CN116898441B

CN116898441B - Character testing method and device based on man-machine conversation and electronic equipment

Info

Publication number: CN116898441B
Application number: CN202310861533.8A
Authority: CN
Inventors: 杨家铭; 郑叔亮; 李文珏
Original assignee: Beijing Lingxin Intelligent Technology Co ltd
Current assignee: Beijing Lingxin Intelligent Technology Co ltd
Priority date: 2022-08-25
Filing date: 2023-07-13
Publication date: 2024-03-22
Anticipated expiration: 2043-07-13
Also published as: CN115101151A; CN116898441A

Abstract

The embodiment of the application provides a character testing method and device based on man-machine conversation and electronic equipment. The method comprises the following steps: dialogue data meeting preset test conditions between a target user and the chat robot are obtained; extracting semantic tendency feature vectors from the dialogue data, wherein the semantic tendency feature vectors represent habitual expressions implicitly existing in the dialogue data; acquiring a scale test result of a target user; extracting character bias characteristic vectors from the test results of the scale, wherein the character bias characteristic vectors represent character bias implicitly existing in the test results of the scale; and inputting the character bias feature vector and the semantic tendency feature vector into the test model to obtain character test results of the target user. Because the semantic tendency feature vector and the character bias feature vector are hidden features embedded into the original data, the hidden features are fused and input into the test model, so that the output result of the prediction model can be optimized, and the accuracy of character test is improved.

Description

Character testing method and device based on man-machine conversation and electronic equipment

Technical Field

The embodiment of the application relates to the technical field of computers, in particular to a personality test method and device based on man-machine conversation and electronic equipment.

Background

In the related art, a character testing tool generally performs character testing on a tested person by pushing test questions to the tested person, and then based on the response of the tested person to the test questions.

Specifically, the test questions pushed to the testee generally include a series of preset questions, and each question corresponds to a plurality of answers. Wherein, each answer under each question corresponds to representing a character representation, so that the character test result of the tested person can be evaluated through the answer selected by the tested person. However, the test questions are single in form, the test process is boring and tedious, the test experience is poor, the response condition of the tested person is easily affected, and the accuracy of the test result is reduced.

Therefore, a new solution is needed to be designed for evaluating the character of the tested person, improving the evaluation experience of the tested person and improving the accuracy of the test result.

Disclosure of Invention

The embodiment of the application provides an improved character test method, device and electronic equipment based on man-machine conversation, which are used for realizing character test of a target user, optimizing character test experience and improving accuracy of character test.

The embodiment of the application expects to provide a character testing method and device based on man-machine conversation and electronic equipment.

In a first aspect of the present application, a personality testing method based on a human-machine conversation is provided, including:

dialogue data meeting preset test conditions between a target user and the chat robot are obtained;

extracting semantic tendency feature vectors from the dialogue data; wherein the semantic-predisposing feature vector characterizes a habitual expression implicitly present in the dialog data; the semantic-predisposing feature vector includes at least one of: information density, sentence complexity, questioning habits, self-disclosure degree, emotion polarity change, and dialogue synchronism;

acquiring a scale test result of the target user; the scale test result is obtained based on psychology scale feedback;

extracting character bias characteristic vectors from the scale test results; the character bias characteristic vector characterizes character bias implicitly existing in the test result of the scale; the character biasing feature vector includes at least one of: exogenously, preferably humanized, responsible, neuro-cytoplasmic and openness;

and inputting the character bias characteristic vector and the semantic tendency characteristic vector into a test model to obtain a character test result of the target user.

In a second aspect of the present application, there is provided a personality testing apparatus based on a human-machine conversation, the apparatus comprising:

the receiving and transmitting module is used for acquiring dialogue data between the target user and the chat robot, wherein the dialogue data meets preset test conditions;

a processing module for extracting semantic tendency feature vectors from the dialogue data; wherein the semantic-predisposing feature vector characterizes a habitual expression implicitly present in the dialog data; the semantic-predisposing feature vector includes at least one of: information density, sentence complexity, questioning habits, self-disclosure degree, emotion polarity change, and dialogue synchronism;

the receiving and transmitting module is also used for acquiring the meter test result of the target user; the scale test result is obtained based on psychology scale feedback;

the processing module is also used for extracting character bias characteristic vectors from the scale test results; the character bias characteristic vector characterizes character bias implicitly existing in the test result of the scale; the character biasing feature vector includes at least one of: exogenously, preferably humanized, responsible, neuro-cytoplasmic and openness;

And the processing module is used for inputting the character bias characteristic vector and the semantic tendency characteristic vector into a test model to obtain character test results of the target user.

In a third aspect of the present application, there is provided a computer readable storage medium comprising instructions which, when run on a computer, cause the computer to perform the character testing method based on man-machine conversation as described in the fourth aspect.

In a fourth aspect of the present application, there is provided a computing device configured to: the system comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the character testing method based on man-machine conversation in the fourth aspect when executing the computer program.

In the technical scheme provided by the embodiment of the application, dialogue data meeting preset test conditions between the target user and the chat robot, for example, dialogue data of more than 40 rounds, is obtained. Further, semantic-predisposing feature vectors are extracted from the dialogue data by a machine learning model, wherein the semantic-predisposing feature vectors characterize a habitual expression implicitly present in the dialogue data. Illustratively, the semantic-predisposing feature vector includes at least one of: information density, sentence complexity, questioning habits, self-disclosure level, emotion polarity change, dialog synchronicity. Meanwhile, the scale test result of the target user can be obtained, and the scale test result is obtained based on psychology scale feedback. Further, character bias feature vectors are extracted from the scale test results. The character bias feature vector characterizes character bias implicitly present in the test results of the scale, e.g., the character bias feature vector includes at least one of: exogenously, humanity, responsibility, neuro, openness. And finally, inputting the character bias feature vector and the semantic bias feature vector into a test model to obtain character test results of the target user.

