阅读说明：本技术 一种基于人工智能的心理咨询（会话）系统及其方法 (Psychological consultation (conversation) system and method based on artificial intelligence ) 是由 黄峥 *** 马海刚 于 2020-06-06 设计创作，主要内容包括：本发明公开了一种基于人工智能的心理咨询(会话)系统及其方法,系统包括输入模块、语言分析模块、逻辑树对话模块、语料数据库和小结与反馈模块；输入模块采集用户会话输入；通过使用心理知识图谱模块及多轮对话管理模块对用户会话输入进行意图分析；输入模块和语言分析模块在人机交互中循环迭代,通过多轮对话管理模块管理多轮对话之间的关系；语言分析模块按照智能心理咨询逻辑树的逻辑流程及对用户对话的意图分析,在多轮对话管理模块的管理下组织指导用户进行多轮对话的逻辑走向；结合机器人与用户间的会话信息,反馈对话时用户与心理、情绪问题相关的信息,并输出,制定心理干预方案。本发明通过对话机器人为用户提供心理咨询服务,使用户感觉更像是在和人(咨询师)进行会话,提高用户体验,大大提高了机器人回复的效率。(The invention discloses a psychological consultation (conversation) system and a method thereof based on artificial intelligence, wherein the system comprises an input module, a language analysis module, a logic tree dialogue module, a corpus database and a summary and feedback module; the input module collects user session input; analyzing intention of the user session input by using a mental knowledge map module and a multi-turn dialogue management module; the input module and the language analysis module are iterated circularly in man-machine interaction, and the relationship among multiple rounds of conversations is managed through the multiple rounds of conversation management modules; the language analysis module organizes and guides the logic trend of the multi-turn conversation of the user under the management of the multi-turn conversation management module according to the logic flow of the intelligent psychological consultation logic tree and the intention analysis of the user conversation; and (4) combining the conversation information between the robot and the user, feeding back the information related to the psychological and emotional problems of the user during the conversation, outputting the information, and making a psychological intervention scheme. The invention provides psychological consultation service for the user through the conversation robot, so that the user feels like a conversation with a person (a consultant), the user experience is improved, and the reply efficiency of the robot is greatly improved.)

1. An artificial intelligence based psychological counseling (conversation) system, the system comprising:

the input module is used for collecting user session input;

the language analysis module is connected with the input module and comprises a mental knowledge map module and a multi-round dialogue management module; the multi-turn dialogue management module is used for managing the dialogue between the robot and the user, and the psychological knowledge map module and the multi-turn dialogue management module are used for analyzing the intention of user session input; the input module and the language analysis module are iterated circularly in man-machine interaction, and the relationship among multiple rounds of conversations is managed through the multiple rounds of conversation management modules;

the logic tree dialogue module is connected with the language analysis module and used for guiding multi-turn dialogue to be performed in order, an intelligent psychological consultation logic tree is arranged in the logic tree dialogue module, and the language analysis module organizes and guides a user to perform logic trend of the multi-turn dialogue under the management of the multi-turn dialogue management module according to the logic flow of the intelligent psychological consultation logic tree and the intention analysis of user dialogue;

the corpus database is connected with the language analysis module, is used as the input of the language analysis module and is used for storing psychological consultation session data;

and the summary and feedback module is connected with the language analysis module, and the language analysis module collects information related to psychological and emotional problems generated by the user when the user has a conversation with the robot and outputs the information to be fed back to the user through the summary and feedback module.

2. The artificial intelligence based psychological counseling (conversational) system of claim 1, further comprising a user side chart module for information transmission with the linguistic analysis module and the summary and feedback module, which gets user information and outputs linguistic analysis results through the linguistic analysis module, and serves as input for the multi-turn dialogue management module and the summary and feedback module.

3. The artificial intelligence based psychological counseling (conversation) system of claim 2, wherein the language analysis module comprises a BERT language classification model, any dialog given by the user is input through the trained BERT language classification model, and the key dialog information is obtained by classifying and extracting the dialog input information through the BERT language classification model.

