WO2021000400A1 - 导诊相似问题对生成方法、系统及计算机设备 - Google Patents

导诊相似问题对生成方法、系统及计算机设备 Download PDF

Info

Publication number
WO2021000400A1
WO2021000400A1 PCT/CN2019/102784 CN2019102784W WO2021000400A1 WO 2021000400 A1 WO2021000400 A1 WO 2021000400A1 CN 2019102784 W CN2019102784 W CN 2019102784W WO 2021000400 A1 WO2021000400 A1 WO 2021000400A1
Authority
WO
WIPO (PCT)
Prior art keywords
question
diagnosis
similar
guidance
questions
Prior art date
Application number
PCT/CN2019/102784
Other languages
English (en)
French (fr)
Inventor
黎旭东
林桂
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2021000400A1 publication Critical patent/WO2021000400A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/332Query formulation
    • G06F16/3329Natural language query formulation or dialogue systems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/20ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms

Definitions

  • the embodiments of the present application relate to the field of computer data processing, and in particular, to a method, system, computer device, and non-volatile computer-readable storage medium for generating similar question pairs for diagnosis.
  • Intelligent customer service is an industry-oriented application developed on the basis of large-scale knowledge processing, including: a variety of question answering systems and dialogue robots have emerged as the times require. People can communicate with devices in natural language to obtain The required information also establishes a fast and effective technical means based on natural language for the communication between enterprises and a large number of users, which can greatly reduce the labor costs of enterprises in customer service.
  • the question and answer system usually searches the database for similar questions of the user's current question, and uses the answer corresponding to the similar question as the reply content of the current question.
  • traditional techniques often use database retrieval or manual methods to obtain similar problem pairs.
  • the former has a relatively single data set and the latter has a cumbersome process, which is not conducive to the realization of similar problem pairs in a large number of data sets. Occasionally, it is unable to provide stable and high-quality similar question pairs for the Q&A system.
  • the purpose of the embodiments of the present application is to provide a method, system, computer equipment and non-volatile computer-readable storage medium for generating similar problem pairs for guidance diagnosis, which are used to generate high-quality similar problem pairs for training data, which are effective Improve the processing accuracy of the guidance question and answer model.
  • the embodiment of the present application provides a method for generating a diagnosis-guided similar question pair, which includes the following steps:
  • diagnosis guide question set including a plurality of diagnosis questions
  • a plurality of second diagnosis similar question pairs are generated by generating a confrontation network similar question pair generation model
  • a plurality of third diagnosis similar question pairs are generated by a random similar question pair generation module
  • diagnosis-guided question-pair data set may be input into a diagnosis-guided question-and-answer model for training the diagnosis-guided question-and-answer model.
  • the embodiment of the present application also provides a system for generating a diagnosis similar question pair, including:
  • An obtaining module for obtaining a set of diagnosis guidance questions includes multiple guidance questions;
  • the first generation module is configured to generate multiple first diagnosis similar question pairs through the SOLR system based on the diagnosis guide question set;
  • the second generation module is configured to generate multiple second diagnosis similar question pairs by generating a confrontation network similar question pair generation model based on the diagnosis guidance question set;
  • the third generation module is configured to generate multiple third diagnosis similar question pairs through the random similar question pair generation module based on the diagnosis guide question set;
  • the mixing module is used to mix multiple first guidance similar question pairs, multiple second guidance similar question pairs and multiple third guidance similar question pairs to obtain guidance question pair data sets;
  • diagnosis-guided question-pair data set may be input into a diagnosis-guided question-and-answer model for training the diagnosis-guided question-and-answer model.
  • an embodiment of the present application further provides a computer device, the computer device memory, a processor, and computer-readable instructions stored in the memory and running on the processor, the computer When the readable instructions are executed by the processor, the following steps are implemented:
  • diagnosis guide question set including a plurality of diagnosis questions
  • a plurality of second diagnosis similar question pairs are generated by generating a confrontation network similar question pair generation model
  • a plurality of third diagnosis similar question pairs are generated by a random similar question pair generation module
  • diagnosis-guided question-pair data set may be input into a diagnosis-guided question-and-answer model for training the diagnosis-guided question-and-answer model.
  • embodiments of the present application also provide a non-volatile non-volatile computer-readable storage medium, in which computer-readable instructions are stored ,
  • the computer-readable instructions may be executed by at least one processor, so that the at least one processor executes the following steps:
  • diagnosis guide question set including a plurality of diagnosis questions
  • a plurality of second diagnosis similar question pairs are generated by generating a confrontation network similar question pair generation model
  • a plurality of third diagnosis similar question pairs are generated by a random similar question pair generation module
  • diagnosis-guided question-pair data set may be input into a diagnosis-guided question-and-answer model for training the diagnosis-guided question-and-answer model.
  • the method, system, computer equipment, and non-volatile computer-readable storage medium for generating similar problem pairs for guided diagnosis can generate a model and a random similar problem pair generation module through the SOLR system, generating a confrontation network similar problem pair It is not difficult to understand that multiple first guide similar question pairs and multiple second guide similar question pairs with high-quality features, and multiple third guide similar question pairs with diverse characteristics are obtained.
  • a guide question pair data set composed of one guide diagnosis similar question pair, multiple second guide diagnosis similar question pairs, and multiple third guide diagnosis similar question pairs. It has the characteristics of high quality and diversity, and is a guide question answering model Provide high-quality training data for similar question pairs, effectively improving the processing accuracy of the diagnostic question-and-answer model.
  • FIG. 1 is a schematic flowchart of Embodiment 1 of the method for generating similar question pairs for guided diagnosis of this application.
  • FIG. 2 is a schematic diagram of a specific flow of step S102 in FIG. 1.
  • FIG. 3 is a schematic diagram of a specific flow of step S102C in FIG. 2.
  • Fig. 4 is a schematic diagram of a specific flow of step S104 in Fig. 1.
  • FIG. 5 is a schematic diagram of a specific flow of step S104C in FIG. 4.
  • Fig. 6 is a schematic diagram of the program modules of the second embodiment of the system for generating similar question pairs for diagnosis guidance according to this application.
  • FIG. 7 is a schematic diagram of the hardware structure of the third embodiment of the computer equipment of this application.
  • FIG. 1 shows a flowchart of steps of a method for generating a diagnosis-guided similar question pair in Embodiment 1 of the present application. It can be understood that the flowchart in this method embodiment is not used to limit the order of execution of the steps. details as follows.
  • Step S100 Obtain a diagnosis guide question set, where the diagnosis guide question set includes multiple diagnosis questions.
  • the multiple diagnosis guidance questions can be crawled from the medical question and answer database, or the user's diagnosis guidance questions can be collected through the terminal device.
  • Step S102 based on the diagnosis guide question set, generate a plurality of first diagnosis similar question pairs through the SOLR system.
  • the SOLR system is a search engine whose main functions can include full-text search, hit marking, faceted search, dynamic clustering, database integration, and processing of rich text (such as Word and PDF). It can provide distributed search and index replication, and can calculate the similarity of two sentences or documents through statistical methods based on the TF-IDF clustering method.
  • the step S102 further includes steps S102A to S102D:
  • Step S102A performing word segmentation operations on each diagnosis question to obtain multiple lemmas corresponding to the multiple diagnosis questions, and each lemma includes multiple lemmas extracted from the corresponding diagnosis question;
  • the multiple word sets corresponding to the multiple diagnosis questions are stored in a database
  • Step S102B retrieve a quasi-similar diagnosis guidance question set from the database for each diagnosis question, and the quasi-similar diagnosis guidance question set is the diagnosis guidance question set.
