WO2023236252A1

WO2023236252A1 - Answer generation method and apparatus, electronic device, and storage medium

Info

Publication number: WO2023236252A1
Application number: PCT/CN2022/100568
Authority: WO
Inventors: 段沛宸
Original assignee: 来也科技(北京)有限公司
Priority date: 2022-06-07
Filing date: 2022-06-22
Publication date: 2023-12-14
Also published as: CN114936276A

Abstract

The present invention relates to the technical fields of robotic process automation (RPA) and artificial intelligence (AI), and relates to an answer generation method and apparatus, an electronic device, and a storage medium. The method comprises: obtaining a query statement and a question type to which the query statement belongs; obtaining a target content fragment matching the query statement from among a plurality of content fragments comprised in at least one document; and according to a response policy corresponding to the question type, on the basis of the target content fragment, generating a target answer corresponding to the query statement. Therefore, an answer is automatically generated instead of manual work, thereby reducing labor and time costs required for generating the answer; and the target content fragment capable of answering a question of a user is accurately determined from the document, and the answer corresponding to the query statement is generated according to the target content fragment, thereby improving the accuracy of the generated answer. According to the present invention, a content fragment in an acquisition document of IA can be achieved by combining the RPA and the AI, thereby further reducing the labor costs for generating an answer.

Description

Answer generation method, device, electronic device and storage medium

Cross-references to related applications

This application is filed based on a Chinese patent application with application number 2022106357290 and a filing date of June 7, 2022, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby incorporated by reference into this application.

Technical field

The present disclosure relates to the technical fields of robotic process automation and artificial intelligence, and specifically relates to an answer generation method, device, electronic equipment and storage medium.

Background technique

Robotic Process Automation (RPA) uses specific "robot software" to simulate human operations on a computer and automatically execute process tasks according to rules.

Artificial Intelligence (AI for short) is a technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.

Intelligent Automation (IA) is a general term for a series of technologies from robotic process automation to artificial intelligence. It combines RPA with Optical Character Recognition (OCR), Intelligent Character Recognition (ICR), and process mining. (Process Mining), Deep Learning (DL), Machine Learning (ML), Natural Language Processing (NLP), Speech Recognition (Automatic Speech Recognition, ASR), Speech Synthesis (Text To Speech) , TTS), Computer Vision (CV) and other AI technologies are combined to create end-to-end business processes that can think, learn and adapt, covering from process discovery, process automation, to automatic and continuous The entire process of data collection, understanding the meaning of data, and using data to manage and optimize business processes.

At present, in many business scenarios, such as power question and answer systems, it is necessary to find specific content that can answer the questions raised by users from a large number of documents, such as a certain sentence or certain cells in a table. content, etc., and then give accurate answers based on the content. In related technologies, after obtaining questions raised by users, a large number of documents are usually manually queried to find specific content that can answer the user's questions, and answers are given based on the specific content, or from FAQ (Frequently Asked Questions) Answers) library to find answers that match the user's questions. The above-mentioned method of answering questions through manual query will waste a lot of labor costs and time costs, and the method of answering questions through FAQ can only answer questions that already exist in the FAQ, and cannot answer questions that do not exist in the FAQ. Accurate answer. How to accurately answer user questions with lower labor costs and time costs has become an urgent problem to be solved.

Contents of the invention

The present disclosure provides an answer generation method, device, electronic device and storage medium to solve the technical problems of high labor and time costs and poor accuracy of the answer generation method.

An embodiment of the first aspect of the present disclosure provides an answer generation method. The method includes: obtaining a query statement and a question type to which the query statement belongs; and obtaining a target content fragment matching the query statement from a plurality of content fragments included in at least one document. ; According to the response strategy corresponding to the question type, based on the target content fragment, the target answer corresponding to the query statement is generated.

In some embodiments, the question type includes one of numeric type, extraction type, and judgment type; the number of target content fragments is multiple; according to the response strategy corresponding to the question type, based on the target content fragment, a target answer corresponding to the query statement is generated. , including: for each target content fragment, input the query statement and the target content fragment into the extraction model in the field of natural language processing NLP to extract the candidate answer fragment corresponding to the query statement from the target content fragment, and obtain the corresponding confidence level; according to The confidence level corresponding to each candidate answer fragment is used to obtain the target answer fragment from each candidate answer fragment; according to the response strategy corresponding to the question type, the target answer is generated based on the target answer fragment.

In some embodiments, the question type includes an extraction class; according to the response strategy corresponding to the question type, the target answer is generated based on the target answer fragment, including: using the target answer fragment as the target answer.

In some embodiments, the question type includes a judgment type; according to the response strategy corresponding to the question type, generating the target answer based on the target answer fragment includes: inputting the target answer fragment and the query statement into a judgment model in the NLP field to obtain the query statement corresponding to Judgment result; use the judgment result and/or target answer fragment as the target answer.

In some embodiments, the question type includes a numeric type; according to the response strategy corresponding to the question type, generating the target answer based on the target answer fragment includes: obtaining the target number from the target answer fragment according to preset rules, and obtaining the unit corresponding to the target number. ; Generate the target answer based on the target number and the corresponding unit.

In some embodiments, the question type includes a statistical type; according to the response strategy corresponding to the question type, based on the target content fragment, a target answer corresponding to the query statement is generated, including: extracting the target content fragment through a regular expression extraction rule, so as to Get the target answer.

In some embodiments, before obtaining the target content fragment that matches the query statement from the plurality of content fragments included in at least one document, the method further includes: obtaining the target question that matches the query statement from a preset question and answer set; based on the NLP field The first correlation model is used to obtain the first correlation between the query statement and the target question; it is determined that the first correlation is not greater than the preset threshold.

In some embodiments, the method further includes: when the first correlation is greater than a preset threshold, obtaining the answer corresponding to the target question from the question and answer set; determining the answer corresponding to the target question as the target answer corresponding to the query statement.

In some embodiments, obtaining target content fragments matching the query statement from multiple content fragments included in at least one document includes: querying based on the query statement to obtain from the multiple content fragments related to the query statement. Multiple candidate content segments; based on the second correlation model in the NLP field, obtain the second correlation between the query statement and each candidate content segment; based on each second correlation, obtain the target content segment from each candidate content segment.

In some embodiments, before obtaining the target content fragment matching the query statement from multiple content fragments included in at least one document, the method further includes: based on optical character recognition OCR technology in the field of artificial intelligence (AI), performing on each document Recognize to obtain the recognition results of each document; perform structured processing on each recognition result to obtain multiple content fragments included in each document; and store each content fragment in correspondence with the corresponding content field.

In some embodiments, based on the optical character recognition OCR technology in the field of artificial intelligence, each document is recognized to obtain the recognition results of each document, including: calling the RPA robot to upload each document to the document processing platform for document processing. The platform uses optical character recognition (OCR) technology to identify each document; and obtains the recognition results of each document returned by the document processing platform.

A second embodiment of the present disclosure provides an answer generation device, including: a first acquisition module for acquiring a query statement and the question type to which the query statement belongs; and a second acquisition module for obtaining multiple contents included in at least one document. In the fragment, the target content fragment matching the query statement is obtained; the generation module is used to generate the target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type.

In some embodiments, the question type includes one of a numerical class, an extraction class, and a judgment class; the number of target content segments is multiple; the generation module includes: a first acquisition unit, used for each target content segment, The query statement and the target content fragment are input into the extraction model in the field of natural language processing NLP to extract the candidate answer fragments corresponding to the query statement from the target content fragment and obtain the corresponding confidence level; the second acquisition unit is used to extract the candidate answer fragments according to each candidate answer fragment. The corresponding confidence level is used to obtain the target answer fragment from each candidate answer fragment; the generation unit is used to generate the target answer based on the target answer fragment according to the response strategy corresponding to the question type.

The third embodiment of the present disclosure provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned first step of the present disclosure is implemented. Methods described in aspect embodiments.

The fourth embodiment of the present disclosure provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method described in the first embodiment of the present disclosure is implemented.

The fifth aspect embodiment of the present disclosure proposes a computer program product, which includes a computer program. When executed by a processor, the computer program implements the method described in the above first aspect embodiment of the present disclosure.