4. The artificial intelligence based psychological counseling (conversational) system of claim 1, wherein the logical tree dialogue module comprises an ingestion interview module and a diagnostic evaluation module;

the ingestion session module is used for completing the session of the basic information with the user;

the diagnostic evaluation module at least comprises 9 modules, namely a personal characteristic module, an overtime module, a body health module, an emotion module, an interpersonal relationship model, a sleep module, a competency module, a coping style module and a resource module, and each of the ingestion session module and the diagnostic evaluation module is respectively connected with the corpus database.

5. The artificial intelligence based psychological counseling (conversational) system of claim 4, wherein the corpus database comprises:

the independent tree corpus is used for self-correlation introduction of the robot;

the ingestion session corpus is used for matching with the ingestion session module and the diagnostic evaluation module, collecting the basic information of the user and completing the user profile writing;

the follow-up corpus is used for following up the topic which the user is talking about at present and encouraging the user to elaborate the key problem;

a query corpus used for further querying details on the current topic of the user talk or clarifying other related key information;

a collaboration dialog corpus, which realizes collaboration dialog with a user through a 4W model (what/why/work/wishi), helps the user to perform backstepping, and takes corresponding action;

a corpus of clauses that are reviewed generally around keywords in a user dialog.

6. The artificial intelligence based psychological counseling (conversational) system of claim 5, wherein the follow-up corpus, the chase corpus, the collaborative dialog corpus, and the golden sentence corpus are used in the logical tree dialog module at a frequency of 1:2:1: 6.

7. A psychological counseling (conversation) method based on artificial intelligence, characterized in that the method comprises the following steps:

s1, giving user session input;

s2, performing language identification and intention analysis on the user session input by using the natural language based on the mental knowledge graph, and extracting key information in the user session input;

s3, according to the key information extracted by the robot, mapping the user session input to the user intention of the priority, and carrying out multiple rounds of logic dialogue with the corresponding intelligent psychological consultation logic tree;

and S4, combining the conversation information between the robot and the user, feeding back information related to psychological and emotional problems of the user during conversation, outputting the information, and making a psychological intervention scheme.

8. The artificial intelligence based psychological counseling (conversational) method of claim 7, wherein in step S2, the intention analysis is performed on the user conversational input through the trained BERT language classification model and the extraction model, the user conversational input is classified into one of the extraction categories after being processed by the multi-classifier, and the key information of the user conversational input is extracted from the extraction categories.

9. The artificial intelligence based psychological counseling (conversation) method according to claim 7, wherein in the step S3, when a plurality of rounds of logical dialogue are performed through the intelligent psychological counseling logical tree, an ingestion conversation is first performed, collecting basic information of the user; and then carrying out diagnostic evaluation, and carrying out multi-turn interactive conversation on 9 subjects of personality characteristics, working conditions, physical health, emotional disturbance, interpersonal relationship, sleeping conditions, working competence conditions, coping strategies and social support of the user, wherein each subject simulates psychological consultation conversation through an embedded corpus database mode.

10. A psychological counseling (conversation) method based on artificial intelligence according to claim 9, wherein the interactive dialogue of 9 subjects is performed to the user as follows:

preferentially identifying psychological keywords in the user session input;

if the recognizable psychological keywords do not exist, identifying the chatting keywords in the user session input;

if the user has no identifiable new keyword in the question responses of the follow-up corpus, the question-pursuing corpus and the cooperation dialogue corpus, randomly selecting a sentence from the golden sentence corpus of the keywords of the previous sentence for response; and if no new keyword can be identified after the user responds, jumping out of the corpus database mode, and returning to the intelligent psychological consulting logic tree.

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a psychological consultation (conversation) system and a psychological consultation (conversation) method based on artificial intelligence.

