CN112395399A

CN112395399A - Specific personality dialogue robot training method based on artificial intelligence

Info

Publication number: CN112395399A
Application number: CN202011270983.2A
Authority: CN
Inventors: 杨帅; 卫静雯; 田而慷; 殷莉; 王双文; 蒋泓杰; 周嘉茗; 王玥
Original assignee: Sichuan University
Current assignee: Sichuan University
Priority date: 2020-11-13
Filing date: 2020-11-13
Publication date: 2021-02-23

Abstract

The invention belongs to the field of artificial intelligence, and relates to a training robot system and a training robot method with specific personality and conversation functions. The virtual robot corresponds to 5 factors in the five-big personality factor model: extroversion, concordance, conscientity, nervousness, patency. Based on the psychological sample example, the language sample of each personality mental patient is processed by intelligent natural language, and a language prediction model with a specific personality is established by using a deep learning technology. And then extracting a candidate answer group from a retrieval document set obtained by retrieval according to the user question by adopting an IF-IDF (intermediate frequency-inverse discrete frequency) calculation method, selecting standard answers from the candidate answer group by adopting a correlation similarity calculation method, and feeding the standard answers back to the user, thereby providing a direct and interactive virtual robot with a specific personality for the teaching and simulated training of psychology.

Description

Specific personality dialogue robot training method based on artificial intelligence

Technical Field

The invention belongs to the field of artificial intelligence, and relates to a training robot system and a training robot method with specific personality and conversation functions.

Background

Psychiatry, an important subject in clinical medicine, relates to wide scope, strong specialization and has its own particularity. The diagnosis of psychotic disorders follows mainly the thinking approach of "symptom-syndrome-diagnosis", relying on the identification and comprehensive judgment of psychotic symptoms, without corresponding signs or laboratory evidence, and therefore requiring that the psychiatrist must have a solid theoretical basis and clinical skills, as well as a rich practical experience. Among them, the cultivation of mental skills is important. The mental examination is to sample and evaluate information of mental disorder individuals by a psychiatrist through language communication and observation, and is divided into free conversation and inquiry, wherein the communication capacity is the key. However, domestic common medical students and some psychiatrists rarely pay attention to training of communication skills, and meanwhile, the aspects of privacy, social influence, personal safety and the like of patients are considered, so that some difficulties exist in practice teaching, and the opportunity of direct communication with the patients is low. In addition, the intelligent social robot in China starts late, although some intelligent social robots have voice and face recognition capabilities and simple emotional modalities, most of the current conversation functions are limited to a certain topic field, and the intelligent social robot does not have anthropomorphic personality under a specific scene, and has little effect on clinical teaching and training.

The "big five" personality factor model, namely the basic structure of the personality, is composed of the five big factors, and is called as "big five". In the last 10 years, the study of five-factor models has made dramatic progress, with stability verified in a large number of studies from both old specials questionnaires and other person evaluations, various samples of lexical studies and questionnaire measurements, as well as different cultural backgrounds and different analytical methods, which have been considered by many psychologists as the best paradigm of personality architecture. According to McCrae & Costa et al (1985), the five major factors that make up the personality are respectively, extroversion, concordance, conscientiveness, nervousness, patency. While extroversion represents enthusiasm, confidence, vitality, and also has the characteristics of happiness and sociality, these manifestations of the insiders are not prominent, but not equal to self-centering and lack of energy. The concordance indicates that he is friendly, lovely and lovely, people with high scores are willing to assist people, reliable and rich in the same mood, and pay attention to cooperation without emphasizing competition. Conscientiousness represents restriction and stringency, and is related to achievements motivation and organizational planning, also known as "achievements will" or "working" latitude. Those who feel worries frequently and whose emotions are easy to fluctuate get high scores in measurement of the nervousness, and the low scores of people mostly show calm self-adaptation and are not easy to have extreme and bad emotional reactions. Studies have shown that the "five-factor" model is valuable for diagnosing clinical disorders and treating psychological diseases, as well as being useful for predicting and identifying healthy behaviors and problems.

Disclosure of Invention

The invention aims to provide a more accurate and intelligent interactive virtual robot for psychological discipline teaching. The technical scheme of the invention is as follows:

an intelligent question-answering system with a specific personality, comprising:

the database establishing module is used for establishing a psychological domain database, and a large number of mental patient language example samples are stored in the psychological domain database;

the problem acquisition module is used for acquiring user problems;

the question processing module is used for carrying out natural language processing on the user question to obtain query expression about the user question;

the candidate answer extraction module is used for calculating question-answer pair similarity by adopting an IF-IDF calculation method based on the retrieval document set;

and the standard answer obtaining module extracts the most relevant answer according to the candidate answer, and the answer feedback module feeds the standard answer back to the user.