The sixth embodiment of the present disclosure provides a computer program. The computer program includes computer program code. When the computer program code is run on a computer, it causes the computer to execute the method described in the first embodiment of the present disclosure. method.

The technical solutions provided by the embodiments of the present disclosure may include the following beneficial effects:

After obtaining the query statement and the question type to which the query statement belongs, obtain the target content fragment that matches the query statement from multiple content fragments included in at least one document, and then generate based on the target content fragment according to the response strategy corresponding to the question type. The target answer corresponding to the query statement. Therefore, by automatically generating answers instead of manual work, the labor cost and time cost required to generate answers are reduced, and by accurately determining the target content fragment that can answer the user's question from the document, and generating the query statement corresponding to the target content fragment. answers, improving the accuracy of the generated answers.

Additional aspects and advantages of the disclosure will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the disclosure.

Description of the drawings

In the drawings, unless otherwise specified, the same reference numbers refer to the same or similar parts or elements throughout the several figures. The drawings are not necessarily to scale. It should be understood that these drawings depict only some embodiments in accordance with the disclosure and are not to be considered limiting of the scope of the disclosure.

Figure 1 is a schematic flowchart of an answer generation method according to the first embodiment of the present disclosure;

Figure 2 is an example diagram of an interactive interface provided by an answer generation device according to the first embodiment of the present disclosure;

Figure 3 is a schematic flowchart of an answer generation method according to a second embodiment of the present disclosure;

Figure 4 is a schematic flowchart of an answer generation method according to a third embodiment of the present disclosure;

Figure 5 is a schematic flowchart of an answer generation method according to a fourth embodiment of the present disclosure;

Figure 6 is an example diagram of an interactive interface of a document processing platform and a recognition result of a document according to the fourth embodiment of the present disclosure;

Figure 7 is an example diagram of text recognition results and corresponding content fragments according to the fourth embodiment of the present disclosure;

Figure 8 is an example diagram of table recognition results and corresponding content fragments according to the fourth embodiment of the present disclosure;

Figure 9 is a schematic structural diagram of an answer generation device according to a fifth embodiment of the present disclosure;

FIG. 10 is a block diagram of an electronic device used to implement the answer generation method of an embodiment of the present disclosure.

Detailed ways

Embodiments of the present disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present disclosure and are not to be construed as limitations of the present disclosure.

These and other aspects of embodiments of the present disclosure will become apparent with reference to the following description and accompanying drawings. In these descriptions and drawings, some specific implementations of the embodiments of the disclosure are specifically disclosed to represent some of the ways of implementing the principles of the embodiments of the disclosure, but it should be understood that the scope of the embodiments of the disclosure is not limited by this restriction. On the contrary, the disclosed embodiments include all changes, modifications and equivalents falling within the spirit and scope of the appended claims.

It should be noted that in the technical solution of this disclosure, the acquisition, storage and application of user personal information involved are in compliance with relevant laws and regulations and do not violate public order and good customs.

Embodiments of the present disclosure provide an answer generation method. After obtaining a query statement and the question type to which the query statement belongs, target content fragments matching the query statement are obtained from multiple content fragments included in at least one document, and then the target content fragments matching the query statement are obtained according to the question type. The corresponding response strategy generates the target answer corresponding to the query statement based on the target content fragment. Therefore, by replacing the manual automatic generation of the answer, the labor cost and time cost required to generate the answer are reduced, and the answer can be accurately determined from the document. The target content fragment of the user's question is generated, and the answer corresponding to the query statement is generated based on the target content fragment, which improves the accuracy of the generated answer.

In order to clearly explain the various embodiments of the present disclosure, technical terms involved in the embodiments of the present disclosure are first explained.

In the description of the present disclosure, the term "plurality" means two or more.

In the description of this disclosure, "RPA robot" refers to a software robot that can combine AI technology and RPA technology to automatically perform business processing. RPA robots have two characteristics: "connector" and "non-intrusion". By simulating human operation methods, they can extract, integrate and connect data from different systems in a non-intrusive way without changing the information system.

In the description of this disclosure, "query statement" refers to the statement input by the user for query, that is, the question the user wants to ask. It can be a statement in text form or a statement in voice form. This disclosure does not make any comment on this. limit.

In the description of this disclosure, a "document" is an electronic document used to retrieve specific content that can answer a user's question and generate an answer to the user's question accordingly. It can be a PDF obtained by scanning a paper document. (Portable Document Format, Portable Document Format) format documents can also be documents edited on smart devices such as computers and mobile phones, and this disclosure does not limit this.

In the description of this disclosure, a "content fragment" is a fragment composed of part of the content in the document. The content fragment can be one sentence or several sentences, or it can be a paragraph in the document, or a table in the document, or Partial content in a table, etc., this disclosure does not limit this. In some embodiments of the present disclosure, the number of characters included in the content fragments can be set in advance, so that by processing all documents to be retrieved, the content in all documents is divided into multiple content fragments, and the characters included in each content fragment are The number is less than or equal to the preset number of characters.

In the description of this disclosure, "candidate content fragments" refer to content fragments related to the query statement obtained from all content fragments included in all documents. "Target content fragment" refers to the content fragment matching the query statement obtained from the candidate content fragment or all content fragments included in all documents, that is, the specific content that can accurately answer the user's question.

In the description of this disclosure, "answer fragments" are more fine-grained fragments in content fragments, and answers to user questions can be generated based on the answer fragments. "Candidate answer fragment" is an answer fragment obtained from the target content fragment. "Target answer fragment" is an answer fragment obtained from candidate answer fragments.

In the description of this disclosure, a "question and answer set" is a preset set including multiple candidate questions and corresponding answers, such as FAQ.

In the description of this disclosure, "attribute information" is information that represents the attributes of a content fragment, such as the document name of the document where the content fragment is located, the chapter title and chapter number corresponding to the content fragment, the parent titles of each level of the chapter title, etc.

In the description of this disclosure, "correlation degree" is used to express the magnitude of the degree of correlation.

In the description of this disclosure, "correlation model" is any machine model that can calculate the degree of correlation, such as Bert (Bidirectional Encoder Representations from Transformers, a bidirectional encoder representation model) and other neural network models. In some embodiments, the relevance model can be obtained by fine-tuning a pre-trained model in the NLP field.

In the description of this disclosure, "judgment model" is any machine model that can realize judgment, such as a neural network model, and this disclosure does not limit this.

In the description of this disclosure, "extraction model" is any machine model that can realize information extraction, such as a neural network model, and this disclosure does not limit this.

In the description of this disclosure, "preset rules" are preset extraction rules, which may be in the form of regular expressions or other forms, and this disclosure does not limit this. In order to facilitate distinction in this disclosure, the preset extraction rules for extracting target numbers from target answer segments are called first preset rules, and the preset rules for extracting target answers from target content segments are called second preset rules. Default rules.

In the description of the present disclosure, "content relevance" is the correlation between the query statement and the content fragment determined based on the content contained in the content fragment, and is used to represent the correlation between the content contained in the content fragment and the query statement. The size of the degree.

In the description of this disclosure, "attribute correlation" is the correlation between the query statement and the content fragment determined based on the attribute information corresponding to the content fragment, and is used to represent the correlation between the attribute information corresponding to the content fragment and the query statement. The size of the degree.

In the description of this disclosure, "segmented fragments" refer to fragments composed of content obtained by dividing the document. For example, after the document is divided into multiple sentences according to the punctuation marks used at the end of the sentence, each sentence is A split fragment. Each content segment in the embodiment of the present disclosure may include one or more segmented segments.

In the description of this disclosure, a "document processing platform" is an intelligent automation platform for intelligently processing documents. Among them, Intelligent Document Processing (IDP) is one of the core capabilities of the intelligent automation platform. Intelligent document processing (IDP) is based on AI technologies such as Optical Character Recognition (OCR), Computer Vision (CV), Natural Language Processing (NLP), and Knowledge Graph (KG). , a new generation of automation technology that identifies, classifies, extracts elements, verifies, compares, and corrects errors in various types of documents, helping enterprises realize the intelligence and automation of document processing.