Background

Currently, the more mature conversation robots in the market mainly include the following three types:

(1) functional robots, such as customer service AI, office assistant AI, etc., mainly implement some set functions, such as ordering, invoicing, returning goods, etc., or appointing a meeting room, providing policy guidance, etc., as an assistant. The technology mainly adopted by the robot is purposefully understood, and the technology such as slot position extraction and the like is realized. However, most task robots have a limited number of dialogue rounds and do not make reasonable use of user portraits and knowledge maps. Most of the robots are also completed by a single task, so that the information of the user cannot be recorded for a long time, and different conversations and push can be carried out according to the information of the user every time.

(2) The robot is chatted. The method mainly collects and marks reply corpora from text libraries such as forums, microblogs, ancient poems and the like, and automatically selects replies with high matching degree from the corpora according to user conversations. Technically, such robots typically use text similarity techniques to find the most similar responses from an already given corpus. The robots generally focus on open-domain conversation, have weak multi-round functions, and often have unreasonable context. Meanwhile, these robots usually do not record user images and cannot perform targeted dialog based on user input.

(3) Other types of chat and entertainment robots. For example, a user sends a group of photos, and the robot can automatically generate a poem through picture recognition. The robot generally fuses a plurality of images and voice technologies, and the dialogue is only one presentation form of the robot, so that the robot is greatly different from the dialogue treatment method adopted by the invention.

The existing conversation robot can be divided into three conversation forms according to the purpose and the form of conversation with a user:

A. task-driven dialog: understanding a task instruction issued by a user, and realizing a corresponding function, such as searching a certain song;

B. question-answer dialogue: answering questions about facts classes, such as how tomorrow is;

C. chat conversation: and performing personalized chatting with emotion with the user in the open domain.

The question-answer dialog of the information retrieval type completed based on the user question is generally in a single turn, and cannot be continuously conversed with the user aiming at one type of subject.

The existing computer technology cannot understand the user's conversation deeply, and when the user narrates the situation or thing of a complex scene, the computer cannot understand; the services provided by the existing conversation robot are mainly focused on an assistant function and a chatting function, and the existing information of the client, such as occupation, work, character and other factors, cannot be fully utilized by the existing algorithm to carry out more efficient psychological consultation.

Disclosure of Invention

Therefore, the invention aims to provide psychological consultation services for users through the conversation robot, so that the users feel more like in conversation with people (consultants), and the user experience is improved. To this end, the present invention provides a psychological counseling (conversation) system based on artificial intelligence and a method thereof.

The adopted technical scheme is as follows:

in one aspect, the present invention provides an artificial intelligence based psychological counseling (conversation) system, the system comprising:

the input module is used for collecting user session input;

the language analysis module is connected with the input module and comprises a mental knowledge map module and a multi-round dialogue management module; the multi-turn dialogue management module is used for managing the dialogue between the robot and the user, and the psychological knowledge map module and the multi-turn dialogue management module are used for analyzing the intention of user session input; the input module and the language analysis module are iterated circularly in man-machine interaction, and the relationship among multiple rounds of conversations is managed through the multiple rounds of conversation management modules;

the logic tree dialogue module is connected with the language analysis module and used for guiding multi-turn dialogue to be performed in order, an intelligent psychological consultation logic tree is arranged in the logic tree dialogue module, and the language analysis module organizes and guides a user to perform logic trend of the multi-turn dialogue under the management of the multi-turn dialogue management module according to the logic flow of the intelligent psychological consultation logic tree and the intention analysis of user dialogue;

the corpus database is connected with the language analysis module, is used as the input of the language analysis module and is used for storing psychological consultation session data;

and the summary and feedback module is connected with the language analysis module, and the language analysis module collects information related to psychological and emotional problems generated by the user when the user has a conversation with the robot and outputs the information to be fed back to the user through the summary and feedback module.

The system also comprises a user side drawing writing module which realizes information transmission with the language analysis module and the summary and feedback module, obtains user information through the language analysis module, outputs a language analysis result and is used as the input of the multi-turn dialogue management module and the summary and feedback module.