In the embodiment of the application, on one hand, character test is carried out on a user through a dialogue form, so that the interestingness of the character test is increased, and the problems of single form, boring and boring test process and poor test experience of test questions are solved. On the other hand, compared with the scheme that in the related art, the binding relation between the scale answer and the character is relied on to simply calculate the test conclusion, the semantic tendency feature vector and the character bias feature vector adopted in the embodiment of the application are hidden features embedded into the original data, so that the two hidden features are fused and input into the test model, a character test result with pertinence and accuracy can be obtained, the output result of the prediction model is optimized, and the accuracy of the character test is improved.

Drawings

The above, as well as additional purposes, features, and advantages of exemplary embodiments of the present application will become readily apparent from the following detailed description when read in conjunction with the accompanying drawings. Several embodiments of the present application are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:

FIG. 1 schematically illustrates a flow diagram of a personality testing method based on human-machine interactive communications in accordance with the present application;

FIG. 2 schematically illustrates a schematic diagram of a predictive model training method according to the present application;

FIG. 3 schematically illustrates a schematic structural diagram of a personality testing apparatus based on human-machine interaction according to the present application;

FIG. 4 schematically illustrates a structural schematic diagram of a computing device according to the present application;

fig. 5 schematically shows a schematic structural diagram of a server according to the present application.

In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

Detailed Description

The principles and spirit of the present application will be described below with reference to several exemplary embodiments. It should be understood that these examples are given solely to enable those skilled in the art to better understand and practice the present application and are not intended to limit the scope of the present application in any way. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

Those skilled in the art will appreciate that embodiments of the present application may be implemented as a system, apparatus, device, system, or computer program product. Accordingly, the present disclosure may be embodied in the following forms, namely: complete hardware, complete software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.

Therefore, a new solution is needed to be designed for realizing the personality test of the target user, optimizing the personality test experience, and improving the accuracy of the personality test.

In order to overcome the technical problems, according to the embodiments of the present application, a personality test method, a personality test device and an electronic device based on man-machine interaction are provided.

The applicant finds that compared with the scheme of simply calculating the test conclusion by means of the binding relation between the table answers and the characters in the related art, the technical scheme of the embodiment of the application not only combines the dialogue data between the user and the chat robot, but also extracts the hidden features embedded into the dialogue data and the table test result, and takes the hidden features as the input of the test model, thereby obtaining the character test result with pertinence and accuracy, and greatly improving the accuracy of the character test.

Specifically, dialogue data between the target user and the chat robot satisfying a preset test condition, for example, dialogue data of more than 40 rounds, is acquired. Further, semantic-predisposing feature vectors are extracted from the dialogue data by a machine learning model, wherein the semantic-predisposing feature vectors characterize a habitual expression implicitly present in the dialogue data. Illustratively, the semantic-predisposing feature vector includes at least one of: information density, sentence complexity, questioning habits, self-disclosure level, emotion polarity change, dialog synchronicity. Meanwhile, the scale test result of the target user can be obtained, and the scale test result is obtained based on psychology scale feedback. Further, character bias feature vectors are extracted from the scale test results. The character bias feature vector characterizes character bias implicitly present in the test results of the scale, e.g., the character bias feature vector includes at least one of: exogenously, humanity, responsibility, neuro, openness. And finally, inputting the character bias feature vector and the semantic bias feature vector into a test model to obtain character test results of the target user.

According to the technical scheme, on one hand, character tests are conducted on users in a dialogue mode, interestingness of the character tests is increased, and the problems that the test questions are single in mode, tedious in test process and poor in test experience are solved. On the other hand, compared with the scheme that in the related art, the binding relation between the scale answer and the character is relied on to simply calculate the test conclusion, the semantic tendency feature vector and the character bias feature vector adopted in the embodiment of the application are hidden features embedded into the original data, so that the two hidden features are fused and input into the test model, a character test result with pertinence and accuracy can be obtained, the output result of the prediction model is optimized, and the accuracy of the character test is improved.

As an alternative embodiment, the number of character testing devices based on man-machine interaction may be one or more. The personality test device based on the man-machine conversation can be deployed in a test system of the man-machine conversation, and can be deployed in other forms in application programs used in various psychological tests or personality test scenes, and the personality test device based on the man-machine conversation is not limited. For example, the personality testing apparatus based on the human-machine interaction may be provided in processing devices of various devices (e.g., terminal devices, servers).

Any number of elements in the figures are for illustration and not limitation, and any naming is used for distinction only, and not for any limiting sense.

A method for personality testing based on human-machine interaction according to an exemplary embodiment of the present application is described below with reference to fig. 1 in conjunction with a specific application scenario. It should be noted that the above application scenario is only shown for the convenience of understanding the spirit and principles of the present application, and embodiments of the present application are not limited in this respect. Rather, embodiments of the present application may be applied to any scenario where applicable.

The following describes the execution of the character test method based on man-machine conversation with reference to the following examples. Fig. 1 is a flowchart of a personality testing method based on man-machine interaction according to an embodiment of the present invention. The method is applied to a processing module in a character test scene, and it is worth noting that the character test scene can be a dialogue scene taking character test as a theme or other theme dialogue scenes. In the dialogue scene of other topics, the character test result can be used as a part of output information in other topic dialogues, can be used as a generation basis of other output dialogues in other topic dialogues, and can be used as other forms of information to participate in the dialogues, and is not limited herein. As shown in fig. 1, the character testing method based on man-machine conversation includes the following steps:

Step 101, dialogue data meeting preset test conditions between a target user and the chat robot is obtained.