In the description of this disclosure, a "content field" is a field composed of a single character or multiple consecutive characters. The "content field" can be understood as the attribute item key, and the content contained in the content fragment can be understood as the attribute value value. The content fields and corresponding content fragments together form a piece of structured data. In addition, the content field and the fields corresponding to the attribute information of the content fragment, such as the field named "Document Name", the field named "Chapter Title", and the field named "Parent Title at All Levels", can form a structure. .

Answer generation methods, devices, electronic devices and storage media according to embodiments of the present disclosure are described below with reference to the accompanying drawings.

First, the answer generation method in the embodiment of the present disclosure will be described with reference to the accompanying drawings.

Figure 1 is a flowchart of an answer generation method according to the first embodiment of the present disclosure. As shown in Figure 1, the method may include the following steps: 101-103.

Step 101: Obtain the query statement and the question type to which the query statement belongs.

It should be noted that the answer generation method in the embodiment of the present disclosure can be executed by an answer generation device. In some embodiments, the answer generating device can be implemented by software and/or hardware, and the answer generating device can be an electronic device, or can be configured in an electronic device to automatically generate accurate answers to user questions instead of manual work. In some embodiments, the electronic device may include but is not limited to a terminal device, a server, etc. This embodiment does not specifically limit the electronic device. In some embodiments, the answer generating device may be an intelligent answering system.

In the embodiment of the present disclosure, the answer generation device can provide an interactive interface, so that the user can input a query statement in the interactive interface to perform a query, and accordingly, the answer generation device can obtain the query statement.

In the embodiment of the present disclosure, the classification model can be pre-trained, so that the query statement can be input into the classification model, and the question type to which the query statement belongs can be obtained based on the output of the classification model. The classification model can be any model in related technologies that can realize classification, such as a neural network model, and this disclosure does not limit this.

In the embodiment of the present disclosure, the question type to which the query statement belongs may include numerical type, statistical type, extraction type, judgment type, etc.

In the embodiment of the present disclosure, the numerical type means that the corresponding answer is a specific number. For example, if the query statement is "A newly put into operation 220KV transformer, the rest time should be no less than how many hours before voltage is applied?" If a specific number needs to be answered, the question type to which the query statement belongs is numeric. "KV" means kilovolts.

Statistical category means that the corresponding answers need to be counted. For example, if the query statement is "How many types of chip radiators can be divided according to cooling methods?" and the corresponding answer needs to count the types of chip radiators, then the question type to which the query statement belongs is statistical.

Extraction type means that the corresponding answer needs to be extracted from a piece of text or a table. For example, if the query statement is "What are the replacement cycle requirements for wearing parts?" and the corresponding answer needs to be extracted from a paragraph of text or a table, then the question type to which the query statement belongs is the extraction class.

Judgment type means that the corresponding answer is "yes" or "no". For example, if the query statement is "Does the 750KV oil-immersed transformer meet the requirements for 72 hours of rest after oil change?" and the corresponding answer is "yes" or "no", then the question type to which the query statement belongs is a judgment type. where "h" refers to the hour.

Step 102: Obtain a target content fragment matching the query statement from multiple content fragments included in at least one document.

In this embodiment of the disclosure, the number of target content segments may be one or multiple, and the disclosure does not limit this.

In the embodiment of the present disclosure, a large number of documents to be retrieved (that is, documents that need to retrieve specific content that can answer the user's questions and provide answers accordingly) can be processed in advance to obtain multiple content fragments, and then obtain the query statement. Afterwards, the target content fragment matching the query statement can be obtained from multiple content fragments.

In the embodiment of the present disclosure, the number of target content fragments can be set in advance, so that the answer generation device can obtain the correlation between the query statement and each content fragment, and arrange each content fragment in order from high to low according to the corresponding correlation. Sorting is performed, and the preset number of content fragments that are sorted first are determined as the target content fragments.

In the embodiment of the present disclosure, the correlation threshold can be set in advance (for ease of differentiation, it can be called the first correlation threshold), so that the answer generation device can obtain the correlation between the query statement and each content segment, and combine each content Among the fragments, the content fragment whose corresponding correlation degree is greater than the first correlation degree threshold is determined as the target content fragment. The first correlation threshold can be set arbitrarily as needed, and this disclosure does not limit this.

Step 103: Generate a target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type.

In the embodiment of the present disclosure, the response strategy is a preset strategy for generating target answers corresponding to query statements based on target content segments. Among them, different response strategies can be set for different question types.

In the embodiment of the present disclosure, the answer generation device can provide an interactive interface, so that after generating the target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type, the target answer can be displayed through the interactive interface. In addition, the answer generation device can also display the question type to which the query statement belongs, the target content fragment, the attribute information corresponding to the target content fragment, and the paragraph or table containing the target content fragment through the interactive interface while displaying the target answer (where the target Content fragments or paragraphs or tables containing target content fragments as the basis for answers) and other information, so that users can more clearly understand the source of the target answer to the query statement.

For example, refer to Figure 2, taking the answer generation device as an intelligent response system as an example. The intelligent response system can provide an interactive interface. After the user enters the query statement "Does the 750KV oil-immersed transformer meet the requirements for resting for 72 hours after oil change" on the interactive interface, The intelligent response system can determine that the question type to which the query statement belongs is a judgment type, and then obtain the target content fragment that matches the query statement "transformers after new installation, overhaul, accident maintenance or oil change" from multiple content fragments included in at least one document , the resting time before applying voltage should not be less than the following provisions: a) 110KV 24h b) 220KV 48h c) 500(330)KV 72h d) 750KV 96h", and obtain the chapter number "5.2.6" corresponding to the content fragment, Then, according to the response strategy corresponding to the judgment type, based on the target content fragment, the target answer "No, 96h" corresponding to the query statement is generated, and as shown in Figure 2, the target answer, question type, target content fragment and corresponding are displayed through the interactive interface Chapter number.

In summary, the answer generation method provided by the embodiment of the present disclosure, after obtaining the query statement and the question type to which the query statement belongs, obtains the target content fragment matching the query statement from multiple content fragments included in at least one document, and then according to the The response strategy corresponding to the question type generates the target answer corresponding to the query statement based on the target content fragment. Therefore, by automatically generating answers instead of manual work, the labor cost and time cost required to generate answers are reduced, and by accurately determining the target content fragment that can answer the user's question from the document, and generating the query statement corresponding to the target content fragment. answers, improving the accuracy of the generated answers.

The following is a further explanation of the process of generating the target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type in the answer generation method provided by the embodiment of the present disclosure with reference to FIG. 3 .

Figure 3 is a flow chart of an answer generation method according to the second embodiment of the present disclosure. As shown in Figure 3, the method includes steps 301-306.

Step 301: Obtain the query statement and the question type to which the query statement belongs.

Step 302: Obtain the target content fragment matching the query statement from multiple content fragments included in at least one document.

For the specific implementation process and principle of step 302, please refer to the descriptions of other embodiments and will not be described again here.

Step 303: When the question type includes one of numerical class, extraction class, and judgment class, for each target content segment, input the query statement and the target content segment into the extraction model in the field of natural language processing NLP to extract the target content from the target content segment. Extract candidate answer fragments corresponding to the query statement from the fragments, and obtain the corresponding confidence levels.

Wherein, when the question type includes one of numerical type, extraction type, and judgment type, the number of target content fragments can be multiple, for example, it can be 20, 30, etc.

Among them, confidence represents the probability that the target content fragment can answer the query statement.

In the embodiment of the present disclosure, the extraction model can be trained in advance. For each target content fragment, after the answer generation device inputs the query statement and the target content fragment into the trained extraction model, the extraction model can determine that the target answer corresponding to the query statement is in Input the starting position and ending position in the target content segment, and then determine the segment between the starting position and the ending position in the target content segment as the candidate answer segment, determine the corresponding confidence level, and output the candidate answer segment and the corresponding The confidence level, so that the answer generation device can obtain the candidate answer fragments corresponding to the query statement and the corresponding confidence level based on the output of the extraction model.

It should be noted that the step of obtaining the question type to which the query statement belongs can be performed before step 302 or after step 302. This disclosure does not limit this and it only needs to be performed before step 303.

Step 304: Obtain the target answer segment from each candidate answer segment according to the confidence level corresponding to each candidate answer segment.