  • R(q im ,d j ) represents the correlation value of each lemma q im of the diagnosis problem i and another diagnosis problem j, and W im is the weight of each lemma in the diagnosis problem i.
  • k 1 and k 2 are adjustment factors
  • q im f im is the frequency of occurrence of lemma q im in the diagnosis problem i
  • f im is the frequency of occurrence of lemma q im in the diagnosis problem j.
  • N is the total number of diagnosis questions
  • n(q im ) is the number of diagnosis questions including the lemma q im
  • W im is equal to IDF(q im ).
  • step S102C a preset rule is used to screen out multiple similar guidance questions from the corresponding quasi-similar guidance questions for each guidance question to obtain multiple first guidance similar question pairs, and each first guidance question
  • the similar question pair includes a guide question and multiple similar guide questions filtered from the quasi-similar guide question set of this guide question.
  • step S102C further includes the following steps:
  • Step S102C1 according to the similarity scores of each quasi-similar guidance question of each quasi-similar guidance question set and the corresponding guidance question, sorting operation of each quasi-similar guidance question of each quasi-similar guidance question set;
  • step S102C2 a corresponding similar diagnosis question set is screened for each diagnosis question according to a preset ratio, and the similar diagnosis question set is a subset of the quasi-similar diagnosis question set corresponding to the corresponding diagnosis question;
  • Step S102C3 according to the multiple guidance questions and the similar guidance question set corresponding to each of the multiple guidance questions, a plurality of first guidance similar question pairs are formed, and each first guidance question is similar
  • the question pair includes multiple similar guidance questions in a set of similar guidance questions including corresponding guidance questions and corresponding guidance questions.
  • Step S104 based on the diagnosis guide question set, generate multiple second diagnosis similar question pairs by generating a confrontation network similar question pair generation model.
  • the generative model for the similar problem of the generative confrontation network includes a generative model and a discriminant model
  • the generative model includes N generative sub-models connected in sequence, and each generative sub-model includes an LSTM module, a Softmax module, and a Markov decision module connected in sequence; the discriminant model includes a CNN model.
  • step S104 further includes the following steps:
  • Step S104A Perform word segmentation operations on each diagnosis question to obtain multiple lemmas corresponding to the multiple diagnosis questions, and each lemma includes multiple lemmas extracted from the corresponding diagnosis question.
  • step S104B each word element is mapped to a corresponding word vector to obtain multiple word vectors corresponding to each diagnosis question.
  • Step S104C input the multiple word vectors corresponding to each diagnosis question to the generation model of generating confrontation network similarity question pairs, and obtain multiple similarities corresponding to each diagnosis question through the generation confrontation network similarity question pair generation model Guidance issues.
  • step S104C further includes the following steps:
  • Step a Map each word element to a corresponding word vector to obtain a word vector matrix corresponding to each diagnosis question.
  • Step b Input multiple word vectors corresponding to each diagnosis question into the generation model in order.
  • Step c Obtain multiple target words through the generation model, and the multiple target words constitute a target sentence.
  • Step c1 when the word vector of one of the word elements is received, input the word vector into the LSTM model;
  • Step c2 Obtain the corresponding output vector from the LSTM module
  • Step c3 input the output vector into the softmax module, and output multiple probabilities corresponding to multiple candidate words through the softmax module, where each probability is used to indicate the confidence level with the corresponding candidate word;
  • Step c4 output multiple probabilities corresponding to multiple candidate words according to the softmax module, select and output one of the target words from the candidate words through a Markov decision model.
  • Step d Input the target sentence and pre-stored diagnosis guidance standard questions into the discrimination model, judge the similarity between the target sentence and each pre-stored diagnosis guidance question, and feed back the similarity degree to the generation model.
  • Step e Adjust the model parameters of the generative model according to the degree of similarity between the target sentence fed back by the discriminant model and each pre-stored diagnosis guidance question, and repeat steps c to e by the generative model after adjusting the parameters to obtain the desired result
  • One or more target sentences of the, the one or more target sentences and the corresponding diagnosis question form a second diagnosis similar question pair.
  • step S104D each diagnosis guide question is mapped with corresponding multiple similar diagnosis questions to obtain a plurality of second diagnosis similar question pairs.
  • Step S106 based on the diagnosis question set, generate a plurality of third diagnosis similar question pairs through a random similar question pair generation module.
  • each guide question a plurality of other questions in the guide question set are randomly matched to form a plurality of third guide similar question pairs, and each third guide similar question pair includes a corresponding question and a corresponding question Multiple other questions with random matching.
  • Step S108 mixing a plurality of first guide similar question pairs, a plurality of second guide similar question pairs, and a plurality of third guide similar question pairs to obtain a guide question pair data set.
  • diagnosis-guided question-pair data set may be input into a diagnosis-guided question-and-answer model for training the diagnosis-guided question-and-answer model. It is not difficult to understand that the above-mentioned diagnosis questions have the characteristics of high quality and diversity in the data set, and provide high-quality similar question pair training data for the diagnosis question and answer model, which effectively improves the processing accuracy of the diagnosis question and answer model.
  • FIG. 6 shows a schematic diagram of the program modules of the second embodiment of the system for generating similar problems in the guidance of this application.
  • the system 20 for generating similar question pairs for diagnosis may include or be divided into one or more program modules, and the one or more program modules are stored in a storage medium and executed by one or more processors.
  • the program module referred to in the embodiment of the present application refers to a series of computer-readable instruction segments that can complete specific functions, and is more suitable than the program itself to describe the execution process of the generation system 20 in the storage medium of the similar diagnosis problem. The following description will specifically introduce the functions of each program module in this embodiment:
  • the acquiring module 200 is configured to acquire a diagnosis guide question set, and the diagnosis guide question set includes a plurality of diagnosis questions.
  • the first generation module 202 is configured to generate a plurality of first diagnosis similar question pairs through the SOLR system based on the diagnosis guide question set.
  • the first generation module 202 is further configured to: perform word segmentation operations on each diagnosis question to obtain multiple lemmas corresponding to the multiple diagnosis questions, and each word The meta set includes multiple word elements extracted from the corresponding diagnosis question; according to the word element set of each diagnosis question, a quasi-similar diagnosis question set is retrieved from the database for each diagnosis question.
  • the quasi-similar guidance question set is a subset of the guidance question set, and the similarity score of each quasi-similar guidance question set in the quasi-similar guidance question set to the corresponding guidance question is greater than a preset score;
  • the preset rule is for each guidance question to filter out multiple similar guidance questions from the corresponding quasi-similar guidance questions, and obtain multiple first guidance similar question pairs, and each first guidance similar question pair includes A guide question and multiple similar guide questions selected from the quasi-similar guide question set of this guide question.
  • the preset rule is used to screen out multiple similar guidance questions from the corresponding quasi-similar guidance questions for each guidance question to obtain multiple first guidance similar question pairs.
  • the steps include: sorting each quasi-similar guidance question in each quasi-similar guidance question set according to the similarity scores of each quasi-similar guidance question and the corresponding guidance question in each quasi-similar guidance question set; Set a ratio to filter out the corresponding similar guide question set for each guide question, and the similar guide question set is a subset of the quasi-similar guide question set corresponding to the corresponding guide question; Describe the set of similar diagnosis questions corresponding to each of the multiple diagnosis questions, forming multiple first diagnosis similar question pairs, and each first diagnosis similar question pair includes the corresponding diagnosis question and the corresponding diagnosis Multiple similar guidance questions in a set of similar guidance questions.