In the embodiment of the application, the chat robot may be disposed in a server or a local terminal of the dialogue system, so as to perform real-time dialogue with the target user. Of course, chat robots may also be plug-ins in instant chat tools, web pages, and other various forms of applications. Further alternatively, the dialogue system may be an online learning architecture system, so that the feature extraction model and the prediction model in the dialogue system can be learned and dynamically updated in real time, and are continuously adapted to newly-added data and various character test scenes, so that the performance of the feature extraction model and the prediction model is improved.

Since the dialogue between the target user and the chat robot also comprises a part of other contents which are irrelevant to the character test, such as call, function introduction, inquiry of basic information, problem retrieval and the like, a screening mechanism needs to be set so as to select dialogue data for subsequent predictive analysis from the dialogue contents, or a guarantee mechanism needs to be set so as to ensure that the dialogue data contains enough effective dialogue contents which can be used for subsequent predictive analysis.

Specifically, preset test conditions may be set so as to screen dialogue data for character test from the dialogue between the target user and the chat robot. In practical applications, the preset test conditions include, but are not limited to: the dialogue response times between the target user and the chat robot exceed the preset response times, the dialogue response times between the target user and the chat robot in the set time period reach the preset response frequency, and the dialogue response of the target user comprises preset keywords.

As an alternative embodiment, it is assumed that the preset test condition is that the number of conversational replies between the target user and the chat robot exceeds the preset number of replies. Since the dialog exceeding the preset number of replies usually contains a sufficient amount of valid dialog content, dialog data meeting this condition can be used for the subsequent predictive analysis. Based on the above assumption, in step 101, the obtaining of dialogue data between the target user and the chat robot that satisfies the preset test condition may be implemented as:

receiving real-time dialogue data between a target user and the chat robot, wherein the real-time dialogue data carries a dialogue identifier; storing the real-time dialogue data into a message queue according to dialogue identification, and recording the response times between a target user and the chat robot; and detecting that the response times between the target user and the chat robot exceeds preset response times, and storing the real-time dialogue data into a persistent storage system.

For example, assume that the preset test condition is that the number of conversational answers between the target user and the chat robot exceeds 40 times (i.e., the preset number of answers). Based on this, a chat data queue is established in the conversation system for storing real-time conversation data between the target user and the chat robot. After detecting that the target user starts to talk to the chat robot, real-time conversation data between the target user and the chat robot is received. And storing the real-time dialogue data into a corresponding message queue according to the dialogue identifier carried in the real-time dialogue data, and recording the real-time response times between the target user and the chat robot. If the number of real-time replies reaches 40, it can be ensured that the real-time dialogue data contains a sufficient amount of valid dialogue content, so that it can be ensured that the real-time dialogue data can be used for executing effective analysis later, in which case the real-time dialogue data can be stored in a persistent storage system.

Step 102, extracting semantic tendentiousness feature vectors from the dialogue data.

Wherein the semantic-predisposing feature vector characterizes a habitual expression implicitly present in the dialog data. The semantic tendency feature vector is feature information implicitly existing in the original dialogue data and is used for describing more personalized expression habits of users in the dialogue process, so that a target user can be predicted more accurately by an auxiliary prediction model in the subsequent prediction process.

Specifically, in an alternative embodiment, high-level semantic features are extracted from dialogue data as semantic-predisposed feature vectors. For example, a recurrent neural network (Recurrent Neural Network, RNN) is employed to extract high-level semantic features from dialog text (i.e., dialog data) and to vector the extracted high-level semantic features into semantic-predisposing feature vectors. The network model used here is not limited to the cyclic neural network in the example, but may be other machine learning models, which are not developed here.

The processing flow of the recurrent neural network to the dialogue text specifically comprises the following steps: preprocessing the dialogue text to obtain a preprocessed text. Furthermore, by the steps of word list construction, vector embedding, sequence filling and the like, a cyclic neural network is constructed based on a pre-training language model (Bidirectional Encoder Representations from Transfo-mers, BERT). Furthermore, the word embedding vector is used as input, an output sequence is generated through the cyclic neural network, the feature extraction and vectorization operations are executed, and the hidden state of the last layer in the cyclic neural network is used as a semantic tendency feature vector. In practice, the BERT model includes, but is not limited to: deBERTa model, LEBERT model, cogBERT model, syntaxBERT model, sentence-BERT model. Of course, these BERT models are merely examples and are not limited herein.

In practical application, the semantic-predisposing feature vector includes at least one of the following: information density, sentence complexity, questioning habits, self-disclosure level, emotion polarity change, dialog synchronicity.

The concept and extraction of the above-described semantic-predisposed feature vector is described below.

Information density: for measuring the amount of effective information conveyed in the conversation. Wherein the effective information accounts for the proportion of the whole information. Effective information refers to information that is meaningful in a conversation, such as information with actual conversation content, rather than insignificant details or lengthy expressions. The invalid information is usually expression mode, stop word, etc., and has no influence on sentence semantics.

As an alternative embodiment, characters whose validity vector features satisfy the validity information recognition condition are determined from the validity vector features of the respective characters in the dialogue data by the BERT model. Further, the proportion of these effective characters in the entire dialogue data is calculated as the information density of the dialogue data.

Statement complexity: for analyzing the grammar structure used by the target user to evaluate the language ability and the depth of thinking of the target user. Specifically, the frequency or proportion of various grammar structures in the dialogue text is counted through a machine learning model, so that whether the grammar using habit of a user frequently uses more complex grammar structures or not is judged. If more complex grammar structures are often used, the sentence used by the user is considered to be more complex. If the sentence complexity used by the user is higher, the user has higher language capability and logic thinking capability.