The language analysis module comprises a BERT language classification model, any dialog given by a user is input through the trained BERT language classification model, and dialog input information is classified and extracted through the BERT language classification model to obtain key dialog information.

The logic tree dialogue module comprises an ingestion session module and a diagnostic evaluation module;

the ingestion session module is used for completing the session of the basic information with the user;

the diagnostic evaluation module at least comprises 9 modules, namely a personal characteristic module, an overtime module, a body health module, an emotion module, an interpersonal relationship model, a sleep module, a competency module, a coping style module and a resource module, and each of the ingestion session module and the diagnostic evaluation module is respectively connected with the corpus database.

The corpus database includes:

the independent tree corpus is used for self-correlation introduction of the robot;

the ingestion session corpus is used for matching with the ingestion session module and the diagnostic evaluation module, collecting the basic information of the user and completing the user profile writing;

the follow-up corpus is used for following up the topic which the user is talking about at present and encouraging the user to elaborate the key problem;

a query corpus used for further querying details on the current topic of the user talk or clarifying other related key information;

a collaboration dialog corpus, which realizes collaboration dialog with a user through a 4W model (what/why/work/wishi), helps the user to perform backstepping, and takes corresponding action;

a corpus of clauses that are reviewed generally around keywords in a user dialog.

The follow-up corpus, the question-pursuing corpus, the cooperative dialogue corpus and the golden sentence corpus are used in the logic tree dialogue module at a frequency of 1:2:1: 6.

In another aspect, the present invention also provides a psychological counseling (conversation) method based on artificial intelligence, the method comprising the steps of:

s1, giving user session input;

s2, performing language identification and intention analysis on the user session input by using the natural language based on the mental knowledge graph, and extracting key information in the user session input;

s3, according to the key information extracted by the robot, mapping the user session input to the user intention of the priority, and carrying out multiple rounds of logic dialogue with the corresponding intelligent psychological consultation logic tree;

and S4, combining the conversation information between the robot and the user, feeding back information related to psychological and emotional problems of the user during conversation, outputting the information, and making a psychological intervention scheme.

In step S2, the trained BERT language classification model and extraction model are used to analyze the intention of the user session input, the user session input is classified into one of the extraction categories after being processed by the multi-classifier, and the key information of the user session input is extracted from the extraction categories.

In step S3, when performing multiple rounds of logic dialogue through the intelligent psychological consultation logic tree, first performing an ingestion session and collecting basic information of the user; and then carrying out diagnostic evaluation, and carrying out multi-turn interactive conversation on 9 subjects of personality characteristics, working conditions, physical health, emotional disturbance, interpersonal relationship, sleeping conditions, working competence conditions, coping strategies and social support of the user, wherein each subject simulates psychological consultation conversation through an embedded corpus database mode.

When the user is in interactive dialogue with 9 subjects, the following method is carried out:

preferentially identifying psychological keywords in the user session input;

if the recognizable psychological keywords do not exist, identifying the chatting keywords in the user session input;

if the user has no identifiable new keyword in the question responses of the follow-up corpus, the question-pursuing corpus and the cooperation dialogue corpus, randomly selecting a sentence from the golden sentence corpus of the keywords of the previous sentence for response; and if no new keyword can be identified after the user responds, jumping out of the corpus database mode, and returning to the intelligent psychological consulting logic tree.

The technical scheme of the invention has the following advantages:

A. according to the invention, a natural language understanding and multi-turn dialogue technology based on a psychological knowledge map and text classification technology are adopted, so that on one hand, the input of a user is fully understood, and the intention analysis is carried out on the input of the user, and compared with the traditional robot technology, the replying efficiency of the robot is greatly improved.