  • the second generation module 204 is configured to generate multiple second diagnosis similar question pairs by generating a confrontation network similar question pair generation model based on the diagnosis guidance question set.
  • the second generation module 204 is further configured to: perform word segmentation operations on each diagnosis question to obtain multiple lemmas corresponding to the multiple diagnosis questions, and each word
  • the meta set includes multiple word elements extracted from the corresponding diagnosis question; each word element is mapped to a corresponding word vector to obtain multiple word vectors corresponding to each diagnosis question; each of the diagnosis questions corresponds to The multiple word vectors of is input into the generation model of the similar problem pair generation of the generation confrontation network, and multiple similar diagnosis problems corresponding to each diagnosis problem are obtained through the generation model of the generation confrontation network similarity problem pairs; The corresponding multiple similar diagnosis guide questions are mapped to obtain multiple second diagnosis similar question pairs.
  • the generative confrontation network similarity problem pair generative model includes a generative model and a discriminant model; the generative model includes N generative sub-models connected in sequence, and each generative sub-model includes a sequence of The connected LSTM module, Softmax module, Markov decision module; the discriminant model includes a CNN model.
  • the multiple word vectors corresponding to each of the diagnosis questions are input into a generation model of generating a confrontation network similarity question pair, and each of the word vectors is obtained through the generation of the confrontation network similarity question pair generation model.
  • the steps of multiple similar diagnosis questions corresponding to the diagnosis question include: step a, map each word element to the corresponding word vector, and obtain the word vector matrix corresponding to each diagnosis question; step b: The multiple word vectors corresponding to the diagnosis questions are sequentially input into the generative model; step c, obtain multiple target words through the generative model, and the multiple target words constitute a target sentence; step d, the target sentence And pre-stored diagnosis guidance standard questions are input into the discriminant model, the degree of similarity between the target sentence and each pre-stored diagnosis guidance question is judged, and the degree of similarity is fed back to the generation model; step e, feedback according to the discriminant model The degree of similarity between the target sentence and each pre-stored diagnosis guidance question is adjusted, the model parameters of the generative model are
  • the third generation module 206 is configured to generate a plurality of third diagnosis similar question pairs through the random similar question pair generation module based on the diagnosis guide question set.
  • the third generation module 206 is further configured to: randomly match a plurality of other questions in the diagnosis question set for each diagnosis question to form a plurality of third diagnosis similar question pairs , Each third guide similar question pair includes multiple other questions randomly matching the corresponding question with the corresponding question.
  • the mixing module 208 is used to mix a plurality of first guide similar question pairs, a plurality of second guide similar question pairs, and a plurality of third guide similar question pairs to obtain a guide question pair data set.
  • diagnosis-guided question-pair data set may be input into a diagnosis-guided question-and-answer model for training the diagnosis-guided question-and-answer model.
  • the computer device 2 is a device that can automatically perform numerical calculation and/or information processing according to pre-set or stored instructions.
  • the computer device 2 may be a PC, a rack server, a blade server, a tower server, or a cabinet server (including an independent server, or a server cluster composed of multiple servers).
  • the computer device 2 at least includes, but is not limited to, a memory 21, a processor 22, a network interface 23, and a generation system 20 for guiding and generating similar question pairs through a system bus that can communicate with each other. among them:
  • the memory 21 includes at least one type of non-volatile computer-readable storage medium.
  • the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), Random access memory (RAM), static random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk Wait.
  • the memory 21 may be an internal storage unit of the computer device 2, such as a hard disk or memory of the computer device 2.
  • the memory 21 may also be an external storage device of the computer device 2, for example, a plug-in hard disk, a smart media card (SMC), and a secure digital (Secure Digital, SD card, Flash Card, etc.
  • the memory 21 may also include both the internal storage unit of the computer device 2 and its external storage device.
  • the memory 21 is generally used to store the operating system and various application software installed in the computer device 2, such as the program code of the generating system 20 for guiding similar problems in the second embodiment.
  • the memory 21 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 22 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments.
  • the processor 22 is generally used to control the overall operation of the computer device 2.
  • the processor 22 is used to run the program code or processing data stored in the memory 21, for example, to run the diagnostic similar question pair generation system 20, so as to implement the guidance similar question pair generation method of the first embodiment.
  • the network interface 23 may include a wireless network interface or a wired network interface, and the network interface 23 is generally used to establish a communication connection between the computer device 2 and other electronic devices.
  • the network interface 23 is used to connect the computer device 2 with an external terminal through a network, and establish a data transmission channel and a communication connection between the computer device 2 and the external terminal.
  • the network may be Intranet, Internet, Global System of Mobile Communication (GSM), Wideband Code Division Multiple Access (WCDMA), 4G network, 5G Network, Bluetooth (Bluetooth), Wi-Fi and other wireless or wired networks.
  • FIG. 7 only shows the computer device 2 with the components 20-23, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.
  • the diagnostic similar question pair generation system 20 stored in the memory 21 may also be divided into one or more program modules, and the one or more program modules are stored in the memory 21, and It is executed by one or more processors (the processor 22 in this embodiment) to complete the application.
  • FIG. 6 shows a schematic diagram of the program modules of the second embodiment of the system for generating similar question pairs based on diagnosis guidance.
  • the generating system 20 based on similar question pairs on diagnosis guidance can be divided into acquisition modules 200, The first generation module 202, the second generation module 204, the third generation module 206, and the mixing module 208.
  • the program module referred to in this application refers to a series of computer-readable instruction segments that can complete specific functions. The specific functions of the program modules 200-208 have been described in detail in the second embodiment, and will not be repeated here.
  • This embodiment also provides a non-volatile computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory ( SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, server, App application mall, etc., on which storage There are computer-readable instructions, and the corresponding functions are realized when the program is executed by the processor.
  • the non-volatile computer-readable storage medium of this embodiment is used to store the diagnosis guidance similar question pair generation system 20, and when executed by the processor, the following steps are implemented:
  • Step 1 Obtain a diagnosis guide question set, and the diagnosis question set includes multiple diagnosis questions.
  • Step 2 Based on the diagnostic guidance question set, multiple first guidance similar question pairs are generated through the SOLR system.
  • Step 3 Based on the diagnosis question set, a plurality of second diagnosis similar question pairs are generated by generating a confrontation network similar question pair generation model.
  • Step 4 Based on the diagnosis guide question set, a plurality of third diagnosis guide similar question pairs are generated through the random similar question pair generation module.
  • Step 5 Mix a plurality of first guide similar question pairs, a plurality of second guide similar question pairs, and a plurality of third guide similar question pairs to obtain a guide question pair data set.
  • diagnosis-guided question-pair data set may be input into a diagnosis-guided question-and-answer model for training the diagnosis-guided question-and-answer model.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Business, Economics & Management (AREA)
  • General Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Primary Health Care (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Epidemiology (AREA)
  • Human Computer Interaction (AREA)
  • Databases & Information Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Instructional Devices (AREA)