For example, dialogue information is input into a sentence structure feature extraction model to obtain sentence structure feature vectors corresponding to respective sentences, and thereby, various grammar structures included in respective sentences and the proportion of the respective grammar structures in the respective sentences are calculated based on the sentence structure feature vectors corresponding to the respective sentences. Furthermore, the grammar structure distribution and the grammar structure duty ratio in each sentence are input into a sentence complexity evaluation model to obtain the sentence complexity score of the user. Thus, the language ability and the thinking depth of the user are judged by the scores.

Asking habits: the method is used for observing the frequency, the type and the question mode of the target user in the dialogue so as to explore curiosity, the awareness and the listening degree of the target user in the communication process.

Specifically, the number, type and mode of questions presented in the dialogue are counted by the machine learning model, and the user's habit of questions is determined from the dimensions. The questioning habit can show the dialogue purpose and attitudes of the user. Further, the field, the opportunity and the talking object of the questions preferred by the user can be identified, so that the dialogue style of the user is further analyzed to assist in determining the character attribute of the user.

For example, based on the specific domain keywords in the dialogue information, the dialogue information is classified according to the related domain, the number of questions is counted in various dialogue information, and the question sentence feature vector corresponding to each question sentence is obtained through the feature extraction model. Further, a first degree of association of the current question sentence with the corresponding domain is analyzed based on the question sentence feature vector, and a second degree of association of the current question sentence with the context information is analyzed. And judging the question asking mode and the interest preference degree of the user according to the association degree.

For example, based on the first relevance, whether the user is interested in the topic in the field or not is judged through a preset interest algorithm, and an interest score is obtained. That is, the higher the interest score, the more interested the user is in the topic of this field. If the interest scores of the users in the multiple fields exceed the threshold value, the user is interested in the multiple fields, and the exploration desire and curiosity of the user can be obtained.

For example, according to the second association degree, a question asking mode of the user can be judged, including whether the user actively inquires about related information in the current field. The question sentence has a larger second association with the context, and the association with the context information is tighter. Therefore, whether other information in the current field appears in the question sentence can be further judged, if the other information in the current field appears, the user is judged to want to actively inquire about the related information in the current field, the information in the current field is known, and the interest in the current field is large.

Or, whether the user continues the topic in the current field can be judged according to the second association degree. If the second association degree of the question sentence and the context is larger, the association with the context information is tighter, so that the user can be judged to want to continue the current topic, and the user is interested in the current field.

Degree of self disclosure: the method is used for analyzing the frequency and the depth of the personal information of the target user in the dialogue so as to evaluate the openness and the trust degree of the target user in the communication. In particular, the degree of self-disclosure of the user is obtained by calculating the amount and depth of personal information disclosure in the dialog. A high level of self-disclosure indicates that the user prefers to share personal information and display true self.

As an alternative embodiment, the dialogue information is input into the privacy feature extraction model, and the privacy feature vector related to the personal information in the dialogue is output. And calculating the privacy depth corresponding to the privacy feature vector based on the corresponding relation between the pre-configured privacy information type and the privacy class. And calculates the information density of the private information in the dialogue information. The self-disclosure degree of the user in the dialogue is determined by combining the privacy depth and the information density of the privacy information, and the self-disclosure degree is used for representing the openness degree and the credibility degree of the user in the dialogue.

Affective polarity change: the method is used for observing the change of the emotion positive and negative of the target user in the dialogue so as to explore the emotion stability and adaptability of the target user.

Specifically, the change in emotion polarity is obtained by analyzing the change in emotion polarity in the dialogue. For example, user emotion changes from positive to negative or vice versa. The frequency and magnitude of the change in emotion polarity may reveal the change in emotion during the user's conversation. For example, dialogue information is input into the emotion polarity labeling model, and emotion polarity feature vectors corresponding to each sentence in the dialogue are output. The emotion polarity feature vector includes at least: the emotion polarity labels corresponding to the sentence, for example, emotion polarity labels are classified into: positive, neutral, negative. And counting emotion polarity label change trends among a plurality of sentences, for example, emotion polarity labels in unit sentences, so as to obtain emotion polarity change condition curves among the plurality of sentences in a summarizing way.

Further, the number of sentences required from one emotion polarity tag to another emotion polarity tag, the context information of the sentence where the change is located, the sentence type, the sentence topic, the sentence grammar structure, and the like can also be counted. Thus, based on the statistical information, the influence factors related to the emotion polarity change condition can be analyzed, and further, based on the analyzed influence factors, the character attribute of the user can be determined.

Dialog synchronicity: the method is used for analyzing the communication rhythm and topic switching of the target user in the dialogue so as to evaluate the interaction default degree and the communication capacity of the target user. Specifically, by calculating the response speed and the fluency of topic conversion in the dialogue, the assessment of the dialogue synchronization capability can be obtained.

As an alternative embodiment, the reply speed of the user in the dialogue and the change trend information generated by the reply speed along with topic conversion are obtained. And inputting the reply speed and the change trend information of the user into a first dialogue synchronicity feature extraction model to obtain an exchange rhythm feature vector used for representing the exchange rhythm characteristics of the user in the dialogue. Topics related by a user in a conversation are acquired, and topic switching sentences are identified and marked in conversation information. Inputting topic switching sentences and context information of the topic switching sentences into a second dialogue synchronicity feature extraction model to obtain topic switching feature vectors used for representing topic switching modes of users in dialogs. Further, the communication rhythm feature vector and the topic switching feature vector implicit in the dialogue information are input into a dialogue synchronization prediction model, and a dialogue synchronization score of the user is output, wherein the dialogue synchronization score characterizes dialogue synchronization performance exhibited by the user in the dialogue. Therefore, the social capacity of the user can be judged based on the dialogue synchronization score, and the adaptability degree of the user to social occasions can be judged.