B. The invention adopts a guided intelligent logic tree dialogue module and adopts (a plurality of) logic trees conforming to psychological consultation dialogue, so that the robot breaks through a single sentence dialogue mode with a user, can carry out a group of associated dialogue, and completes basic starting, ingestion conversation, diagnostic evaluation and intervention introduction of psychological consultation; compared with the prior robot technology, the method greatly enhances the capacity and efficiency of multi-turn conversation, has no limit on the number of conversation turns, and enhances the capacity of understanding the meaning of the user; meanwhile, the conversation theme is clear, and compared with the common chatting robot, the conversation theme has purposiveness and direction feeling, and the conversation theme is not easy to be far away from the field range of psychological consultation.

C. The branch in the logic tree in the invention adopts the BERT language classification model to identify the natural language of the user, thereby breaking through the prior art that the identification is mainly carried out by adopting a tab mode, and the user experience is more intelligent. The traditional tab technology is rigid, and the invention can make the user really feel that the answer is true people instead of a logically simple machine.

D. In the process of carrying out logic tree conversation, the invention collects the information of the user related to psychological and emotional problems, completes the side writing of the user and saves the history of the user characteristics; the user side writing information is used in the subsequent dialogue, and different language material responses are carried out for different users. Therefore, the matching between the corpus database and the user is higher, the user can feel that the robot knows the user and remembers the user, and compared with the traditional robot, the method fully considers the conversation history and the existing answer of the user in each conversation, is favorable for searching the shortest path to guide the user, and enables psychological consultation to be more efficient.

E. The invention defines 4-6 type corpus by adopting corpus database, independent and embedded logic tree dialogue module, simulates psychological consultation dialogue according to psychological consultation method, and enables users to have target, interest, thinking and harvest according to the modes of number of rounds of definition, corpus use priority, corpus extraction frequency and the like.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

FIG. 1 is a diagram of a mental knowledgebase module provided by the present invention;

FIG. 2 is a diagram of the multi-turn dialog state provided by the present invention;

FIG. 3 is an example of a robotic reply;

FIG. 4 is a flow diagram of a logical tree provided by the present invention;

FIG. 5 is an ingestible meeting module provided by the present invention;

FIG. 6 is a personal feature module provided in the present invention;

FIG. 7 is an overtime module provided in the present invention;

FIG. 8 is a body health module provided in the present invention;

FIG. 9 is a mood module provided in the present invention;

FIG. 10 is an interpersonal relationship module provided in the present invention;

FIG. 11 is a sleep module provided in the present invention;

FIG. 12 is a competency module provided in the present invention;

FIG. 13 is a coping style module provided in the present invention;

FIG. 14 is a resource module provided in the present invention;

fig. 15 is a block diagram of a psychological counseling system provided in the present invention.

Detailed Description

The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it should be understood that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 15, the present invention provides an artificial intelligence based psychological counseling (conversation) system, comprising: the system comprises an input module, a language analysis module, a logic tree dialogue module, a corpus database and a summary and feedback module, wherein the input module is used for collecting user session input; the robot asks questions to users according to the established knowledge nodes, the multi-turn dialogue management module is used for managing the dialogue between the robot and the users, and intention analysis is carried out on user conversation input by using the psychological knowledge map module and the multi-turn dialogue management module; the input module and the language analysis module are iterated circularly in man-machine interaction, and the relationship among multiple rounds of conversations is managed through the multiple rounds of conversation management modules; the logic tree dialogue module is connected with the language analysis module and used for guiding multi-turn dialogue to be carried out in order, an intelligent psychological consultation logic tree is arranged in the logic tree dialogue module, and the language analysis module organizes and guides a user to carry out logic trend of the multi-turn dialogue under the management of the multi-turn dialogue management module according to the logic flow of the intelligent psychological consultation logic tree and the intention analysis of user dialogue;

and the corpus database is connected with the language analysis module and used as the input of the language analysis module for storing psychological consultation session data.