In the embodiment of the present disclosure, the corresponding candidate answer segment with the highest confidence among each candidate answer segment may be determined as the target answer segment.

Step 305: Generate a target answer based on the target answer fragment according to the response strategy corresponding to the question type.

In the embodiment of the present disclosure, when the question type includes an extraction class, the target answer fragment can be directly used as the target answer. That is, step 305 includes: using the target answer fragment as the target answer.

For example, assuming that the query statement is "What are the replacement cycle requirements for wearing parts" of the extraction class, the answer generation device obtains a target content fragment that matches the query statement from multiple content fragments included in at least one document. "5.1.6 Replacement cycle of wearing parts. If the oil pump bearing or cooling fan bearing that has been used for more than 10 years makes abnormal noise during operation, it should be replaced when the transformer or shunt reactor is out of operation; if it has been used for more than 15 years, it should be replaced according to the specific conditions. "Replace all gaskets", according to the process of step 303, the candidate answer fragment extracted from the target content fragment is "When the oil pump bearing or cooling fan bearing that has been used for more than 10 years makes abnormal noise during operation, the transformer or parallel connection Replace the reactor when it is out of operation; when it is used for more than 15 years, replace all seals according to specific conditions."

Assuming that the confidence corresponding to the candidate answer fragment is the highest among the candidate answer fragments, the candidate answer fragment can be determined as the target answer fragment, and the target answer fragment can be used as the target answer corresponding to the query statement.

Thus, when the query statement is an extracted class, the target answer corresponding to the query statement can be accurately generated from the document.

In the embodiment of the present disclosure, when the question type includes a judgment class, step 305 can be implemented in the following manner: input the target answer fragment and the query statement into the judgment model in the NLP field to obtain the judgment result corresponding to the query statement, and convert the judgment The result and/or target answer fragment serves as the target answer.

Among them, the judgment result can be "yes" or "no".

Specifically, the probability threshold can be set in advance, such as 0.5, and a judgment model in the NLP field can be pre-trained. After inputting the target answer fragment and query statement into the judgment model, the judgment model can determine and output the answer corresponding to the query statement as "yes" "The probability. The answer generating device can determine the judgment result to be "yes" when the probability is greater than the probability threshold 0.5, and determine the judgment result to be "no" when the probability is not greater than the probability threshold 0.5, and then combine the judgment result with/ or a target answer fragment as the target answer.

For example, assuming that the query statement is of the judgment type "Does a 750KV oil-immersed transformer meet the requirements for 72 hours of rest after oil change?", the answer generation device obtains one that matches the query statement from multiple content fragments included in at least one document. The target content fragment is "For newly installed, overhauled, accident-repaired or oil-changed transformers, the resting time before applying voltage should not be less than the following provisions: a) 110KV 24h b) 220KV 48h c) 500 (330) KV 72h d) 750KV 96h", according to the process of step 303, the candidate answer segment extracted from the target content segment is "96h".

Assuming that the confidence corresponding to the candidate answer fragment is the highest among the candidate answer fragments, the candidate answer fragment "96h" can be determined as the target answer fragment, and then the target answer fragment "96h" and the query statement can be input into the judgment model in the NLP field , to obtain the judgment results corresponding to the query statement. Since the target answer fragment "96h" is greater than "72h" in the query statement, the probability that the answer corresponding to the query statement output by the judgment model is "yes" is less than 0.5, so the answer generation device can determine that the judgment result is "no", and then The judgment result "No" and the target answer fragment "96h" can be used as the target answer.

Thus, when the query statement is a judgment type, the target answer corresponding to the query statement can be accurately generated from the document.

In the embodiment of the present disclosure, when the question type includes numbers, step 305 can be implemented in the following manner: according to the first preset rule, obtain the target number from the target answer fragment, and obtain the unit corresponding to the target number; according to The target number and the corresponding unit are used to generate the target answer.

The first preset rule may be in the form of a regular expression.

Specifically, the answer generation device can extract the target number from the target answer fragment based on the regular expression, and at the same time extract the unit corresponding to the target number, and then splice the target number and the corresponding unit into the target answer. Alternatively, the units corresponding to the target answer fragments can also be set in advance, so that after the answer generating device extracts the target number from the target answer fragment, the target number and the preset unit can be spliced into the target answer.

For example, assuming that the query statement is a numeric type "A newly put into operation 220KV transformer, the rest time should be no less than how many hours before applying voltage?", the answer generation device obtains it from multiple content fragments included in at least one document. One of the target content fragments that matches the query statement is "3.0.3 The insulation test of oil-immersed transformers and reactors should be filled with qualified oil and allowed to stand for a certain period of time until the bubbles are eliminated. The standing time should be determined by the manufacturer. When the manufacturer does not specify, the relationship between the voltage level of oil-immersed transformers and reactors and the resting time after oil filling should be determined according to Table 3.0.3. Table 3.0.3 The voltage levels of oil-immersed transformers and reactors and the relationship between oil filling and charging The relationship between the resting time after oil filling >= 48", according to the process of step 303, the candidate answer fragment extracted from the target content fragment is "The relationship between the voltage level of oil-immersed transformer and reactor and the resting time after oil filling >= 48 ".

Assuming that the confidence level corresponding to this candidate answer fragment is the highest among all candidate answer fragments, it can be determined that the candidate answer fragment "The relationship between the voltage level of oil-immersed transformers and reactors and the rest time after oil filling >= 48" is the target answer fragment , and then based on regular expressions, the target number "48" can be extracted from the target answer fragment. Assuming that the preset unit is "h", the target number "48" and the preset unit "h" can be spliced into the target answer "48h".

As a result, when the query statement is numeric, the target answer corresponding to the query statement can be accurately generated from the document.

Step 306: If the question type includes statistics, extract the target answer from the target content segment according to the second preset rule.

The second preset rule may be in the form of a regular expression.

Wherein, when the question type is statistics, the number of target content fragments may be one.

In the embodiment of the present disclosure, the target content fragment can be extracted based on the regular expression to obtain the target answer.

For example, assuming that the number of target content fragments is one and the query statement is a statistical "chip radiator can be divided into several categories according to cooling methods", the answer generation device obtains from multiple content fragments included in at least one document. The target content fragment matching the query statement is "4.1.2 According to the cooling method, it is divided into: a) self-cooling (ONAN); b) air-cooling (ONAF); c) strong oil air-cooling (OFAF)".

Then the answer generation device can extract the target content fragment based on the regular expression and obtain the target answer "self-cooling (ONAN), air-cooling (ONAF), strong oil air-cooling (OFAF)".

As a result, when the query statement is of statistical type, the target answer corresponding to the query statement can be accurately generated from the document.

In summary, the answer generation method provided by the embodiment of the present disclosure, after obtaining the query statement and the question type to which the query statement belongs, obtains the target content fragment matching the query statement from multiple content fragments included in at least one document. When the type includes one of numeric class, extraction class, and judgment class, for each target content fragment, input the query statement and the target content fragment into an extraction model in the field of natural language processing NLP to extract the query statement from the target content fragment. Corresponding candidate answer fragments and obtain the corresponding confidence. According to the confidence corresponding to each candidate answer fragment, obtain the target answer fragment from each candidate answer fragment. According to the response strategy corresponding to the question type, generate the target answer based on the target answer fragment. When the question type includes statistics, the target answer is extracted from the target content segment according to the second preset rule. As a result, by automatically generating answers instead of manual work, the labor cost and time cost required to generate answers are reduced. For query statements of each question type, the target content fragment that can answer the user's question can be accurately determined from the document, and based on The target content fragment generates answers corresponding to the query statements, which improves the accuracy of the generated answers.

In the embodiment of the present disclosure, the target answer corresponding to the query statement can also be generated based on the answer generation process in the above embodiment and a preset question and answer set such as FAQ. In view of the above situation, the answer generation method provided by the embodiment of the present disclosure will be further described below with reference to FIG. 4 .

Figure 4 is a flow chart of an answer generation method according to the third embodiment of the present disclosure. As shown in Figure 4, the method includes steps 401-408.

Step 401: Obtain the query statement and the question type to which the query statement belongs.

For the specific implementation process and principle of step 401, reference can be made to the description of the above embodiments and will not be described again here.