Furthermore, analysis modules for different fields can be added, so that the change trend or score distribution probability of dialogue synchronism when the user switches between different fields is output, and the distinguishing expression of the user in different types of topics is further judged, so that the character attribute characteristics of the user are further evaluated.

Further optionally, the conversation strategy of the chat robot can be dynamically adjusted according to the semantic tendency feature vector, so that more conversation information is obtained, and the expression habit of the target user in the conversation process and the character, emotion and thinking mode reflected by the expression habit are further mined. In addition, the dialogue strategy is dynamically adjusted, so that the flexibility of dialogue can be increased, the interest of character test is improved, and the man-machine experience is improved.

For example, when the self-disclosure degree of the target user is detected to be lower than the threshold value, the dialogue strategy of the chat robot is changed, the question content proportion of the chat information is reduced, or the occurrence density of the question content is reduced, or the generation strategy of the question sentence pattern is changed, so that the comfort of the target user in the dialogue process is improved, and more dialogue data is acquired. Specifically, the generation strategy of changing the question sentence mode can be to change the question sentence to the exclamation sentence so as to improve the affinity to cause the co-emotion.

In addition to the above semantic-oriented feature vectors, in the embodiment of the present application, low-level semantic feature vectors may be further extracted from dialogue data as language habit feature vectors, including but not limited to: subject keywords, idiomatic features, punctuation idiomatic features, field length of unit text, dialogue time interval, idiomatic change feature conditions. For example, key fields capable of representing the subject content of the current dialogue are extracted from the training sample data, and the key words or words representing the subject of the current dialogue are extracted, so that the use condition of the word of the language in the dialogue process of the user, for example, whether each sentence contains: and (5) performing statistics on the Chinese words such as the yao, the morale and the like. The punctuation mark use category and frequency of each sentence can also be extracted from the training sample data, for example, a plurality of exclamation marks or question marks appear in one sentence, and the situation that each sentence is short and does not comprise the punctuation mark is also possible. The length of the dialogues of different rounds, the number of words and the number of fields of each dialogue transmitted, the reply time interval between the different rounds and the habit change condition of the user in the dialogue process can be extracted. For example, the language habit of a user in a certain group of data always repeatedly speaks a word, the dialogue time interval is longer, and the corresponding user can be determined to be a square character by comparing the features through a preset character attribute standard. Through the characteristics, character attributes corresponding to different dialogue characteristics of the person can be comprehensively analyzed, and further character testing of the user is achieved.

Therefore, more data is provided for the prediction process as an analysis basis, and the accuracy of the prediction model is further improved.

And step 103, obtaining a scale test result fed back by the target user based on the psychology scale.

The scale test results are obtained according to the scale test of the professional psychology, and different feedback results correspond to different characters. For example, the user selects the corresponding option (i.e. the above-mentioned meter answer) according to each question in the psychology meter, and then determines the character attribute of the user according to the binding relationship between the option and the character feature, as the final meter test result. Here, the binding relation between the options and the character features is usually set based on the professional psychology theory, so that the scale test result is supported by the professional psychology theory and is more accurate.

Specifically, in the dialogue system, the psychological scale can be pushed to the target user through the chat robot, so that options fed back by the user based on the psychological scale are obtained, and the scale test result of the target user is obtained. Or when the target user logs in the system, pushing the psychological scale to the target user, prompting the target user to fill in the scale, assisting in completing the psychological test, acquiring options fed back by the user based on the psychological scale, and obtaining the scale test result of the target user.

Further optionally, before step 103, a chat robot is employed to push a psychology scale. For example, the chat robot generates corresponding dialogue pushing information based on the psychology scale and pushes the dialogue pushing information to the target user. Next, in step 103, response information of the target user is received as a test result of the target user's scale. In practical application, the dialogue push information may be test questions contained in the psychology scale, chat content generated based on the test questions in the psychology scale, or other forms.

The chat robot related to the embodiment of the application may be a large language model of a generation type, or may be a dialogue generation module of other forms, which is not limited in the application.

And 104, extracting character bias characteristic vectors from the scale test results.

Wherein, character bias characteristic vector characterizes character bias implicitly existing in test results of the table. Character bias feature vectors are feature information implicitly existing in the scale test results and are used for describing more personalized selection preferences of users in the process of feeding back based on the psychology scale, so that a target user can be predicted more accurately by assisting a prediction model in the subsequent prediction process.

In practical application, the character bias characteristic vector comprises at least one of the following: exogenously, humanity, responsibility, neuro, openness. The characteristics are embodied in the character attributes as follows: openness refers to the attitude of an individual to new things, new ideas, and new ideas. The openness is proportional to curiosity and desire to explore new things, unknown things. The responsibility refers to the sense of responsibility and autonomy of the individual to himself. The accountability is proportional to the person's planability and mobility. For example, highly responsible individuals are more orderly, time-keeping, reliable, and have clear goals and plans for work and life. People with low responsibility are more random and loose, and have no clear plan and goal for their own work and life. Exotropy refers to the attitude and behavior of an individual to social activity. The exology is proportional to social ability. Humanization refers to the friendliness, cooperation and concentricity of individuals. The cytoplasm refers to the sensitivity of an individual to stress and anxiety.

Similar to the above way of extracting the semantic-predisposing feature vector, the high-level semantic features may also be extracted from the scale test result as character-predisposing feature vectors. It should be noted that, the network model for extracting the character bias feature vector and the network model for extracting the semantic bias feature vector are the same type of network, and may also be different types of networks, which are all adapted to specific application scenarios, and the application is not limited. Even if the same type of network model is used, certain differences may exist between the network parameters and the network structure so as to adapt to different data extraction requirements.