Abstract

一种导诊相似问题对生成方法,所述方法包括:获取导诊问题集,所述导诊问题集包括多个导诊问题(S100),通过SOLR系统生成多个第一导诊相似问题对(S102),通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对(S104),通过随机相似问题对生成模块生成多个第三导诊相似问题对(S106);将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集(S108);其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。上述导诊问题对数据集,同时具备高质量和多样性等特点,为导诊问答模型提供高质量的相似问题对训练数据,有效提高导诊问答模型的处理准确度。

Description

导诊相似问题对生成方法、系统及计算机设备
本申请申明2019年07月02日递交的申请号为201910587880.X、名称为“导诊相似问题对生成方法、系统及计算机设备”的中国专利申请的优先权,该中国专利申请的整体内容以参考的方式结合在本申请中。
技术领域
本申请实施例涉及计算机数据处理领域,尤其涉及一种导诊相似问题对生成方法、系统、计算机设备及非易失性计算机可读存储介质。
背景技术
随着电子商务和人工智能技术的发展,智能客服越来越常见。智能客服是在大规模知识处理基础上发展起来的一项面向行业应用的,包括:各种各样的问答系统和对话机器人应运而生,人们可通过以自然语言的方式与设备进行沟通,获取所需要的信息,还为企业与海量用户之间的沟通建立了一种基于自然语言的快捷有效的技术手段,可以大大降低企业在客服方面的人工成本。
问答系统,通常是在数据库中查找用户当前提问问题的相似问题,并将相似问题对应的解答作为当前提问问题的答复内容。而发明人意识到,传统技术对相似问题对的获取方式,多采用数据库检索或者人工的方式,前者数据集较为单一,后者过程繁琐,并不利于大量数据集相似问题对的实现,具备一定偶然性,均无法为问答系统提供稳定且质量高的相似问题对。
发明内容
有鉴于此,本申请实施例的目的是提供一种导诊相似问题对生成方法、系统、计算机设备及非易失性计算机可读存储介质,用于生成高质量的相似问题对训练数据,有效提高导诊问答模型的处理准确度。
为实现上述目的,本申请实施例提供了一种导诊相似问题对生成方法,包括以下步骤:
获取导诊问题集,所述导诊问题集包括多个导诊问题;
基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似 问题对;
基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集;
其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
为实现上述目的,本申请实施例还提供了导诊相似问题对生成系统,包括:
获取模块,用于获取导诊问题集,所述导诊问题集包括多个导诊问题;
第一生成模块,用于基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
第二生成模块,用于基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对;
第三生成模块,用于基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
混合模块,用于将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集;
其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
为实现上述目的,本申请实施例还提供了一种计算机设备,所述计算机设备存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,所述计算机可读指令被处理器执行时实现以下步骤:
获取导诊问题集,所述导诊问题集包括多个导诊问题;
基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对;
基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进 行混合,得到导诊问题对数据集;
其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
为实现上述目的,本申请实施例还提供了一种非易失性非易失性计算机可读存储介质,所述非易失性非易失性计算机可读存储介质内存储有计算机可读指令,所述计算机可读指令可被至少一个处理器所执行,以使所述至少一个处理器执行以下步骤:
获取导诊问题集,所述导诊问题集包括多个导诊问题;
基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对;
基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集;
其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
本申请实施例提供的导诊相似问题对生成方法、系统、计算机设备及非易失性计算机可读存储介质,通过SOLR系统、生成对抗网络相似问题对生成模型和随机相似问题对生成模块,可以分别得到高质量特征的多个第一导诊相似问题对和多个第二导诊相似问题对,以及具有多样性特征的多个第三导诊相似问题对,不难理解,由多个第一导诊相似问题对、多个第二导诊相似问题对以及多个第三导诊相似问题对构成的导诊问题对数据集,同时具备高质量和多样性等特点,为导诊问答模型提供高质量的相似问题对训练数据,有效提高导诊问答模型的处理准确度。
附图说明
图1为本申请导诊相似问题对生成方法实施例一的流程示意图。
图2为图1中步骤S102的具体流程示意图。
图3为图2中步骤S102C的具体流程示意图。
图4为图1中步骤S104的具体流程示意图。
图5为图4中步骤S104C的具体流程示意图。
图6为本申请导诊相似问题对生成系统实施例二的程序模块示意图。
图7为本申请计算机设备实施例三的硬件结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处所描述的具体实施例仅用以解释本申请,并不用于限定本申请。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
需要说明的是,在本申请中涉及“第一”、“第二”等的描述仅用于描述目的,而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此,限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外,各个实施例之间的技术方案可以相互结合,但是必须是以本领域普通技术人员能够实现为基础,当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在,也不在本申请要求的保护范围之内。
以下实施例将以计算机设备2为执行主体进行示例性描述。
实施例一
参阅图1,示出了本申请实施例一之导诊相似问题对生成方法的步骤流程图。可以理解,本方法实施例中的流程图不用于对执行步骤的顺序进行限定。具体如下。
步骤S100,获取导诊问题集,所述导诊问题集包括多个导诊问题。
示例性的,可以从医疗问答数据库中爬取所述多个导诊问题,或者通过终端设备收集用户的导诊问题等。
步骤S102,基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对。
SOLR系统,为一种搜索引擎,主要功能可以包括全文检索、命中标示、分面搜索、动态聚类、数据库集成,以及富文本(如Word、PDF)的处理。其可以提供分布式搜索和索引复制,以及可以基于TF-IDF聚类方法通过统计学的方法计算两个句子或者文档的相似度。
在示例性的实施例中,如图2所示,所述步骤S102进一步包括步骤S102A~S102D:
步骤S102A,将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;
示例性的,所述多个导诊问题对应的多个词元集被存储到数据库中;
步骤S102B,根据每个导诊问题的词元集,为所述每个导诊问题从所述数据库中分别检索一个准相似导诊问题集,所述准相似导诊问题集为所述导诊问题集的子集,所述准相似导诊问题集中的各个准相似导诊问题与对应导诊问题的相似分数大于一预设分数;
Figure PCTCN2019102784-appb-000001
R(q im,d j)表示导诊问题i的每个词元q im与另一导诊问题j的相关度值,W im为导诊问题i中各个词元的权重。
Figure PCTCN2019102784-appb-000002
k 1,k 2是调节因子,q imf im是词元q im在导诊问题i中的出现频率,f im是词元q im在导诊问题j中的出现频率。
Figure PCTCN2019102784-appb-000003
N是导诊问题总数量,n(q im)是包括词元q im的导诊问题数量,W im等于IDF(q im)。
步骤S102C,以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对,每个第一导诊相似问题对包括一个导诊问题以及从这个导诊问题的准相似导诊问题集中筛选出的多个相似导诊问题。