Step 402: Obtain target questions matching the query statement from the preset question and answer set.

In the embodiment of the present disclosure, target questions matching the query statement can be obtained from a preset question and answer set based on a search engine.

Specifically, each candidate question included in the preset question and answer set can correspond to the question type to which the annotation belongs, and then based on the search engine, the query statement can be obtained from each candidate question whose annotated question type is the same as the question type to which the query statement belongs. Matching target problem.

Step 403: Based on the first correlation model in the NLP field, obtain the first correlation between the query statement and the target question.

In the embodiment of the present disclosure, the first relevance model can be trained in advance. After obtaining the target question, the answer generation device can input the query statement and the target question into the first relevance model, and the first relevance model can output the query statement and the target question. The correlation score between the questions is scored, so that the answer generating device can obtain the first correlation between the query statement and the target question according to the output of the first correlation model.

Step 404: Determine whether the first correlation degree is greater than the preset threshold. If so, execute step 405; otherwise, execute step 407.

Step 405: Obtain the answer corresponding to the target question from the question and answer set.

Step 406: Determine the answer corresponding to the target question as the target answer corresponding to the query statement.

The preset threshold can be set as needed, and this disclosure does not limit this.

In the embodiment of the present disclosure, when the first correlation is greater than the preset threshold, the answer corresponding to the target question can be obtained from the question and answer set, and the answer corresponding to the target question is determined as the target answer corresponding to the query statement.

As a result, the target answer corresponding to the query statement can be quickly generated based on the preset question and answer set, and the generated target answer is highly accurate.

Step 407: Obtain the target content fragment matching the query statement from multiple content fragments included in at least one document.

Step 408: Generate a target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type.

For the specific implementation process and principles of steps 407-408, please refer to the descriptions of other embodiments and will not be described again here.

In the embodiment of the present disclosure, when the first correlation is not greater than the preset threshold, the target content fragment matching the query statement can be obtained from multiple content fragments included in at least one document, and the target content fragment corresponding to the question type can be obtained. The response strategy generates the target answer corresponding to the query statement based on the target content fragment.

Thus, when the user's question cannot be accurately answered based on the preset question and answer set, the target content fragment that can answer the user's question can be accurately determined from the document, and the answer corresponding to the query statement is generated based on the target content fragment, and the generated The accuracy of the target answer is high. Moreover, by combining obtaining the target answer from the preset question and answer set and generating the target answer based on the target content fragment in the document, these two methods are used to generate the target answer, so that there is no need to waste a lot of labor costs to maintain the preset question and answer set, thereby reducing The cost of manually maintaining a preset question and answer set.

The following is a further explanation of the process of obtaining a target content segment that matches a query statement from multiple content segments included in at least one document in the answer generation method provided by the embodiment of the present disclosure with reference to FIG. 5 .

Figure 5 is a flow chart of an answer generation method according to the fourth embodiment of the present disclosure. As shown in Figure 5, the method includes steps 501-508.

Step 501: Obtain the query statement and the question type to which the query statement belongs.

Step 502: Perform a query based on the query statement to obtain multiple candidate content fragments related to the query statement from multiple content fragments included in at least one document.

In the embodiment of the present disclosure, a large number of documents to be retrieved can be processed in advance to obtain multiple content fragments, and the multiple content fragments can be saved in the retrieval engine. Then, after the answer generation device obtains the query statement, the retrieval can be based on the The engine performs a query based on the query statement to obtain multiple candidate content fragments related to the query statement from multiple content fragments, and returns them to the answer generation device. Correspondingly, the answer generating device can obtain multiple candidate content segments.

The retrieval engine can be any retrieval engine with a retrieval function, and this disclosure does not limit this. In addition, the retrieval engine may be configured in the answer generation device, or the retrieval engine may be configured separately and connected to the answer generation device through an interface, which is not limited by the present disclosure.

In the embodiment of the present disclosure, the number of candidate content fragments can be set in advance, so that the retrieval engine can obtain the correlation between the query statement and each content fragment, and process each content fragment in order from high to low according to the corresponding correlation. Sorting: determine a preset number of content fragments that are ranked first as multiple candidate content fragments.

In the embodiment of the present disclosure, the correlation threshold can be set in advance (for ease of differentiation, it can be called the second correlation threshold), so that the retrieval engine can obtain the correlation between the query statement and each content fragment, and combine each content fragment , multiple content segments whose corresponding correlation degrees are greater than the second correlation threshold are determined as multiple candidate content segments. The second correlation threshold can be set arbitrarily as needed, and this disclosure does not limit this.

In this embodiment of the present disclosure, step 502 can be implemented in the following manner: obtaining the content contained in each content fragment and the attribute information of each content fragment; based on the content contained in each content fragment, obtaining the relationship between the query statement and the corresponding content fragment. content correlation, and based on the attribute information of each content fragment, obtain the attribute correlation between the query statement and the corresponding content fragment; based on the content correlation and attribute correlation between the query statement and each content fragment, from multiple In the content fragments, obtain multiple candidate content fragments related to the query statement.

The attribute information of the content fragment may include at least one of the document name of the document in which the content fragment is located, the chapter title corresponding to the content fragment, and the parent titles at all levels of the chapter title corresponding to the content fragment. When the attribute information of the content fragment includes multiple information such as document name, chapter title, parent title at each level, etc., correspondingly, for each content fragment, each attribute information between the query statement and the corresponding content fragment can be obtained. Attribute correlation.

Taking the attribute information including document name, chapter title, and parent title at all levels as an example, the content contained in each content fragment and the attribute information of the content fragment can be saved in the form of a structure. The fields in the structure can include names. A field named "Document Name", a field named "Chapter Title", a field named "Parent Title at All Levels" and a field named "Content Fragment", so that the answer generation device can obtain the corresponding content based on each structure. The content contained in the fragment and the corresponding attribute information.

In the embodiment of the present disclosure, the query statement can be segmented into words, and the content correlation between the query statement and the content segment can be determined based on the number of times each segment appears in the content contained in the content segment. For example, the more times each segment appears in the content contained in a certain content segment, the higher the content correlation between the query statement and the content segment is determined; when each segment appears in the content contained in a certain content segment, The fewer the occurrences in , the lower the content relevance between the query statement and the content fragment.

Similarly, the query statement can be segmented, and the attribute correlation between the query statement and the content segment can be determined based on the number of times each segment appears in the attribute information of a certain content segment. For example, the more times each segment appears in the document name of a certain content fragment, the higher the attribute correlation of the corresponding document name between the query statement and the content fragment is determined; when each segment appears in the document name of a certain content fragment, The fewer the occurrences in the document name, the lower the correlation of the attribute corresponding to the document name between the query statement and the content fragment.

For example, assuming that the query statement is "transformer type" and the attribute information includes the document name and chapter title, the query statement can be segmented to obtain "transformer" and "type", and then according to the content contained in each content fragment, The number of times "transformer" and "type" are used to determine the content correlation between the query statement "transformer type" and the corresponding content fragment, and based on the number of times "transformer" and "type" appear in the document name of the document where each content fragment is located, Determine the attribute correlation of the corresponding document name between the query statement "Transformer Type" and the corresponding content fragment, and determine the query statement "Transformer Type" based on the number of times "Transformer" and "Type" appear in the chapter title corresponding to each content fragment. The attribute correlation between the corresponding chapter title and the corresponding content fragment.

In the embodiment of the present disclosure, a third correlation threshold corresponding to the content correlation and a fourth correlation threshold corresponding to the attribute correlation can be set, so that among multiple content segments, the corresponding content correlation can be greater than the third correlation The degree threshold, and/or the content fragments whose corresponding attribute correlation is greater than the fourth correlation threshold, are determined as multiple candidate content fragments related to the query statement. The third correlation threshold and the fourth correlation threshold can be set as needed, and are not limited here.

Alternatively, the fifth correlation threshold can be set, and the content correlation and attribute correlation can be set to have corresponding weights (the weights can be the same or different), and then the weighted sum can be determined according to the weight corresponding to the content correlation and attribute correlation, and Content fragments whose weighted sum is greater than the fifth relevance threshold are determined as multiple candidate content fragments related to the query statement. Among them, the fifth correlation threshold can be set as needed, and is not limited here.

