-round question-answering questions, matching the correction text with the knowledge map by using a map matching method, and returning to final reply if matching is successful.
The arbitration module is used for arbitrating the replies returned by the retrieval question-answering module and the map question-answering module so as to return the final reply.
Further, when the arbitration module carries out arbitration, final reply is returned according to the accuracy rate or the business rule of reply of the retrieval question answer and the map question answer.
Further, a slot position extraction unit is arranged in the multi-turn question and answer module and used for extracting slot value information in the correction text.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention processes the single-turn question and answer problem by retrieving the question and answer and the atlas question and answer in a combined manner, ensures that the single-turn question and answer problem can be effectively solved, and simultaneously obtains the most real intention of the user by multiple turns of question and answer to solve the intelligent question and answer under the complex scene, so that the invention can be applied to various scenes in the automobile field, thereby providing high-quality service for the user.
2. The invention can improve the accuracy of the retrieval question and answer under the single-turn question and answer scene by adopting a mode of jointly using the character similarity and the semantic similarity.
Drawings
FIG. 1-schematic flow chart of the present invention.
Fig. 2-an exemplary graph of the query car wash index.
Fig. 3-an exemplary diagram of a remote controlled air conditioner.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
Referring to fig. 1, an intelligent question answering method for the automobile field is characterized by comprising the following steps:
s1: converting voice or text information input by a user into an input text;
s2: correcting the input text by adopting a pre-trained error correction model to obtain a corrected text after error correction;
s3: performing text matching on the corrected text and a pre-trained file matching model so as to judge that the corrected text is a single round of question-answering questions or multiple rounds of question-answering questions; if the corrected text is a single round of question and answer questions, the step goes to S4, and if the corrected text is a multi-round of question and answer questions, the step goes to S5;
the method is mainly used for intention recognition, and the file matching model is divided into a single round of question-answering questions and multiple rounds of question-answering questions facing the technical field of automobiles and chatting.
S4: returning a reply by adopting a retrieval question and answer, wherein the retrieval question and answer are matched with standard questions in the standard corpus and the trained semantic model according to the corrected text, and returning a reply corresponding to the standard question with high similarity as a final reply;
s5: pre-defining slot position information, verifying whether a corresponding slot value exists in a slot position in the communication process with a user, if so, extracting slot value information in a correction text, contacting with a context, and returning a final reply according to the obtained information; if not, the user is asked for the answer, and the slot value information is extracted according to the answer of the user until the final reply is returned.
Therefore, under a plurality of complex scenes, a multi-turn question answering method can be adopted to solve the problem that a single turn of question answering can not be answered. For example, in a car-shopping scenario, multiple interactions with the user are usually required to obtain the real intention of the user.
Taking the car washing index query as an example, as shown in fig. 2, when the text input by the user is "query car washing index", because the problem lacks related entity words, the system does not know the time and place of the car washing index that needs to be queried specifically, and the semantic expression of the user is unclear, the system can obtain necessary entity words through multiple back-questioning, for example, the user inputs "xx city" and "today", the system will contact the context to obtain the complete demand of the user, namely query car washing index of xx city today, and according to the demand, corresponding external service is requested, and relevant information is returned to the user.
As shown in fig. 3, a user says to the mobile phone APP before going up: the system recognizes the intention of turning on the air conditioner and adjusting the temperature in the vehicle to the memory question and answer commonly used by the user.
In particular, the search response includes the following steps:
(1) firstly, matching the corrected text with a first standard problem in a standard corpus, calculating character similarity between the corrected text and the first standard problem, and returning to the first standard problem with higher similarity in the previous N numbers;
(2) matching the corrected text with a second standard problem in a pre-trained semantic model, calculating semantic similarity between the corrected text and the second standard problem in the semantic model, and returning the previous M second standard problems with higher similarity;
(3) recalling the N returned first standard problems and the M returned second standard problems to form a candidate set;
(4) and matching the corrected text with the candidate set, calculating the similarity between the corrected text and the candidate set by adopting an interactive semantic model, sequencing based on the calculation result, and finally returning a final reply according to a preset threshold value.