For example, a convolutional neural network (Convolutional Neural Network, CNN) is used to extract high-level semantic features from the scale test results and to vector the extracted high-level semantic features into character bias feature vectors. Further optionally, a character attention variable cluster associated with each character attribute is extracted from the scale test result data. And further, analyzing the character attention variable clusters associated with each character attribute to determine character trend variables associated with each character attribute. And clustering the character attributes until a clustering end condition is met, determining characteristic distribution of each cluster corresponding to the character bias nodes, outputting character bias based on the characteristic distribution, and determining character attribute bias associated with each character attribute. Therefore, before character bias feature vector extraction, the analysis performance of the character bias mining output process aiming at the relevance features and the non-relevance features can be improved by determining each clustering feature distribution of the character bias nodes based on character bias extraction dimensions, and the reliability of character bias analysis is further improved.

Further, from the scale test result data, the following operations are performed in accordance with at least one of character bias evaluation dimensions of exogenously, humanizedly, disciplinary, neural, openness, and the like, through the convolutional neural network, so as to obtain character bias feature vectors. The specific operation is as follows:

And if analyzing that the to-be-analyzed information related to the character bias estimation dimension exists in the test result data of each table, calculating character attribute bias key information biased to the to-be-analyzed information and a proportion coefficient between standard character attributes associated with the to-be-analyzed information, and determining the corresponding bias degree of the to-be-analyzed information under the current character bias estimation dimension so as to output a corresponding character bias characteristic vector.

The network model used herein is not limited to convolutional neural networks in the example, but may be other machine learning models, which are not developed here.

And 105, inputting the character bias feature vector and the semantic bias feature vector into a test model to obtain character test results of the target user.

Because the semantic tendency feature vector and the character bias feature vector are hidden features embedded into the original data, the two hidden features are fused and input into the test model through the step, and the prediction result which is more in line with the hidden characters shown by the target user in the dialogue can be obtained, so that the purposes of optimizing the output result of the prediction model and improving the accuracy of character test are achieved.

Specifically, in step 105, as an optional embodiment, the character bias feature vector and the semantic bias feature vector are input into the test model to obtain the character test result of the target user, which may be implemented as:

taking character bias characteristic vectors as truth values reflecting character characteristics of a target user under a plurality of evaluation dimensions; fusing the true value with the semantic tendency feature vector to obtain a diversity semantic feature vector of the target user; and predicting psychological characteristic images of the target user based on the diversity semantic feature vectors through the test model, and taking the psychological characteristic images as character test results of the target user.

For example, character bias feature vectors and semantic bias feature vectors are deeply data analyzed. And further, the character bias feature vector extracted based on the large five-personality trait questionnaire result (scale test result) is taken as a true value, and is fused with the semantic tendency feature vector to obtain the diversity semantic feature vector of the target user. Furthermore, the diversity semantic feature vector of the target user is used as the input of the test model, and a prediction model constructed by a convolutional neural network can be adopted to predict the psychological feature image of the target user as the character test result of the target user.

Further optionally, in the step, the psychological characteristic image of the target user is predicted based on the diversity semantic feature vector through the test model, and after the psychological characteristic image is used as the personality test result of the target user, the physiological index data of the target user under the specific psychological activities can be obtained; calculating a dynamic evaluation confidence level of the psychological characteristic portrait of the target user based on the physiological index data and the reference psychological characteristic portrait bound in advance for the specific psychological activity; and if the dynamic evaluation confidence coefficient is higher than the set dynamic evaluation confidence coefficient threshold value, displaying the psychological characteristic image of the target user. The physiological index data of the user under the specific psychological activities can be, for example: heart rate, body temperature, blood oxygen, blood pressure, respiratory frequency, etc. acquired by the wearable device. For example, the index data may change for a user in an anxiety state. Therefore, the state and emotion of the user at present can be introduced into the evaluation of the personality test result through the physiological index data of the target user under the specific psychological activities and the dynamic evaluation confidence mechanism established based on the physiological index data, so that the accuracy of the personality test result is further improved.

In the embodiment of the present application, further alternatively, the test model may be set as a machine learning model having interpretability, which may be referred to herein as a visualization model. The visual model shows the prediction process from the diversified semantic feature vectors to the psychological feature representation. In this way, the visual model with the interpretability can show the basis and the process of the psychological characteristic image prediction process in the machine learning model through the visual technology or the interpretative algorithm.

In the related art, the prediction model is generally required to be integrally trained based on the complete data set, however, the training mode is long in time consumption and is not suitable for application scenes with strong requirements on real-time performance. Therefore, in the embodiment of the application, a more efficient model training mode is also provided.

After step 102 or step 104, the test model is locally updated based on the scale test results or the semantic-predisposed feature vector using an incremental learning (Incremental Learning) algorithm to optimize network parameters in the predictive model. The prediction model can be dynamically optimized through the incremental learning algorithm, the prediction performance of the prediction model is improved, and the method is more suitable for a real-time dialogue data processing scene.