在示例性的实施例中,如图3所示,所述步骤S102C进一步包括以下步骤:
步骤S102C1,根据各个准相似导诊问题集的各个准相似导诊问题与对应导诊问题的相似分数,对各个准相似导诊问题集的各个准相似导诊问题进行排序操作;
步骤S102C2,根据预设比例为每个导诊问题筛选出对应的相似导诊问题集,所述相似导诊问题集为相应导诊问题对应的准相似导诊问题集的子集;
步骤S102C3,根据多个导诊问题和所述多个导诊问题中的每个导诊问题对应的相似导诊问题集,形成多个第一导诊相似问题对,每个第一导诊相似问题对包括相应导诊问题与相应导诊问题的相似导诊问题集中的多个相似导诊问题。
步骤S104,基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第 二导诊相似问题对。
所述生成对抗网络相似问题对生成模型包括生成模型和判别模型;
所述生成模型包括依顺序串接的N个生成子模型,每个生成子模型包括依顺序串接的LSTM模块、Softmax模块、马尔科夫决策模块;所述判别模型包括CNN模型。
在示例性的实施例中,如图4所示,所述步骤S104进一步包括以下步骤:
步骤S104A,将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元。
步骤S104B,将每个词元映射为相应的词向量,得到每个导诊问题对应的多个词向量。
步骤S104C,将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应多个相似导诊问题。
在示例性的实施例中,如图5所示,所述步骤S104C进一步包括以下步骤:
步骤a,将每个词元映射为相应的词向量,得到每个导诊问题对应的词向量矩阵。
步骤b,将每个导诊问题对应的多个词向量依顺序输入到所述生成模型中。
步骤c,通过所述生成模型得到多个目标词,该多个目标词构成一个目标句。
示例性的,以其中一个生成子模型为例:
步骤c1,当接收到其中一个词元的词向量时,将该词向量输入到LSTM模型中;
步骤c2,由LSTM模块得到相应的输出向量;
步骤c3,将该输出向量输入到softmax模块中,通过softmax模块输出多个待选词对应的多个概率,其中,每个概率用于表示与相应待选词的置信度;
步骤c4,根据softmax模块输出多个待选词对应的多个概率,通过马尔科夫决策模型从所述待选词中选择并输出其中一个目标词。
步骤d,将所述目标句和预存导诊标准问题输入到所述判别模型中,判断所述目标句与各个预存导诊问题之间的相似程度,并将相似程度反馈给生成模型。
步骤e,根据所述判别模型反馈的目标句与各个预存导诊问题之间的相似程度,调整生成模型的模型参数,并藉由调整参数后的生成模型重复执行步骤c~e以得到符合预期的一个或多个目标句,所述一个或多个目标句与相应导诊问题形成一个第二导诊相似问题对。
步骤S104D,将每个导诊问题与对应的多个相似导诊问题进行映射,以得到多个第二导诊相似问题对。
步骤S106,基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对。
示例性的,为每个导诊问题随机匹配所述导诊问题集中的多个其他问题,形成多个第三导诊相似问题对,每个第三导诊相似问题对包括相应问题与相应问题随机匹配的多个其他问题。
步骤S108,将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集。
其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。不难理解,上述导诊问题对数据集同时具备高质量和多样性等特点,为导诊问答模型提供高质量的相似问题对训练数据,有效提高导诊问答模型的处理准确度。
实施例二
请继续参阅图6,示出了本申请导诊相似问题对生成系统实施例二的程序模块示意图。在本实施例中,导诊相似问题对生成系统20可以包括或被分割成一个或多个程序模块,一个或者多个程序模块被存储于存储介质中,并由一个或多个处理器所执行,以完成本申请,并可实现上述导诊相似问题对生成方法。本申请实施例所称的程序模块是指能够完成特定功能的一系列计算机可读指令段,比程序本身更适合于描述导诊相似问题对生成系统20在存储介质中的执行过程。以下描述将具体介绍本实施例各程序模块的功能:
获取模块200,用于获取导诊问题集,所述导诊问题集包括多个导诊问题。
第一生成模块202,用于基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对。
在示例性的实施例中,所述第一生成模块202,还用于:将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;根据每个导诊问题的词元集,为所述每个导诊问题从所述数据库中分别检索得到一个准相似导诊问题集,所述准相似导诊问题集为所述导诊问题集的子集,所述准相 似导诊问题集中的各个准相似导诊问题与对应导诊问题的相似分数大于一预设分数;以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对,每个第一导诊相似问题对包括一个导诊问题以及从这个导诊问题的准相似导诊问题集中筛选出的多个相似导诊问题。
在示例性的实施例中,所述以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对的步骤,包括:根据各个准相似导诊问题集的各个准相似导诊问题与对应导诊问题的相似分数,对各个准相似导诊问题集的各个准相似导诊问题进行排序操作;根据预设比例为每个导诊问题筛选出对应的相似导诊问题集,所述相似导诊问题集为相应导诊问题对应的准相似导诊问题集的子集;根据多个导诊问题和所述多个导诊问题中的每个导诊问题对应的相似导诊问题集,形成多个第一导诊相似问题对,每个第一导诊相似问题对包括相应导诊问题与相应导诊问题的相似导诊问题集中的多个相似导诊问题。
第二生成模块204,用于基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对。
在示例性的实施例中,所述第二生成模块204,还用于:将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;将每个词元映射为相应的词向量,得到每个导诊问题对应的多个词向量;将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题;将每个导诊问题与对应的多个相似导诊问题进行映射,以得到多个第二导诊相似问题对。
在示例性的实施例中,所述生成对抗网络相似问题对生成模型包括生成模型和判别模型;所述生成模型包括依顺序串接的N个生成子模型,每个生成子模型包括依顺序串接的LSTM模块、Softmax模块、马尔科夫决策模块;所述判别模型包括CNN模型。
在示例性的实施例中,所述将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题的步骤,包括:步骤a,将每个词元映射为相应的词向量,得到每个导诊问题对应的词向量矩阵;步骤b,将每个导诊问题对应的多个词向量依顺序输入到 所述生成模型中;步骤c,通过所述生成模型得到多个目标词,该多个目标词构成一个目标句;步骤d,将所述目标句和预存导诊标准问题输入到所述判别模型中,判断所述目标句与各个预存导诊问题之间的相似程度,并将相似程度反馈给生成模型;步骤e,根据所述判别模型反馈的目标句与各个预存导诊问题之间的相似程度,调整生成模型的模型参数,并藉由调整参数后的生成模型重复执行步骤c~e以得到符合预期的一个或多个目标句,所述一个或多个目标句与相应导诊问题形成一个第二导诊相似问题对。
第三生成模块206,用于基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对。
在示例性的实施例中,所述第三生成模块206,还用于:为每个导诊问题随机匹配所述导诊问题集中的多个其他问题,形成多个第三导诊相似问题对,每个第三导诊相似问题对包括相应问题与相应问题随机匹配的多个其他问题。
混合模块208,用于将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集。
其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
实施例三
参阅图7,是本申请实施例三之计算机设备的硬件架构示意图。本实施例中,所述计算机设备2是一种能够按照事先设定或者存储的指令,自动进行数值计算和/或信息处理的设备。该计算机设备2可以是PC、机架式服务器、刀片式服务器、塔式服务器或机柜式服务器(包括独立的服务器,或者多个服务器所组成的服务器集群)等。如图所示,所述计算机设备2至少包括,但不限于,可通过系统总线相互通信连接存储器21、处理器22、网络接口23、以及导诊相似问题对生成系统20。其中:
本实施例中,存储器21至少包括一种类型的非易失性计算机可读存储介质,所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。 在一些实施例中,存储器21可以是计算机设备2的内部存储单元,例如该计算机设备2的硬盘或内存。在另一些实施例中,存储器21也可以是计算机设备2的外部存储设备,例如该计算机设备20上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。当然,存储器21还可以既包括计算机设备2的内部存储单元也包括其外部存储设备。本实施例中,存储器21通常用于存储安装于计算机设备2的操作系统和各类应用软件,例如实施例二的导诊相似问题对生成系统20的程序代码等。此外,存储器21还可以用于暂时地存储已经输出或者将要输出的各类数据。