In the specific implementation, in the retrieval answers, the threshold value of the accurate reply is 0.9, the threshold value of the recommended reply is 0.6, when the calculation result is greater than or equal to 0.9, 1 reply with the highest score is returned, when the calculation result is greater than or equal to 0.6 and less than 0.9, the recommended reply is returned, and a plurality of replies with higher scores are returned.
In actual use, the recommended replies may return 5 replies with higher scores.
During specific implementation, for a single round of question and answer, returning a reply by adopting a map question and answer, matching the corrected text with a pre-constructed knowledge map by using a map matching method by the map question and answer, and returning a final reply if matching is successful; at this time, the final reply is returned according to the accuracy rate or the business rule of the reply of the retrieval question-answer and the atlas question-answer.
Because the question and answer retrieval has certain limitations, such as the judgment problems of tire pressure and the like, the retrieval problems can not be processed, and the problems can be effectively solved by using the map question and answer. Therefore, the question-answer of multiple scenes can be solved by adopting the retrieval question-answer and the map question-answer
In specific implementation, the atlas question-answer comprises the following steps:
(1) extracting SPO triples from a corpus in advance, and then constructing a knowledge graph;
(2) carrying out syntactic analysis and SPO extraction on the corrected text, and then generating a graph search language;
(3) and (4) carrying out searching and matching on the graph search language in the knowledge graph, thereby returning a final reply.
An intelligent question-answering system for the field of automobiles comprises a conversion module, an error correction module, an intention identification module, a retrieval question-answering module and a multi-turn question-answering module.
The conversion module is used for converting the voice or text information input by the user into an input text.
The error correction module is used for correcting the error of the input text according to the pre-trained error correction model so as to obtain a corrected text after error correction.
The intention recognition module is used for judging whether the corrected text is a single-turn question-answer question or a multi-turn question-answer question, transmitting the single-turn question-answer question to the search question-answer module, and transmitting the multi-turn question-answer question to the multi-turn question-answer module.
The retrieval question-answering module is used for returning a reply corresponding to a standard question with higher corrected text character similarity or semantic similarity.
The multi-turn question-answering module is used for verifying whether a corresponding slot value exists in a predefined slot position in the communication process with a user, if so, extracting slot value information in the correction text, contacting with the context, and returning a final reply according to the obtained information; if not, the user is asked for the answer, and the slot value information is extracted according to the answer of the user until the final reply is returned.
When the system is specifically implemented, the system also comprises a map question-answering module and an arbitration module.
The map question-answering module is internally provided with a pre-constructed knowledge map and used for receiving single-round question-answering questions, matching the correction text with the knowledge map by using a map matching method, and returning to final reply if matching is successful.
The arbitration module is used for arbitrating the replies returned by the retrieval question-answering module and the map question-answering module so as to return the final reply.
In specific implementation, when the arbitration module arbitrates, the final reply is returned according to the accuracy rate or the business rule of the reply of the retrieval question answer and the map question answer.
In specific implementation, a slot position extraction unit is arranged in the multi-turn question and answer module and is used for extracting slot value information in the correction text.
Finally, it should be noted that the above-mentioned examples of the present invention are only examples for illustrating the present invention, and are not intended to limit the embodiments of the present invention. Variations and modifications in other variations will occur to those skilled in the art upon reading the foregoing description. Not all embodiments are exhaustive. All obvious changes and modifications of the present invention are within the scope of the present invention.
Claims (9)
1. An intelligent question answering method for the field of automobiles is characterized by comprising the following steps:
s1: converting voice or text information input by a user into an input text;
s2: correcting the input text by adopting a pre-trained error correction model to obtain a corrected text after error correction;
s3: performing text matching on the corrected text and a pre-trained file matching model so as to judge that the corrected text is a single round of question-answering questions or multiple rounds of question-answering questions; if the corrected text is a single round of question and answer questions, the step goes to S4, and if the corrected text is a multi-round of question and answer questions, the step goes to S5;
s4: returning a reply by adopting a retrieval question and answer, wherein the retrieval question and answer are matched with standard questions in the standard corpus and the trained semantic model according to the corrected text, and returning a reply corresponding to the standard question with high similarity as a final reply;
s5: pre-defining slot position information, verifying whether a corresponding slot value exists in a slot position in the communication process with a user, if so, extracting slot value information in a correction text, contacting with a context, and returning a final reply according to the obtained information; if not, the user is asked for the answer, and the slot value information is extracted according to the answer of the user until the final reply is returned.