As a result, multiple candidate content segments that are highly relevant to the query statement can be accurately obtained from all content segments included in all documents.

Step 503: Based on the second correlation model in the NLP field, obtain the second correlation between the query statement and each candidate content segment.

In the embodiment of the present disclosure, the second correlation model can be pre-trained. The input of the second correlation model is the candidate content fragment and the query statement, and the output is the correlation score (ie, confidence) between the candidate content fragment and the query statement. , and then for each candidate content fragment, the query statement and the candidate content fragment can be input into the trained second correlation model, so that the second correlation model determines the candidate content based on the content contained in the query statement and the candidate content fragment. The degree of correlation between the fragment and the query statement is determined, and the second correlation degree is output, so that the answer generation device can obtain the second degree of correlation between the query statement and the candidate content fragment according to the output of the second correlation degree model.

In the embodiment of the present disclosure, for each candidate content segment, the corresponding attribute information can be obtained, and the attribute information and the candidate content segment can be spliced to obtain the corresponding splicing result, and the query statement and the splicing result corresponding to the candidate content segment can be obtained. , input the second correlation model, so that the second correlation model determines the degree of correlation between the candidate content fragment and the query statement based on the query statement and the content and attribute information of the candidate content fragment itself, and outputs the second correlation degree, thereby The answer generating device may obtain the second correlation between the query statement and the candidate content segment according to the output of the second correlation model.

The attribute information of the candidate content fragment may include at least one of the name of the document in which the candidate content fragment is located, the chapter title corresponding to the candidate content fragment, and the parent titles of each level of the chapter title.

Step 504: Obtain target content segments from each candidate content segment based on each second correlation degree.

Therefore, through the second correlation model based on the NLP field, the second correlation between each candidate content segment and the query sentence is determined based on the query sentence, the attribute information of each candidate content segment, and the content contained in the candidate content segment itself. , further improving the accuracy of the identified target content segments.

Step 505: Generate a target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type.

For the specific implementation process and principle of step 505, please refer to the descriptions of other embodiments and will not be described again here.

In addition, before step 502, the following steps 506-508 may also be included:

Step 506: Recognize each document based on the optical character recognition OCR technology in the field of artificial intelligence AI to obtain the recognition results of each document.

In the embodiment of the present disclosure, the answer generation device may use optical character recognition (OCR) technology to recognize each document to obtain the recognition results of each document.

In the embodiment of the present disclosure, the answer generation device can also be connected to the document processing platform through an interface, thereby uploading each document to the document processing platform, so as to recognize each document based on the document processing platform and using optical character recognition OCR technology, and then Obtain the recognition results of each document returned by the document processing platform.

In the embodiment of the present disclosure, the answer generation device can also call the RPA robot to upload each document to the document processing platform. Based on the document processing platform, optical character recognition (OCR) technology is used to identify each document, and then obtain the document returned by the document processing platform. Recognition results for each document. Therefore, when there are a large number of documents to be retrieved, the labor costs required for document uploading can be reduced by calling the RPA robot to upload each document one by one to the document processing platform.

Referring to the left drawing of Figure 6, the document processing platform may provide an interactive interface, which may include an "upload document" button for uploading documents and a "start recognition" button for starting the document recognition process. The answer generation device can call the RPA robot to simulate mouse operations, click the "Upload Document" button on the interactive interface for uploading documents to upload the document to be processed to the document processing platform, and then click the "Upload Document" button on the interactive interface for starting Click the "Start Recognition" button of the document recognition process to start the document recognition process on the document processing platform, and then obtain the document recognition results shown in the right side of Figure 6. Among them, "cl_num" in Figure 6 represents the chapter serial number, "cl_name" represents the chapter title, "cl_rank" represents the row where the chapter is located, and "cl_content" represents the content contained in the chapter.

Step 507: Perform structured processing on each recognition result to obtain multiple content fragments included in each document.

In embodiments of the present disclosure, documents may include text and/or tables. Correspondingly, the document recognition results may include text recognition results and/or table recognition results.

Accordingly, step 507 can be implemented in the following ways: segment the text recognition results and/or table recognition results according to a preset segmentation method to obtain multiple segmented segments; aggregate the multiple segmented segments according to a preset aggregation method, To obtain multiple content segments, each content segment is obtained by aggregating at least one segmented segment.

The preset segmentation method is a method of dividing the recognition result of the document into multiple segmented segments, which can be determined according to the type of content contained in the document (such as text type, table type).

The default aggregation method is a method of aggregating divided fragments to obtain content fragments, which can be determined according to the type of content contained in the document (such as text type, table type).

For example, assume that the document recognition results include text recognition results, and the text recognition results include chapter numbers, commas, periods and other punctuation marks. The answer generation device can perform the first segmentation of the text recognition result based on the chapter number, and then perform the second segmentation on the result of the first segmentation based on punctuation marks (usually period and other end-of-sentence punctuation marks), thereby segmenting the text recognition result. It is a plurality of sentences, each sentence is a segmented segment, and each segmented segment is arranged from front to back according to its corresponding position in the document.

Furthermore, a specific length can be given, such as 200 characters, and then gradually accumulated from the first segmented segment backwards. When the accumulated length is greater than 200 characters, the previously accumulated segmented segments are regarded as one content segment. Use the currently accumulated split segment as the first split segment of the next content segment. For example, when the length of the fifth sentence is 203 characters, and the length of the previously accumulated sentences is 197 characters, the four previously accumulated sentences will be regarded as one content fragment, and the fifth sentence will be used as the next content fragment. The first sentence, and then the subsequent sentences are accumulated to determine the next content fragment.

Referring to Figure 7, by performing structured processing on the text recognition results shown in the left figure, multiple content fragments shown in the right figure of Figure 7 can be obtained.

Or, assume that the recognition results of the document include table recognition results, and the table recognition results include delimiter symbols used to distinguish different cells, and the row numbers where the cells are located. The answer generation device can perform the first segmentation of the table recognition result according to the row number, and then the second segmentation of the first segmentation result according to the delimiter symbol, thereby dividing the table recognition result into multiple cell contents, each cell The content of the grid is a segmented segment, and the segmented segments in each row are arranged from front to back according to their corresponding positions in the document. Furthermore, the divided fragments in each row can be spliced into one content fragment.

Referring to Figure 8, by performing structured processing on the table recognition results shown in the left figure, multiple content fragments shown in the right figure of Figure 8 can be obtained.

It should be noted that the above-mentioned ways of segmenting text recognition results or table recognition results, and the ways of aggregating multiple segmented segments obtained by segmentation are only illustrative descriptions and cannot be understood as limitations to the technical solution of the present disclosure. In practical applications, those skilled in the art can set a preset segmentation method for segmenting the recognition results of the document as needed, and a preset aggregation method for aggregating multiple segmented fragments, and this disclosure does not limit this.

Step 508: Save each content segment in correspondence with the corresponding content field.

In the embodiment of the present disclosure, the name of the content field can be set to "content fragment", and each content fragment can be saved corresponding to the corresponding content field, so that when the content contained in the content fragment needs to be obtained later, the content can be obtained through the content The field obtains the content contained in the corresponding content fragment.

In addition, in the embodiment of the present disclosure, the content contained in each content segment and the document name, chapter title, and parent title at each level corresponding to each content segment can also be saved in the form of a structure. The fields in the structure can include corresponding A field named "Content Fragment", a field named "Document Name", a field named "Chapter Title", and a field named "Level Parent Title".

By using optical character recognition (OCR) technology, each document is recognized to obtain the recognition results of each document, and each recognition result is structured to obtain multiple content fragments included in each document. Each content fragment is compared with the corresponding The content fields are saved correspondingly, which enables the document to be retrieved to be processed and multiple content fragments obtained, which lays the foundation for accurately determining the target content fragment that can answer the user's question from the document, and generating the answer corresponding to the query statement based on the target content fragment. Base. And by calling the RPA robot to upload each document to the document processing platform, based on the document processing platform, OCR technology in the AI field is used to identify each document, and then obtain the recognition results of each document returned by the document processing platform, and then analyze each recognition result Through structured processing, multiple content fragments included in each document are obtained, which realizes the combination of RPA and AI to implement IA to obtain the content fragments in the document, further reducing the labor cost required to generate answers.