The summary and feedback module is connected with the language analysis module, and the language analysis module collects information related to psychological and emotional problems generated by the user when the user is in conversation with the robot and outputs the information to be fed back to the user through the summary and feedback module; the feedback form can be common psychological counseling means such as music treatment, meditation and the like.

Knowledge nodes of a plurality of psychological fields are established in the mental knowledge graph module, as shown in fig. 1, each knowledge node has correlation and is used for the robot to ask questions of the user.

In one example of fig. 1, each node represents a knowledge point (circle node) in the psychological domain, which can be used to ask questions of the user, and can be used to help diagnose psychological problems of the user, such as sleep, time of day, overtime, etc. in the map.

The knowledge nodes have some attribute values (yellow boxes in the graph) which can be classified, the values represent values which the knowledge node can take, and the value classification represents the representative values of the knowledge node and is used for helping judgment, such as age, gender, overtime situation and the like in the graph.

The knowledge nodes are connected by edges, the edges represent the correlation among the knowledge nodes, and represent that the knowledge nodes can jump, and the knowledge nodes are correlated. During the inquiry process, relevant questions can be selected for questioning, and the inquiry continuity is kept. For example, the sleep condition and the insomnia history are related, and in the case of obtaining the current insomnia, the user can further want to know whether the user has a long-term insomnia history.

In order to be able to speak with a user in natural language, the user's conversational input needs to be understood deeply, and the user replies when the robot gives a specific question. The expected answer categories can be obtained based on the mental knowledge maps, and the answers of the users need to be classified on the expected values of the expected mental knowledge maps by utilizing the natural language understanding technology.

The understanding of the user session input here is divided into two parts: understanding the user's intent and finding key slots from the user's input.

Understanding the user's intentions, such as understanding whether the user is anxious, whether psychological anxiety has an impact on the user, is considered a multi-category problem. By using the BERT-based language classification model, some positive samples are marked from the prior psychological consultation to form a user-side sketch map for any expected reply, wherein the user-side sketch map is not only an output result of the language analysis module and user information obtained in the language analysis process, but also can be used as input of the multi-turn dialogue management module, and meanwhile, the user-side sketch map is also input of the feedback and summary module.

For example, "i feel anxious" is a positive sample of "negative effects" in psychological effects. Thus there are many samples for the possible values based on which a multi-class classifier is trained using the BERT language classification model. Based on this classifier, given any input, it can be classified to one of the expected values.

For finding key slots from user session input, the present invention also uses a rule-based extraction model. For example, it is to be understood that the "3 years 2 months" inside the "i am 3 years 2 months" input by the user indicates the duration of the user's job, which represents a time period. This time period will be divided into the category "3 years or more" last. For such problems, the present invention uses a rule-based extraction model. For diseases and time with special relation in psychology, the invention extracts various rules based on regular expression and extracts from sentences. For these successful extractions, we use rules to map strings to extraction categories.

The logic tree dialogue module adopted in the invention comprises an ingestion session module and a diagnostic evaluation module;

the ingestion session module is used for completing basic information session with the user; the diagnostic evaluation module comprises 9 modules, namely a personal characteristic module, an overtime module, a body health module, an emotion module, an interpersonal relationship model, a sleep module, a competency module, a coping style module and a resource module, wherein the linguistic data database is embedded in each of the ingestion session module and the diagnostic evaluation module respectively, and is shown in figures 5-14.

The corpus database adopted in the invention comprises: independent tree corpus, ingestion interview corpus, follow-up corpus, question-chasing corpus, collaborative dialog corpus, and gold sentence corpus

The independent tree corpus is used for self-correlation introduction of the robot;

the ingestion session corpus is used for matching with the ingestion session module and the diagnostic evaluation module, collecting the basic information of the user and completing the user profile writing;

the follow-up corpus is used for following up the topic which the user is talking about at present and encouraging the user to elaborate the key problem;

the query corpus is used for further querying details on the current topic of the user talk or clarifying other related key information;

the collaboration dialogue corpus realizes collaboration dialogue with the user through a 4W model (what/why/work/wishi), helps the user to perform backstepping and takes corresponding action;

a corpus of clauses that surround general comments on keywords in a user's conversation.