处理器22在一些实施例中可以是中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器22通常用于控制计算机设备2的总体操作。本实施例中,处理器22用于运行存储器21中存储的程序代码或者处理数据,例如运行导诊相似问题对生成系统20,以实现实施例一的导诊相似问题对生成方法。
所述网络接口23可包括无线网络接口或有线网络接口,该网络接口23通常用于在所述计算机设备2与其他电子装置之间建立通信连接。例如,所述网络接口23用于通过网络将所述计算机设备2与外部终端相连,在所述计算机设备2与外部终端之间的建立数据传输通道和通信连接等。所述网络可以是企业内部网(Intranet)、互联网(Internet)、全球移动通讯系统(Global System of Mobile communication,GSM)、宽带码分多址(Wideband Code Division Multiple Access,WCDMA)、4G网络、5G网络、蓝牙(Bluetooth)、Wi-Fi等无线或有线网络。
需要指出的是,图7仅示出了具有部件20-23的计算机设备2,但是应理解的是,并不要求实施所有示出的部件,可以替代的实施更多或者更少的部件。
在本实施例中,存储于存储器21中的所述导诊相似问题对生成系统20还可以被分割为一个或者多个程序模块,所述一个或者多个程序模块被存储于存储器21中,并由一个或多个处理器(本实施例为处理器22)所执行,以完成本申请。
例如,图6示出了所述实现导诊相似问题对生成系统20实施例二的程序模块示意图,该实施例中,所述基于导诊相似问题对生成系统20可以被划分为获取模块200、第一生成模块202、第二生成模块204、第三生成模块206和混合模块208。其中,本申请所称的程序模块是指能够完成特定功能的一系列计算机可读指令段。所述程序模块200-208的具体 功能在实施例二中已有详细描述,在此不再赘述。
实施例四
本实施例还提供一种非易失性计算机可读存储介质,如闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘、服务器、App应用商城等等,其上存储有计算机可读指令,程序被处理器执行时实现相应功能。本实施例的非易失性计算机可读存储介质用于存储导诊相似问题对生成系统20,被处理器执行时实现如下步骤:
步骤一:获取导诊问题集,所述导诊问题集包括多个导诊问题。
步骤二:基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对。
步骤三:基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对。
步骤四:基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对。
步骤五:将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集。
其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种导诊相似问题对生成方法,所述方法包括:
    获取导诊问题集,所述导诊问题集包括多个导诊问题;
    基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
    基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对;
    基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
    将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集;
    其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
  2. 根据权利要求1所述的导诊相似问题对生成方法,所述基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对的步骤,包括:
    将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;
    根据每个导诊问题的词元集,为所述每个导诊问题从所述数据库中分别检索得到一个准相似导诊问题集,所述准相似导诊问题集为所述导诊问题集的子集,所述准相似导诊问题集中的各个准相似导诊问题与对应导诊问题的相似分数大于一预设分数;
    以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对,每个第一导诊相似问题对包括一个导诊问题以及从这个导诊问题的准相似导诊问题集中筛选出的多个相似导诊问题。
  3. 根据权利要求2所述的导诊相似问题对生成方法,所述以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对的步骤,包括:
    根据各个准相似导诊问题集的各个准相似导诊问题与对应导诊问题的相似分数,对各个准相似导诊问题集的各个准相似导诊问题进行排序操作;
    根据预设比例为每个导诊问题筛选出对应的相似导诊问题集,所述相似导诊问题集为 相应导诊问题对应的准相似导诊问题集的子集;
    根据多个导诊问题和所述多个导诊问题中的每个导诊问题对应的相似导诊问题集,形成多个第一导诊相似问题对,每个第一导诊相似问题对包括相应导诊问题与相应导诊问题的相似导诊问题集中的多个相似导诊问题。
  4. 根据权利要求1所述的导诊相似问题对生成方法,所述基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对的步骤,包括:
    将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;
    将每个词元映射为相应的词向量,得到每个导诊问题对应的多个词向量;
    将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题;
    将每个导诊问题与对应的多个相似导诊问题进行映射,以得到多个第二导诊相似问题对。
  5. 根据权利要求4所述的导诊相似问题对生成方法,所述生成对抗网络相似问题对生成模型包括生成模型和判别模型;
    所述生成模型包括依顺序串接的N个生成子模型,每个生成子模型包括依顺序串接的LSTM模块、Softmax模块、马尔科夫决策模块;所述判别模型包括CNN模型。
  6. 根据权利要求5所述的导诊相似问题对生成方法,所述将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题的步骤,包括:
    步骤a,将每个词元映射为相应的词向量,得到每个导诊问题对应的词向量矩阵;
    步骤b,将每个导诊问题对应的多个词向量依顺序输入到所述生成模型中;
    步骤c,通过所述生成模型得到多个目标词,该多个目标词构成一个目标句;
    步骤d,将所述目标句和预存导诊标准问题输入到所述判别模型中,判断所述目标句与各个预存导诊问题之间的相似程度,并将相似程度反馈给生成模型;
    步骤e,根据所述判别模型反馈的目标句与各个预存导诊问题之间的相似程度,调整生成模型的模型参数,并藉由调整参数后的生成模型重复执行步骤c~e以得到符合预期的 一个或多个目标句,所述一个或多个目标句与相应导诊问题形成一个第二导诊相似问题对。
  7. 根据权利要求1所述的导诊相似问题对生成方法,其特征在于,所述基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对的步骤,包括:
    为每个导诊问题随机匹配所述导诊问题集中的多个其他问题,形成多个第三导诊相似问题对,每个第三导诊相似问题对包括相应问题与相应问题随机匹配的多个其他问题。
  8. 一种导诊相似问题对生成系统,包括:
    获取模块,用于获取导诊问题集,所述导诊问题集包括多个导诊问题;
    第一生成模块,用于基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
    第二生成模块,用于基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对;
    第三生成模块,用于基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
    混合模块,用于将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集;
    其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
  9. 一种计算机设备,所述计算机设备存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,所述计算机可读指令被处理器执行时实现以下步骤:
    获取导诊问题集,所述导诊问题集包括多个导诊问题;
    基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
    基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对;