Specifically, the newly added scale test result or dialogue data is input into the existing feature extraction layer in the prediction model, and the newly added feature vector corresponding to the newly added scale test result or dialogue data is obtained. In the prediction model, a new task layer is constructed for learning key information in the new feature vector. Taking the newly added dialogue data as an example, the task layer is used for learning and mining implicit habitual expressions in the newly added dialogue data, such as sentence structures, question frequencies, personal information depth, emotion polarity change rule curves, dialogue topics and the like in the newly added dialogue data. And the learned key information is used for adjusting and updating network parameters (namely local parameters) in the newly added task layer, such as convolution kernel types, quantity, weight parameters and the like in the model. For example, a new data set is constructed using the newly added meter test results or dialogue data, and then the newly added task layer is trained using the new data set. Optionally, the network parameters in the newly added task layer are updated using random gradient descent. In the process, the existing network parameters in the prediction model are kept unchanged, and only the network parameters of the new task layer are trained and updated by adopting the new data set, so that the synchronous operation of the operation prediction model can be kept, and the influence on the existing basic parameters in the prediction model is avoided.

Besides the incremental learning algorithm, in practical application, the accuracy and generalization capability of the prediction model can be further provided by adopting modes of model fusion, self-adaptive learning, time sequence analysis and the like.

In the above or below embodiment, in step 105, character bias feature vectors and semantic tendency feature vectors are input into a test model, and after obtaining character test results of a target user, probability distribution trends of the test model in a preset period are also obtained; if the test model is in the flat probability distribution or the deflection probability distribution, determining that the test model is in a decay trend; and updating and training the test model of the attenuation trend. Therefore, whether the prediction model needs to be trained and updated is evaluated in real time, the accuracy of the prediction model is further improved, and the adaptability of the model in a real-time character test scene is improved.

In the above or below embodiments of the present application, a method for training a predictive model is also provided. The method comprises the following steps:

in the above or below embodiments, the test model is trained by the following method: acquiring training sample data and table test results corresponding to each group of data in the training sample data, and taking a set of all table test results as character attribute standards; obtaining a training sample dialogue feature set from the training sample data; the training sample dialogue feature set is a feature for representing dialogue character descriptions; selecting a target training rule from a plurality of pre-deployed training rules according to the training sample dialogue feature set; training a test network to be trained by using the training sample dialogue feature set, the character attribute standard and the target training rule to obtain the test model; the test model is used for respectively determining psychological characteristic images corresponding to different dialogue characteristics in the training sample dialogue characteristic set by combining the character attribute standard.

In this embodiment, the correspondence from dialog data to personality attributes is accomplished by building a pre-trained test model. Specific implementation is shown in fig. 2, fig. 2 shows a flow chart of a training and testing method, firstly, training sample data is obtained, a plurality of testees are recruited, and are randomly grouped to perform two-by-two conversations, and conversation data is obtained as the training sample data, however, the training sample data can be obtained in other manners. And performing professional psychology scale test on the testees, and taking the obtained scale test result as the character label of each testee.

Then, a training sample feature set with multiple dimensions is obtained from the training sample data, wherein the training sample feature set mainly comprises: extracting and vectorizing the obtained semantic tendency feature vector from dialogue data, and extracting and vectorizing the obtained character bias feature vector from the scale test result.

The semantic-predisposing feature vector includes at least one of: information density, sentence complexity, questioning habits, self-disclosure level, emotion polarity change, dialog synchronicity. That is, the training sample data may be an evaluation score of each subject in each dimension described above. In addition, the characteristics of theme keywords, idiom word habit characteristics, punctuation habit characteristics, field length of unit text, dialogue time interval, habit change characteristic conditions and the like can also be adopted as a training sample characteristic set.

The character bias characteristic vector comprises at least one of the following: exogenously, humanity, responsibility, neuro, openness. That is, the character label of each tester may be an evaluation score of each tester in each dimension described above.

And then training by adopting a plurality of machine learning methods, performing ten-fold cross validation to determine an optimal target training rule, and training based on the optimal target training rule to obtain a pre-trained test model. Because the character labels corresponding to each tested person are obtained, after the training sample feature set is obtained, the character labels of the tested person corresponding to certain or some specific features can be determined, and the purpose of determining the character labels according to the semantic tendency feature vector and the character bias feature vector is achieved.

In this embodiment, by acquiring enough training sample data and the scale test result corresponding to each user, it is ensured that the obtained training sample feature set includes all the behavioral features of the person, such as personality traits, temporary states, moods, and the like, so as to determine the corresponding user and the corresponding feature. The test model can comprehensively and normally test dialogue data of the target user and output character test results.

In an alternative embodiment, the selecting, according to the training sample dialogue feature set, the target training rule from the plurality of pre-deployed training rules may be implemented as:

and training the training sample dialogue feature set by using the plurality of pre-deployed training rules respectively, and taking the training rule with the training result closest to a preset value as the target training rule.

Wherein the plurality of pre-deployed training rules includes at least two of the following rules: logistic regression, K nearest neighbor KNN, random forest, decision tree.

In this embodiment, multiple training rules are evaluated by cross-validation, so as to determine an optimal training rule, thereby greatly improving the testing efficiency.

In an optional implementation manner, after determining the personality attribute corresponding to the target dialog feature according to a preset personality attribute standard and taking the personality attribute as the personality test result of the target user, test feedback of the target user on the personality test result is also obtained, and the pre-trained test model is optimized according to the test feedback.

In this embodiment, as shown in fig. 2, after the character test result is output to the user, the user can give feedback of the test result according to the situation of the user, the accurate place is kept, the inaccurate place of the test result is modified, the test model is optimized continuously, and the accuracy of the test is increased.

Having described the method of the embodiments of the present application, next, description is made of a personality testing apparatus based on a human-computer conversation of the embodiments of the present application with reference to fig. 3.