Preferably, the follow-up corpus, the question-pursuing corpus, the cooperative dialogue corpus and the golden sentence corpus are used in the logic tree dialogue module at a frequency of 1:2:1: 6.

Psychological consultation is a complex conversation process, and can achieve good diagnosis and treatment effects only by deep question and answer among psychological consultants and users in multiple rounds of interaction. The common single-round conversation can not meet the requirements of psychological consultation at all, and the invention combines a psychological knowledge map module and emphasizes the development of a multi-round conversation technology. The multi-turn dialog comprises two parts: conversation state maintenance and conversation state jumping. The present invention designs and organizes multiple rounds of dialog management in a logical tree fashion, as shown in FIG. 2, which shows a typical dialog jump diagram.

The invention provides a psychological consultation (conversation) method based on artificial intelligence, which comprises the following steps:

(S1) given user input;

(S2) performing language identification and intention analysis on the user session input by using natural language based on a mental knowledge graph, and extracting key information in the user session input;

(S3) mapping user session inputs to prioritized user intentions according to the key information extracted by the robot, and performing multiple rounds of logical dialogue with the corresponding intelligent psychological consultative logic tree;

and (S4) combining the conversation information between the robot and the user, making a user side-writing picture, feeding back the user side-writing picture, and making a psychological intervention scheme.

A typical conversation process is as follows. Given the input of a user, the input is firstly divided into a certain extraction category by using the extraction category and key information in the extraction category is extracted by utilizing a natural language understanding technology based on a mental knowledge graph. For a particular intent and last state, the present invention maps the user's input to a user intent that corresponds to a different dialog logic and execution action. Depending on the specific information and the difference in the execution result, the robot may give different responses, for example, if the robot says differently, it is referred to as a response action here to identify different states. Finally, based on the reply action, the robot may expect different expectations, such as whether the robot expects a response after asking a question. The invention uses the extraction model with priority to express the input sequence expected by the robot. Figure 3 gives an example of how a robot may resolve a particular piece of data, and give a reply.

The present invention defines a dialog state as being based on the state (points in fig. 2) in which all previous dialogs were located, which can be considered a uniquely defined description of all user responses. This dialog state is divided into two parts: now the category of the answers to all questions answered by the user and the question currently being asked, based on this dialog state the invention is able to give the user the next round of interaction, such as the next round of question or question.

The dialog state can be directly jumped, and according to the definition of the mental knowledge map, the invention carries out different jumpers according to the answers of the user. Different answers may lead to different jumps for the same question (lines in fig. 2, multiple lines representing different jump conditions). As shown in fig. 2, a plurality of paths from a point can jump to different points, which correspond to different user input conditions and the user's historical state.

Under the technical support of a multi-turn dialogue management module, the invention adopts a logic tree to organize and guide the logic trend of multi-turn dialogue. In order to enable the robot to provide psychological consultation services for the user, namely enable the robot to perform multiple rounds of interactive conversations with the user in a closed domain aiming at a class of subjects, the invention completes ingestion conversations, diagnostic evaluations and psychological interventions in psychological consultation according to the constructed intelligent psychological consultation logic tree.

The invention can complete two modules of ingestion conversation and diagnosis evaluation in psychological consultation through the logic tree of each module shown in figures 5-14, wherein the topics related to the diagnosis evaluation module comprise 9 types of topics such as personality characteristics, working conditions, physical health, emotional disturbance, interpersonal relationship, sleeping conditions, working competence conditions, coping strategies, social support and the like of a user, and a plurality of rounds of interactive conversations are carried out with the user in a closed domain around the 9 topics through a plurality of classifiers. Each topic simulates a psychological consultation session through an embedded corpus database model. These topics are managed as separate parts, and the order and appearance of items can be adjusted through historical conversation records.