The question processing module comprises a word segmentation sub-module, a word segmentation sub-module and a question processing sub-module, wherein the word segmentation sub-module is used for segmenting a user question and obtaining a group of terms of the user question;

each term in the term set is subjected to named entity recognition, semantic disambiguation and synonym expansion to obtain a term set;

and judging that the user question is complete according to the intention analysis result, if so, combining the terms in the processing term set into a query expression, and if not, requesting the user to supplement the user question, and supplementing the user question after the user question is supplemented.

The standard answer obtaining module optionally comprises a candidate answer correlation calculating submodule for calculating the occurrence frequency of each noun phrase in the candidate answers in the medical field database, obtaining the correlation of each noun phrase in the candidate answer set, and calculating the occurrence frequency of the noun phrase in the query expression in the medical field database by using the query expression correlation calculating submodule to obtain the correlation expressed in the query;

calculating the similarity of the candidate answer relevance and the relevance represented in the query by using a calculation sub-module; candidate answers greater than the second set of similarity values are selected and combined into a standard answer to the user's question.

According to the specific embodiment of the invention, the invention discloses the following technical effects:

the invention discloses a virtual robot training method and a virtual robot training system, wherein based on a psychological sample example, an IF-IDF calculation method is adopted to extract a candidate answer group from a retrieval document set obtained by retrieval according to user problems, a correlation similarity calculation method is adopted to select standard answers from the candidate answer group, and the standard answers are fed back to a user, so that a direct and interactive virtual robot with a specific personality is provided for the teaching of psychology.

Drawings

FIG. 1: a component module of an intelligent question-answering system with a specific personality.

FIG. 2: a training method and a process for a virtual robot are provided.

Detailed Description

S1: a receiving module receives a problem sent by a client through an HTTP request, wherein the problem is described by characters;

s2: the preprocessing module removes special symbols, punctuation marks and stop words in the problem to obtain a processed problem q;

s3: the intention module classifies the question q through a question intention classification model based on BERT to obtain the category Cq of the input question;

s4: the retrieval module firstly refers to a domain dictionary to divide an input question q into w1 and w2 … wi … wn, wherein wi is the ith word which is divided, and a query formula of (w 1U w 2U … U … wn) n Cq is adopted to retrieve a corresponding question and a corresponding answer from a knowledge base (the knowledge base is used for storing a problem set and a corresponding answer set in the domain, and the problem set and the corresponding answer set in the domain are stored in the knowledge base in an index establishing mode), so that a candidate question set and a corresponding answer set are obtained;

s5: the sequencing module sequences the candidate question set Q and the corresponding answer set A by adopting a TF-IDF statistical method to obtain a set Score corresponding to Q and A, wherein a Score threshold Scoremin is set, if Scorein is greater than Scoremin, corresponding Qi and Ai are stored, and the rest question sets and the corresponding answer sets are RQ and RA; s6: the matching algorithm module takes out the problems in the remaining problem set RQ, establishes a matching pair (RQI, q) with the input problem q, inputs the matching pair into a problem similarity matching model based on BERT to obtain a matching probability pi, obtains the probability maximum value pmax of the problems in the RQ through statistics, sets a probability threshold pmin, determines that the RQI is the problem required by the symbol if pmax > pmin, otherwise, does not have the problems meeting the requirements in the remaining problem set; s7: and the Answer configuration module outputs RAi corresponding to RQI or a Special Answer which does not meet the requirement to the client.

S31: and collecting related problem sets Q and corresponding classification sets C in the field to obtain one-to-one corresponding Qi and Ci.

Claims

1. A virtual robot training method and a system are characterized by comprising the following steps: generating a virtual robot that simulates a particular personality of mental patient, the virtual robot corresponding to 5 factors in a "five-large" personality factor model: extroversion, concordance, conscientity, nervousness, patency.

2. The virtual robot training method and system according to claim 1, further comprising: based on the example, the intelligent natural language processing is carried out on the language sample of each personality mental patient, a language prediction model with a specific personality is built by using a deep learning technology, and an intelligent dialogue system with the specific personality is built according to training.

3. The virtual robot training method and system according to claim 1, wherein quantitative clustering is performed on the personality traits of each psychiatric patient's sample.

4. The virtual robot training method and system according to claim 2, wherein: the probability of each answer is selected for each mental patient for each personality forecast, and the dialog of the maximum probability is output.