    基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
    将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集;
    其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答 模型。
  10. 根据权利要求9所述的计算机设备,所述基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对的步骤,包括:
    将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;
    根据每个导诊问题的词元集,为所述每个导诊问题从所述数据库中分别检索得到一个准相似导诊问题集,所述准相似导诊问题集为所述导诊问题集的子集,所述准相似导诊问题集中的各个准相似导诊问题与对应导诊问题的相似分数大于一预设分数;
    以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对,每个第一导诊相似问题对包括一个导诊问题以及从这个导诊问题的准相似导诊问题集中筛选出的多个相似导诊问题。
  11. 根据权利要求10所述的导计算机设备,所述以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对的步骤,包括:
    根据各个准相似导诊问题集的各个准相似导诊问题与对应导诊问题的相似分数,对各个准相似导诊问题集的各个准相似导诊问题进行排序操作;
    根据预设比例为每个导诊问题筛选出对应的相似导诊问题集,所述相似导诊问题集为相应导诊问题对应的准相似导诊问题集的子集;
    根据多个导诊问题和所述多个导诊问题中的每个导诊问题对应的相似导诊问题集,形成多个第一导诊相似问题对,每个第一导诊相似问题对包括相应导诊问题与相应导诊问题的相似导诊问题集中的多个相似导诊问题。
  12. 根据权利要求9所述的计算机设备,所述基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对的步骤,包括:
    将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;
    将每个词元映射为相应的词向量,得到每个导诊问题对应的多个词向量;
    将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中, 通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题;
    将每个导诊问题与对应的多个相似导诊问题进行映射,以得到多个第二导诊相似问题对。
  13. 根据权利要求12所述的计算机设备,所述生成对抗网络相似问题对生成模型包括生成模型和判别模型;
    所述生成模型包括依顺序串接的N个生成子模型,每个生成子模型包括依顺序串接的LSTM模块、Softmax模块、马尔科夫决策模块;所述判别模型包括CNN模型。
  14. 根据权利要求13所述的计算机设备,其特征在于,所述将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题的步骤,包括:
    步骤a,将每个词元映射为相应的词向量,得到每个导诊问题对应的词向量矩阵;
    步骤b,将每个导诊问题对应的多个词向量依顺序输入到所述生成模型中;
    步骤c,通过所述生成模型得到多个目标词,该多个目标词构成一个目标句;
    步骤d,将所述目标句和预存导诊标准问题输入到所述判别模型中,判断所述目标句与各个预存导诊问题之间的相似程度,并将相似程度反馈给生成模型;
    步骤e,根据所述判别模型反馈的目标句与各个预存导诊问题之间的相似程度,调整生成模型的模型参数,并藉由调整参数后的生成模型重复执行步骤c~e以得到符合预期的一个或多个目标句,所述一个或多个目标句与相应导诊问题形成一个第二导诊相似问题对。
  15. 根据权利要求9所述的计算机设备,所述基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对的步骤,包括:
    为每个导诊问题随机匹配所述导诊问题集中的多个其他问题,形成多个第三导诊相似问题对,每个第三导诊相似问题对包括相应问题与相应问题随机匹配的多个其他问题。
  16. 一种非易失性计算机可读存储介质,所述非易失性计算机可读存储介质内存储有计算机可读指令,所述计算机可读指令可被至少一个处理器所执行,以使所述至少一个处理器执行如下步骤:
    获取导诊问题集,所述导诊问题集包括多个导诊问题;
    基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对;
    基于所述导诊问题集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似 问题对;
    基于所述导诊问题集,通过随机相似问题对生成模块生成多个第三导诊相似问题对;
    将多个第一导诊相似问题对、多个第二导诊相似问题对和多个第三导诊相似问题对进行混合,得到导诊问题对数据集;
    其中,所述导诊问题对数据集可以被输入到导诊问答模型中,用于训练所述导诊问答模型。
  17. 根据权利要求16所述的非易失性计算机可读存储介质,所述基于所述导诊问题集,通过SOLR系统生成多个第一导诊相似问题对的步骤,包括:
    将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;
    根据每个导诊问题的词元集,为所述每个导诊问题从所述数据库中分别检索得到一个准相似导诊问题集,所述准相似导诊问题集为所述导诊问题集的子集,所述准相似导诊问题集中的各个准相似导诊问题与对应导诊问题的相似分数大于一预设分数;
    以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对,每个第一导诊相似问题对包括一个导诊问题以及从这个导诊问题的准相似导诊问题集中筛选出的多个相似导诊问题。
  18. 根据权利要求17所述的非易失性计算机可读存储介质,所述以预设规则为每个导诊问题从与之对应的准相似导诊问题集中筛选出多个相似导诊问题,得到多个第一导诊相似问题对的步骤,包括:
    根据各个准相似导诊问题集的各个准相似导诊问题与对应导诊问题的相似分数,对各个准相似导诊问题集的各个准相似导诊问题进行排序操作;
    根据预设比例为每个导诊问题筛选出对应的相似导诊问题集,所述相似导诊问题集为相应导诊问题对应的准相似导诊问题集的子集;
    根据多个导诊问题和所述多个导诊问题中的每个导诊问题对应的相似导诊问题集,形成多个第一导诊相似问题对,每个第一导诊相似问题对包括相应导诊问题与相应导诊问题的相似导诊问题集中的多个相似导诊问题。
  19. 根据权利要求16所述的非易失性计算机可读存储介质,所述基于所述导诊问题 集,通过生成对抗网络相似问题对生成模型生成多个第二导诊相似问题对的步骤,包括:
    将每个导诊问题进行分词操作,以得到所述多个导诊问题对应的多个词元集,每个词元集包括从相应导诊问题中提取的多个词元;
    将每个词元映射为相应的词向量,得到每个导诊问题对应的多个词向量;
    将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题;
    将每个导诊问题与对应的多个相似导诊问题进行映射,以得到多个第二导诊相似问题对。
  20. 根据权利要求19所述的非易失性计算机可读存储介质,所述生成对抗网络相似问题对生成模型包括生成模型和判别模型;
    所述生成模型包括依顺序串接的N个生成子模型,每个生成子模型包括依顺序串接的LSTM模块、Softmax模块、马尔科夫决策模块;所述判别模型包括CNN模型;
    所述将所述每个导诊问题对应的多个词向量输入到生成对抗网络相似问题对生成模型中,通过所述生成对抗网络相似问题对生成模型得到每个导诊问题对应的多个相似导诊问题的步骤,包括:
    步骤a,将每个词元映射为相应的词向量,得到每个导诊问题对应的词向量矩阵;
    步骤b,将每个导诊问题对应的多个词向量依顺序输入到所述生成模型中;
    步骤c,通过所述生成模型得到多个目标词,该多个目标词构成一个目标句;
    步骤d,将所述目标句和预存导诊标准问题输入到所述判别模型中,判断所述目标句与各个预存导诊问题之间的相似程度,并将相似程度反馈给生成模型;
    步骤e,根据所述判别模型反馈的目标句与各个预存导诊问题之间的相似程度,调整生成模型的模型参数,并藉由调整参数后的生成模型重复执行步骤c~e以得到符合预期的一个或多个目标句,所述一个或多个目标句与相应导诊问题形成一个第二导诊相似问题对。
PCT/CN2019/102784 2019-07-02 2019-08-27 导诊相似问题对生成方法、系统及计算机设备 WO2021000400A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910587880.X 2019-07-02
CN201910587880.XA CN110502620B (zh) 2019-07-02 2019-07-02 导诊相似问题对生成方法、系统及计算机设备