After completing the ingestion interview and the diagnosis evaluation with the user, the current psychological condition of the user needs to be fed back through a summary and a feedback module, so that the user can be helped to understand the user better, and the robot can be helped to write according to different sides of each user to make a next psychotherapeutic intervention scheme. The user-side written content of the present invention is shown in the following table:

information about relevant questions such as user psychology and mood, etc. shown in the table above. The user side writing information is used in the subsequent dialogue, and different language material responses are carried out according to the side writing graphs of different users. In addition, due to the adoption of the multiple classifiers, when the spoken words of the user and the robot relate to the key information related to the user side-writing graph, the system can automatically acquire and acquire the related information, log in the side-writing graph used for recording the user characteristics in the background, and conveniently call the information at any time later. With the increase of the interaction time and times of the user and the robot, the system can record more personalized features of the user, so that the user feels that the robot understands the user more and understands the user more, and the important aim of establishing a good treatment union in consultation is fulfilled.

According to the psychological consultation method, the 4+2 type corpus is defined and established, and psychological consultation conversation is simulated according to the modes of round number definition, corpus use priority, corpus extraction frequency and the like. The independent tree corpus and the ingestion session corpus are mainly used for completing ingestion sessions with users according to the logic tree, collecting basic information of the users and completing side writing. In addition, the invention also sets four types of corpora, namely a follow-up corpus, a question-chasing corpus, a cooperative dialogue corpus and a golden sentence corpus according to the consulting technology commonly used by psychological consultants. On one hand, the four types of corpora can be embedded into an ingestion conversation, a semi-structural conversation in consultation is simulated, so that a user can have personalized conversation with the robot when answering the information required to be collected by the system, and instant feedback of the robot is received, so that the conversation process is efficient and easy; in addition, the four types of corpora are used for performing talking intervention on the user after the user side writing is completed, and different intervention effects in consultation are achieved respectively. Each corpus is specifically introduced as follows:

independent tree corpus: the main role is the robot's relative introduction to itself.

And the ingestion session corpus is matched with the ingestion session and diagnosis evaluation logic tree, collects the basic information of the user and completes the user profile.

The follow-up corpus is used for following the topic which the user is talking about at present, and encourages the user to elaborate the key problem.

And the query corpus is used for further querying details or clarifying other related key information on the topic currently discussed by the user.

And the collaboration dialogue corpus realizes collaboration dialogue with the user through a 4W model (what/why/work/wishi), helps the user to perform backstepping and takes corresponding action.

The golden sentence corpus is used for general comment around key words in user conversation and reflecting general psychology principle or life philosophy, and has the main functions of maintaining the conversation, improving the psychological viscosity of the user and helping the user to obtain the insight of soul growth in the golden sentences.

The four major corpora (follow-up corpus, question-pursuing corpus, collaborative dialog corpus, golden sentence corpus) use the following principles/methods:

A. the corpus database is embedded between each of the logical tree dialog modules, which shows when and what corpus is embedded in the logical tree dialog modules.

B. After the corpus is entered, 3-6 rounds of corpus dialogues are carried out randomly. Round 1 refers to identifying a keyword.

C. The principle of 1 round of corpus dialogue is as follows:

(1) preferentially identifying psychological keywords;

(2) if no psychological keywords can be identified, identifying the chatting keywords;

(3) the use frequency of the four types of corpora is 1:2:1: 6;

(4) if the user has no new recognizable keyword in the responses after following the questions of the three corpora, namely the corpus, the question-pursuing corpus and the cooperative dialogue corpus, a sentence of response is randomly selected from the golden sentence corpus of the keywords of the previous sentence.

(5) And (4) if no new key words can be identified in the response sentence of the user after the step (4), jumping out of the corpus database, and returning to the dialogue tree.

D. During a dialogue of a user, the used sentences in the corpus database are not reused, i.e. each sentence is used at most once.

It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications therefrom are within the scope of the invention.