2. The intelligent question-answering method for the automobile field according to claim 1, wherein the search for answers comprises the steps of:
(1) firstly, matching the corrected text with a first standard problem in a standard corpus, calculating character similarity between the corrected text and the first standard problem, and returning to the first standard problem with higher similarity in the previous N numbers;
(2) matching the corrected text with a second standard problem in a pre-trained semantic model, calculating semantic similarity between the corrected text and the second standard problem in the semantic model, and returning the previous M second standard problems with higher similarity;
(3) recalling the N returned first standard problems and the M returned second standard problems to form a candidate set;
(4) and matching the corrected text with the candidate set, calculating the similarity between the corrected text and the candidate set by adopting an interactive semantic model, sequencing based on the calculation result, and finally returning a final reply according to a preset threshold value.
3. The intelligent question answering method for the automobile field according to claim 2, wherein in the retrieval answers, the threshold value of the accurate answer is 0.9, the threshold value of the recommended answer is 0.6, when the calculation result is greater than or equal to 0.9, 1 answer with the highest score is returned, when the calculation result is greater than or equal to 0.6 and less than 0.9, the answer is recommended, and a plurality of answers with higher scores are returned.
4. The intelligent question-answering method for the automobile field according to claim 1, characterized in that for a single round of question-answering, a map question-answering return reply is adopted, the map question-answering matches the correction text with a pre-constructed knowledge map by using a map matching method, and if the matching is successful, a final reply is returned; at this time, the final reply is returned according to the accuracy rate or the business rule of the reply of the retrieval question-answer and the atlas question-answer.
5. The intelligent question-answering method for the automobile field according to claim 4, wherein the atlas question-answering comprises the following steps:
(1) extracting SPO triples from a corpus in advance, and then constructing a knowledge graph;
(2) carrying out syntactic analysis and SPO extraction on the corrected text, and then generating a graph search language;
(3) and (4) carrying out searching and matching on the graph search language in the knowledge graph, thereby returning a final reply.
6. An intelligent question-answering system for the field of automobiles is characterized by comprising a conversion module, an error correction module, an intention identification module, a retrieval question-answering module and a multi-turn question-answering module;
the conversion module is used for converting voice or text information input by a user into an input text;
the error correction module is used for correcting the error of the input text according to a pre-trained error correction model to obtain a corrected text after error correction;
the intention recognition module is used for judging whether the corrected text is a single-turn question-answer question or a multi-turn question-answer question, transmitting the single-turn question-answer question to the search question-answer module and transmitting the multi-turn question-answer question to the multi-turn question-answer module;
the retrieval question-answering module is used for returning a reply corresponding to a standard question with higher corrected text character similarity or semantic similarity;
the multi-turn question-answering module is used for verifying whether a corresponding slot value exists in a predefined slot position in the communication process with a user, if so, extracting slot value information in the correction text, contacting with the context, and returning a final reply according to the obtained information; if not, the user is asked for the answer, and the slot value information is extracted according to the answer of the user until the final reply is returned.
7. The intelligent question-answering system for the automobile field according to claim 6, further comprising a map question-answering module and an arbitration module;
the map question-answering module is internally provided with a pre-constructed knowledge map and is used for receiving single-round question-answering questions, matching the correction text with the knowledge map by using a map matching method, and returning to final reply if matching is successful;
the arbitration module is used for arbitrating the replies returned by the retrieval question-answering module and the map question-answering module so as to return the final reply.