In order to implement the above embodiments, the present disclosure also proposes an answer generating device. Figure 9 is a schematic structural diagram of an answer generating device according to the fifth embodiment of the present disclosure.

As shown in Figure 9, the answer generation device 900 includes: a first acquisition module 901, a second acquisition module 902 and a generation module 903.

Among them, the first acquisition module 901 is used to acquire the query statement and the question type to which the query statement belongs;

The second acquisition module 902 is configured to acquire target content segments that match the query statement from multiple content segments included in at least one document;

The generation module 903 is used to generate a target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type.

It should be noted that the answer generation device 900 in the embodiment of the present disclosure can execute the answer generation method provided in the above embodiment. The answer generating device 900 may be implemented by software and/or hardware. The answer generating device may be an electronic device, or may be configured in an electronic device to automatically generate accurate answers to user questions instead of manual work. The electronic device may include but is not limited to a terminal device, a server, etc., and this embodiment does not specifically limit the electronic device.

In one embodiment of the present disclosure, the question type includes one of numerical type, extraction type, and judgment type; the number of target content segments is multiple; the generation module 903 includes:

The first acquisition unit is used to input the query statement and the target content fragment into an extraction model in the field of natural language processing NLP for each target content fragment, so as to extract the candidate answer fragment corresponding to the query statement from the target content fragment, and obtain the corresponding Confidence;

The second acquisition unit is used to acquire the target answer segment from each candidate answer segment according to the confidence level corresponding to each candidate answer segment;

The generation unit is used to generate the target answer based on the target answer fragment according to the response strategy corresponding to the question type.

In one embodiment of the present disclosure, the question type includes an extraction class; a generation unit for:

Use the target answer fragment as the target answer.

In one embodiment of the present disclosure, the question type includes a judgment class; a generation unit for:

Input the target answer fragment and query statement into the judgment model in the NLP field to obtain the judgment result corresponding to the query statement;

Use the judgment result and/or the target answer fragment as the target answer.

In one embodiment of the present disclosure, the question type includes a numeric class; a generation unit for:

According to the first preset rule, obtain the target number from the target answer fragment and obtain the unit corresponding to the target number;

Generate the target answer based on the target number and the corresponding unit.

In one embodiment of the present disclosure, the question type includes a statistical class; the generation module 903 includes:

The extraction unit is used to extract the target answer from the target content fragment according to the second preset rule.

In one embodiment of the present disclosure, the answer generation device 900 may also include:

The third acquisition module is used to acquire target questions matching the query statement from the preset question and answer set;

The fourth acquisition module is used to obtain the first correlation between the query statement and the target question based on the first correlation model in the NLP field;

The first determination module is used to determine that the first correlation degree is not greater than the preset threshold.

The fifth acquisition module is used to obtain the answer corresponding to the target question from the question and answer set when the first correlation degree is greater than the preset threshold;

The second determination module is used to determine the answer corresponding to the target question as the target answer corresponding to the query statement.

In one embodiment of the present disclosure, the second acquisition module 902 includes:

The third acquisition unit is used to query based on the query statement to obtain multiple candidate content fragments related to the query statement from multiple content fragments;

The fourth acquisition unit is used to obtain the second correlation between the query statement and each candidate content fragment based on the second correlation model in the NLP field;

The fifth acquisition unit is used to acquire the target content segment from each candidate content segment based on each second correlation degree.

The recognition module is used to identify each document based on the optical character recognition OCR technology in the field of artificial intelligence to obtain the recognition results of each document;

The processing module is used to perform structured processing on each recognition result to obtain multiple content fragments included in each document;

The saving module is used to save each content fragment correspondingly to the corresponding content field.

In one embodiment of the present disclosure, the identification module includes:

The upload unit is used to call the RPA robot to upload each document to the document processing platform, so that based on the document processing platform, optical character recognition OCR technology can be used to identify each document;

The sixth acquisition unit is used to acquire the recognition results of each document returned by the document processing platform.

It should be noted that the foregoing explanation of the embodiment of the answer generation method also applies to the answer generation device of this embodiment. Unpublished details of the embodiments of the answer generation device of the present disclosure will not be described again here.

In summary, the answer generation device of the embodiment of the present disclosure, after obtaining the query statement and the question type to which the query statement belongs, obtains the target content fragment matching the query statement from multiple content fragments included in at least one document, and then according to the question The response strategy corresponding to the type generates the target answer corresponding to the query statement based on the target content fragment. Therefore, by automatically generating answers instead of manual work, the labor cost and time cost required to generate answers are reduced, and by accurately determining the target content fragment that can answer the user's question from the document, and generating the query statement corresponding to the target content fragment. answers, improving the accuracy of the generated answers.

In order to implement the above embodiments, embodiments of the present disclosure also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, The answer generation method as described in any of the aforementioned method embodiments.

In order to implement the above embodiments, embodiments of the present disclosure also provide a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the answer generation method as described in any of the foregoing method embodiments is implemented. In some embodiments, the computer-readable storage medium is a non-transitory computer-readable storage medium.

In order to implement the above embodiments, embodiments of the present disclosure also provide a computer program product. When the instruction processor in the computer program product is executed, the answer generation method as described in any of the foregoing method embodiments is implemented.

In order to implement the above embodiments, an embodiment of the present disclosure also proposes a computer program. The computer program includes computer program code. When the computer program code is run on a computer, it causes the computer to execute as described in any of the foregoing method embodiments. answer generation method.

Figure 10 illustrates a block diagram of an exemplary electronic device suitable for implementing embodiments of the present disclosure. The electronic device 10 shown in FIG. 10 is only an example and should not bring any limitations to the functions and scope of use of the embodiments of the present disclosure.

As shown in Figure 10, electronic device 10 is embodied in the form of a general computing device. The components of electronic device 10 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and a bus 18 connecting various system components, including memory 28 and processing unit 16.

Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics accelerated port, a processor, or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (hereinafter referred to as: ISA) bus, Micro Channel Architecture (Micro Channel Architecture; hereafter referred to as: MAC) bus, enhanced ISA bus, video electronics Standards Association (Video Electronics Standards Association; hereinafter referred to as: VESA) local bus and Peripheral Component Interconnection (hereinafter referred to as: PCI) bus.

Electronic device 10 typically includes a variety of computer system readable media. These media may be any available media that can be accessed by electronic device 10, including volatile and nonvolatile media, removable and non-removable media.

The memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (Random Access Memory; hereinafter referred to as: RAM) 30 and/or cache memory 32. Electronic device 10 may further include other removable/non-removable, volatile/non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write to non-removable, non-volatile magnetic media (not shown in Figure 10, commonly referred to as a "hard drive"). Although not shown in FIG. 10, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk") and a removable non-volatile optical disk (e.g., a compact disk read-only memory) may be provided. Disc Read Only Memory (hereinafter referred to as: CD-ROM), Digital Video Disc Read Only Memory (hereinafter referred to as: DVD-ROM) or other optical media) read and write optical disc drives. In these cases, each drive may be connected to bus 18 through one or more data media interfaces. Memory 28 may include at least one program product having a set (eg, at least one) of program modules configured to perform the functions of embodiments of the present disclosure.

A program/utility 40 having a set of (at least one) program modules 42, including but not limited to an operating system, one or more application programs, other program modules, and program data, may be stored, for example, in memory 28 , each of these examples or some combination may include the implementation of a network environment. Program modules 42 generally perform functions and/or methods in the embodiments described in this disclosure.

Electronic device 10 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), may also communicate with one or more devices that enable a user to interact with electronic device 10, and/or with Any device (eg, network card, modem, etc.) that enables the electronic device 10 to communicate with one or more other computing devices. This communication may occur through input/output (I/O) interface 22. Moreover, the electronic device 10 can also communicate with one or more networks (such as a local area network (Local Area Network; hereinafter referred to as: LAN), a wide area network (Wide Area Network; hereinafter referred to as: WAN)) and/or a public network, such as the Internet, through the network adapter 20 ) communication. As shown in FIG. 10 , network adapter 20 communicates with other modules of electronic device 10 via bus 18 . It should be understood that, although not shown in Figure 10, other hardware and/or software modules may be used in conjunction with electronic device 10, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tapes drives and data backup storage systems, etc.