Publications (1)

Publication Number Publication Date
WO2021000400A1 true WO2021000400A1 (zh) 2021-01-07

Family

ID=68586179

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/102784 WO2021000400A1 (zh) 2019-07-02 2019-08-27 导诊相似问题对生成方法、系统及计算机设备

Country Status (2)

Country Link
CN (1) CN110502620B (zh)
WO (1) WO2021000400A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113270181A (zh) * 2021-06-22 2021-08-17 中国平安人寿保险股份有限公司 指标数据的区分方法、装置、设备及存储介质
CN113270181B (zh) * 2021-06-22 2024-06-07 中国平安人寿保险股份有限公司 指标数据的区分方法、装置、设备及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112417096B (zh) * 2020-11-17 2024-05-28 平安科技(深圳)有限公司 问答对匹配方法、装置、电子设备及存储介质
CN114969291B (zh) * 2022-05-31 2023-08-08 湖南工商大学 一种自动问答方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127702A (zh) * 2016-06-17 2016-11-16 兰州理工大学 一种基于深度学习的图像去雾算法
CN106789595A (zh) * 2017-01-17 2017-05-31 北京诸葛找房信息技术有限公司 信息推送方法和装置
CN108509463A (zh) * 2017-02-28 2018-09-07 华为技术有限公司 一种问题的应答方法及装置
CN109271505A (zh) * 2018-11-12 2019-01-25 深圳智能思创科技有限公司 一种基于问题答案对的问答系统实现方法
CN109857850A (zh) * 2019-01-18 2019-06-07 深圳壹账通智能科技有限公司 咨询请求处理方法、装置、计算机设备及存储介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10104027B2 (en) * 2015-04-16 2018-10-16 The Board Of Trustees Of The Leland Stanford Junior University Systems and methods for inquiry-based learning including collaborative question generation
CN108509411B (zh) * 2017-10-10 2021-05-11 腾讯科技(深圳)有限公司 语义分析方法和装置
CN108345585A (zh) * 2018-01-11 2018-07-31 浙江大学 一种基于深度学习的自动问答方法
CN109460462B (zh) * 2018-11-15 2021-10-19 中通天鸿(北京)通信科技股份有限公司 一种中文相似问题生成系统与方法
CN109783617B (zh) * 2018-12-11 2024-01-26 平安科技(深圳)有限公司 用于答复问题的模型训练方法、装置、设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127702A (zh) * 2016-06-17 2016-11-16 兰州理工大学 一种基于深度学习的图像去雾算法
CN106789595A (zh) * 2017-01-17 2017-05-31 北京诸葛找房信息技术有限公司 信息推送方法和装置
CN108509463A (zh) * 2017-02-28 2018-09-07 华为技术有限公司 一种问题的应答方法及装置
CN109271505A (zh) * 2018-11-12 2019-01-25 深圳智能思创科技有限公司 一种基于问题答案对的问答系统实现方法
CN109857850A (zh) * 2019-01-18 2019-06-07 深圳壹账通智能科技有限公司 咨询请求处理方法、装置、计算机设备及存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113270181A (zh) * 2021-06-22 2021-08-17 中国平安人寿保险股份有限公司 指标数据的区分方法、装置、设备及存储介质
CN113270181B (zh) * 2021-06-22 2024-06-07 中国平安人寿保险股份有限公司 指标数据的区分方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN110502620B (zh) 2023-05-30
CN110502620A (zh) 2019-11-26

Similar Documents

Publication Publication Date Title
CN110162593B (zh) 一种搜索结果处理、相似度模型训练方法及装置
US11537820B2 (en) Method and system for generating and correcting classification models
US11017178B2 (en) Methods, devices, and systems for constructing intelligent knowledge base
CN108804641B (zh) 一种文本相似度的计算方法、装置、设备和存储介质
WO2019105432A1 (zh) 文本推荐方法、装置及电子设备
US11100124B2 (en) Systems and methods for similarity and context measures for trademark and service mark analysis and repository searches
WO2020119063A1 (zh) 专家知识推荐方法、装置、计算机设备及存储介质
US11521713B2 (en) System and method for generating clinical trial protocol design document with selection of patient and investigator
CN105183833B (zh) 一种基于用户模型的微博文本推荐方法及其推荐装置
US10593422B2 (en) Interaction network inference from vector representation of words
US9183285B1 (en) Data clustering system and methods
US20130304730A1 (en) Automated answers to online questions
WO2022048363A1 (zh) 网站分类方法、装置、计算机设备及存储介质
CN109241243B (zh) 候选文档排序方法及装置
WO2021213160A1 (zh) 基于图神经网络的医疗查询方法、装置、计算机设备及存储介质
WO2018056423A1 (ja) シナリオパッセージ分類器、シナリオ分類器、及びそのためのコンピュータプログラム
CN112632261A (zh) 智能问答方法、装置、设备及存储介质
CN110990533A (zh) 确定查询文本所对应标准文本的方法及装置
WO2022222942A1 (zh) 问答记录生成方法、装置、电子设备及存储介质
WO2019201024A1 (zh) 用于更新模型参数的方法、装置、设备和存储介质
US11501547B1 (en) Leveraging text profiles to select and configure models for use with textual datasets
US11379527B2 (en) Sibling search queries
WO2021000400A1 (zh) 导诊相似问题对生成方法、系统及计算机设备
WO2016122575A1 (en) Product, operating system and topic based recommendations
WO2021012958A1 (zh) 原创文本甄别方法、装置、设备与计算机可读存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19936373

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19936373

Country of ref document: EP

Kind code of ref document: A1