8. The system according to claim 7, wherein the arbitration module returns the final response according to the accuracy or business rules of the responses of the search question-answers and the map question-answers during arbitration.
9. The intelligent question answering system for the automobile field according to claim 6, wherein a slot position extracting unit is arranged in the multi-turn question answering module, and the slot position extracting unit is used for extracting slot value information in the correction text.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210061029.5A CN114328880A (en) | 2022-01-19 | 2022-01-19 | Intelligent question and answer method and system for automobile field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210061029.5A CN114328880A (en) | 2022-01-19 | 2022-01-19 | Intelligent question and answer method and system for automobile field |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114328880A true CN114328880A (en) | 2022-04-12 |
Family
ID=81027971
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210061029.5A Pending CN114328880A (en) | 2022-01-19 | 2022-01-19 | Intelligent question and answer method and system for automobile field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114328880A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115132178A (en) * | 2022-07-15 | 2022-09-30 | 科讯嘉联信息技术有限公司 | Semantic endpoint detection system based on deep learning |
| CN115238101A (en) * | 2022-09-23 | 2022-10-25 | 中国电子科技集团公司第十研究所 | Multi-engine intelligent question-answering system oriented to multi-type knowledge base |
-
2022
- 2022-01-19 CN CN202210061029.5A patent/CN114328880A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115132178A (en) * | 2022-07-15 | 2022-09-30 | 科讯嘉联信息技术有限公司 | Semantic endpoint detection system based on deep learning |
| CN115132178B (en) * | 2022-07-15 | 2023-01-10 | 科讯嘉联信息技术有限公司 | Semantic endpoint detection system based on deep learning |
| CN115238101A (en) * | 2022-09-23 | 2022-10-25 | 中国电子科技集团公司第十研究所 | Multi-engine intelligent question-answering system oriented to multi-type knowledge base |
| CN115238101B (en) * | 2022-09-23 | 2023-01-03 | 中国电子科技集团公司第十研究所 | Multi-engine intelligent question-answering system oriented to multi-type knowledge base |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110196901B (en) | Method and device for constructing dialog system, computer equipment and storage medium | |
| CN111309889B (en) | Method and device for text processing | |
| CN114328880A (en) | Intelligent question and answer method and system for automobile field | |
| CN111177310B (en) | Intelligent scene conversation method and device for power service robot | |
| CN111177359A (en) | Multi-turn dialogue method and device | |
| CN111079418B (en) | Named entity recognition method, device, electronic equipment and storage medium | |
| CN108763355B (en) | User-based intelligent robot interactive data processing system and method | |
| CN110825865A (en) | Multi-round conversation intelligent customer service system based on special word correction and cold start | |
| CN117271753B (en) | Intelligent property question-answering method and related products | |
| CN110727776A (en) | Automobile question-answer interaction system and method based on artificial intelligence | |
| CN111199149A (en) | Intelligent statement clarifying method and system for dialog system | |
| CN112185187B (en) | Learning method and intelligent device for social language | |
| CN109684357B (en) | Information processing method and device, storage medium and terminal | |
| CN111651554A (en) | Insurance question-answer method and device based on natural language understanding and processing | |
| CN115952260A (en) | Question and answer method and device in automobile field, electronic equipment and storage medium | |
| CN114328846B (en) | Intelligent customer service response system and method for policy consultation in colleges and universities | |
| CN115168558A (en) | Method for realizing multi-round man-machine conversation | |
| CN113963235A (en) | Cross-category image recognition model reusing method and system | |
| CN109559739B (en) | Content recommendation method and system based on voice search | |
| CN113449091A (en) | Intelligent question and answer method, device, terminal and computer readable storage medium based on automobile field label | |
| CN112905785A (en) | Question-answer knowledge base construction method based on E-commerce dialogue corpus | |
| CN113569022B (en) | Method for realizing dialogue robot response engine based on cascade search | |
| CN110704585A (en) | Question answering method, device and computer readable medium | |
| CN114579606B (en) | Pre-training model data processing method, electronic device and computer storage medium | |
| CN111488431B (en) | Hit determination method, device and system |
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 |