The processing unit 16 executes programs stored in the memory 28 to perform various functional applications and data processing, such as implementing the methods mentioned in the previous embodiments.

It should be noted that the foregoing explanations of the embodiments of the answer generation method are also applicable to the electronic devices, computer-readable storage media, computer program products and computer programs of the embodiments of the present disclosure, and will not be described again here.

In the description of this specification, reference to the terms "one embodiment," "some embodiments," "an example," "specific examples," or "some examples" or the like means that specific features are described in connection with the embodiment or example. , structures, materials, or features are included in at least one embodiment or example of the present disclosure. In this specification, the schematic expressions of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, those skilled in the art may combine and combine different embodiments or examples and features of different embodiments or examples described in this specification unless they are inconsistent with each other.

In addition, the terms “first” and “second” are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of indicated technical features. Therefore, features defined as "first" and "second" may explicitly or implicitly include at least one of these features. In the description of the present disclosure, "plurality" means at least two, such as two, three, etc., unless otherwise expressly and specifically limited.

Any process or method descriptions in flowcharts or otherwise described herein may be understood to represent modules, segments, or portions of code that include one or more executable instructions for implementing customized logical functions or steps of the process. , and the scope of the preferred embodiments of the present disclosure includes additional implementations in which functions may be performed out of the order shown or discussed, including in a substantially simultaneous manner or in the reverse order, depending on the functionality involved, which shall It should be understood by those skilled in the art to which embodiments of the present disclosure belong.

The logic and/or steps represented in the flowcharts or otherwise described herein, for example, may be considered a sequenced list of executable instructions for implementing the logical functions, and may be embodied in any computer-readable medium, For use by, or in combination with, instruction execution systems, devices or devices (such as computer-based systems, systems including processors or other systems that can fetch instructions from and execute instructions from the instruction execution system, device or device) or equipment. For the purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer readable media include the following: electrical connections with one or more wires (electronic device), portable computer disk cartridges (magnetic device), random access memory (RAM), Read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium may even be paper or other suitable medium on which the program may be printed, as the paper or other medium may be optically scanned, for example, and subsequently edited, interpreted, or otherwise suitable as necessary. process to obtain the program electronically and then store it in computer memory.

It should be understood that various parts of the present disclosure may be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if it is implemented in hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: discrete logic gate circuits with logic functions for implementing data signals; Logic circuits, application specific integrated circuits with suitable combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

Those of ordinary skill in the art can understand that all or part of the steps involved in implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. The program can be stored in a computer-readable storage medium. When executed, one of the steps of the method embodiment or a combination thereof is included.

In addition, each functional unit in various embodiments of the present disclosure may be integrated into one processing module, each unit may exist physically alone, or two or more units may be integrated into one module. The above integrated modules can be implemented in the form of hardware or software function modules. If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

The storage media mentioned above can be read-only memory, magnetic disks or optical disks, etc. Although the embodiments of the present disclosure have been shown and described above, it can be understood that the above-mentioned embodiments are illustrative and should not be construed as limitations of the present disclosure. Those of ordinary skill in the art can make modifications to the above-mentioned embodiments within the scope of the present disclosure. The embodiments are subject to changes, modifications, substitutions and variations.

Claims

An answer generation method that includes:

Obtain the query statement and the question type to which the query statement belongs;

Obtain a target content fragment that matches the query statement from a plurality of content fragments included in at least one document;

According to the response strategy corresponding to the question type and based on the target content segment, a target answer corresponding to the query statement is generated.
The method according to claim 1, wherein the question type includes one of a numerical type, an extraction type, and a judgment type; the number of the target content segments is multiple;

According to the response strategy corresponding to the question type, based on the target content fragment, the target answer corresponding to the query statement is generated, including:

For each target content segment, input the query statement and the target content segment into an extraction model in the field of natural language processing NLP to extract candidate answer segments corresponding to the query statement from the target content segment, and Get the corresponding confidence level;

Obtain target answer segments from each of the candidate answer segments according to the confidence level corresponding to each of the candidate answer segments;

The target answer is generated based on the target answer fragment according to the response strategy corresponding to the question type.
The method of claim 2, wherein the question type includes an extraction class;

Generating the target answer based on the target answer fragment according to the response strategy corresponding to the question type includes:

Use the target answer fragment as the target answer.
The method according to claim 2, wherein the question type includes a judgment type;

Generating the target answer based on the target answer fragment according to the response strategy corresponding to the question type includes:

Enter the target answer fragment and the query statement into a judgment model in the NLP field to obtain the judgment result corresponding to the query statement;

The judgment result and/or the target answer fragment are used as the target answer.
The method of claim 2, wherein the question type includes a numeric type;

Generating the target answer based on the target answer fragment according to the response strategy corresponding to the question type includes:

According to the first preset rule, obtain the target number from the target answer fragment, and obtain the unit corresponding to the target number;

The target answer is generated based on the target number and the corresponding unit.
The method of claim 1, wherein the question type includes a statistical class;

According to the response strategy corresponding to the question type, based on the target content fragment, the target answer corresponding to the query statement is generated, including:

According to the second preset rule, the target answer is extracted from the target content segment.
The method according to any one of claims 1 to 6, wherein before obtaining the target content fragment matching the query statement from a plurality of content fragments included in at least one document, it further includes:

Obtain target questions matching the query statement from the preset question and answer set;

Based on the first correlation model in the NLP field, obtain the first correlation between the query statement and the target question;

It is determined that the first correlation degree is not greater than a preset threshold.
The method of claim 7, further comprising:

When the first correlation is greater than the preset threshold, obtain the answer corresponding to the target question from the question and answer set;

The answer corresponding to the target question is determined as the target answer corresponding to the query statement.
The method according to any one of claims 1-6, wherein said obtaining a target content segment matching the query statement from a plurality of content segments included in at least one document includes:

Perform a query based on the query statement to obtain a plurality of candidate content fragments related to the query statement from the plurality of content fragments;

Based on the second correlation model in the NLP field, obtain the second correlation between the query statement and each of the candidate content fragments;

Based on each of the second correlations, the target content segment is obtained from each of the candidate content segments.
The method according to any one of claims 1 to 6, wherein before obtaining the target content fragment matching the query statement from a plurality of content fragments included in at least one document, it further includes:

Based on the optical character recognition OCR technology in the field of artificial intelligence, identify each of the documents to obtain the recognition results of each of the documents;

Perform structured processing on each of the recognition results to obtain a plurality of content fragments included in each of the documents;

Each content segment is stored in correspondence with the corresponding content field.
The method according to claim 10, wherein the optical character recognition (OCR) technology based on the field of artificial intelligence (AI) recognizes each of the documents to obtain the recognition results of each of the documents, including:

Call the RPA robot to upload each of the documents to the document processing platform, so as to identify each of the documents based on the document processing platform and using the optical character recognition OCR technology;

Obtain the identification results of each document returned by the document processing platform.
An answer generating device including:

The first acquisition module is used to obtain the query statement and the question type to which the query statement belongs;

a second acquisition module, configured to acquire a target content segment that matches the query statement from a plurality of content segments included in at least one document;

A generation module, configured to generate a target answer corresponding to the query statement based on the target content fragment according to the response strategy corresponding to the question type.
The device according to claim 12, wherein the question type includes one of a numerical type, an extraction type, and a judgment type; the number of the target content segments is multiple;

The generation module includes:

A first acquisition unit configured to input the query statement and the target content fragment into an extraction model in the field of natural language processing NLP for each target content fragment, so as to extract the query statement from the target content fragment. Corresponding candidate answer fragments, and obtain the corresponding confidence;

a second acquisition unit, configured to acquire a target answer segment from each of the candidate answer segments according to the confidence level corresponding to each of the candidate answer segments;

A generating unit configured to generate the target answer based on the target answer fragment according to the response strategy corresponding to the question type.
An electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the implementation as described in any one of claims 1-11 is achieved. Methods.
A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method according to any one of claims 1-11 is implemented.
A computer program product, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.
A computer program, wherein the computer program includes computer program code, which when run on a computer causes the computer to perform the method according to any one of claims 1 to 11.