The personality testing device 30 based on man-machine interaction in the embodiment of the present application may implement the steps corresponding to the personality testing method based on man-machine interaction in the embodiment corresponding to fig. 1. The functions realized by the personality test device 30 based on the man-machine interaction can be realized by hardware, or can be realized by executing corresponding software by hardware. The hardware or software includes one or more modules corresponding to the functions described above, which may be software and/or hardware. The personality testing device 30 based on man-machine interaction is applied to a server device or a terminal device. The personality testing device 30 based on man-machine interaction may include a transceiver module 301 and a processing module 302, where the function implementation of the transceiver module 301 and the processing module 302 may refer to operations performed in the embodiment corresponding to fig. 1, which are not described herein. For example, the processing module 302 may be configured to control data transceiving operations of the transceiver module 301.

In some embodiments, the transceiver module 301 is configured to obtain dialogue data between the target user and the chat robot that meets a preset test condition; acquiring a scale test result of the target user; the scale test result is obtained based on psychology scale feedback;

a processing module 302 configured to extract a semantic-predisposed feature vector from the dialogue data; wherein the semantic-predisposing feature vector characterizes a habitual expression implicitly present in the dialog data; the semantic-predisposing feature vector includes at least one of: information density, sentence complexity, questioning habits, self-disclosure degree, emotion polarity change, and dialogue synchronism; and

extracting character bias characteristic vectors from the scale test results; the character bias characteristic vector characterizes character bias implicitly existing in the test result of the scale; the character biasing feature vector includes at least one of: exogenously, preferably humanized, responsible, neuro-cytoplasmic and openness; and

In some embodiments, when the processing module 302 inputs the character bias feature vector and the semantic bias feature vector into a test model to obtain a character test result of the target user, the processing module is configured to:

reflecting true values of character characteristics of the target user under a plurality of evaluation dimensions by taking the character bias characteristic vector;

fusing the true value with the semantic tendency feature vector to obtain a diversity semantic feature vector of the target user;

and predicting psychological characteristic images of the target user based on the diversity semantic feature vectors through the test model, and taking the psychological characteristic images as character test results of the target user.

In some embodiments, the processing module 302 is further configured to:

predicting psychological characteristic images of the target user based on the diversified semantic feature vectors through the test model, and taking the psychological characteristic images as character test results of the target user,

acquiring the physiological index data of the target user under specific psychological activities;

calculating the dynamic evaluation confidence level of the psychological characteristic portrait of the target user based on the physiological index data and the reference psychological characteristic portrait bound in advance for the specific psychological activities;

And if the dynamic evaluation confidence coefficient is higher than a set dynamic evaluation confidence coefficient threshold value, displaying the psychological characteristic image of the target user.

In some embodiments, the processing module 302 is further configured to:

after the scale test result or the semantic tendency feature vector is obtained, adopting an incremental learning algorithm to update the test model locally based on the scale test result or the semantic tendency feature vector so as to optimize network parameters in the prediction model.

In some embodiments, the processing module 302 is further configured to:

inputting the character bias feature vector and the semantic tendency feature vector into a test model, and acquiring probability distribution trend of the test model in a preset period after obtaining character test results of the target user;

if the test model is in the flat probability distribution or the deflection probability distribution, determining that the test model is in a decay trend;

and updating and training the test model of the attenuation trend.

In some embodiments, when the transceiver module 301 obtains dialogue data between the target user and the chat robot that meets the preset test condition, the transceiver module is configured to:

Receiving real-time dialogue data between the target user and the chat robot; the real-time dialogue data carries dialogue identification;

storing the real-time dialogue data into a message queue according to the dialogue identifier, and recording the response times between the target user and the chat robot;

and detecting that the response times between the target user and the chat robot exceeds preset response times, and storing the real-time dialogue data into a persistent storage system.

In some embodiments, the test model is a visual model with interpretability; the visual model shows the prediction process from the diversified semantic feature vectors to the psychological feature representation.

Having described the method and apparatus of the embodiments of the present application, next, a description will be given of a computer readable storage medium of the embodiments of the present application, which may be an optical disc, on which a computer program (i.e., a program product) is stored, where the computer program, when executed by a processor, implements the steps described in the foregoing method embodiments, for example, obtaining dialogue data between a target user and a chat robot that satisfies preset test conditions; extracting semantic tendency feature vectors from the dialogue data; wherein the semantic-predisposing feature vector characterizes a habitual expression implicitly present in the dialog data; the semantic-predisposing feature vector includes at least one of: information density, sentence complexity, questioning habits, self-disclosure degree, emotion polarity change, and dialogue synchronism; acquiring a scale test result of the target user; the scale test result is obtained based on psychology scale feedback; extracting character bias characteristic vectors from the scale test results; the character bias characteristic vector characterizes character bias implicitly existing in the test result of the scale; the character biasing feature vector includes at least one of: exogenously, preferably humanized, responsible, neuro-cytoplasmic and openness; and inputting the character bias characteristic vector and the semantic tendency characteristic vector into a test model to obtain a character test result of the target user. The specific implementation of each step is not repeated here.

It should be noted that examples of the computer readable storage medium may also include, but are not limited to, a phase change memory (PRAM), a Static Random Access Memory (SRAM), a Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a flash memory, or other optical or magnetic storage medium, which will not be described in detail herein.

The personality test apparatus 30 based on man-machine interaction in the embodiment of the present application is described above from the point of view of the modularized functional entity, and the server and the terminal device for executing the personality test method based on man-machine interaction in the embodiment of the present application are described below from the point of view of hardware processing, respectively.

It should be noted that, in the embodiment of the personality testing apparatus based on man-machine conversation of the present application, the entity devices corresponding to the transceiver module 301 shown in fig. 3 may be an input/output unit, a transceiver, a radio frequency circuit, a communication module, an input/output (I/O) interface, and the like, and the entity devices corresponding to the processing module 302 may be a processor. The personality testing device 30 based on the man-machine conversation shown in fig. 3 may have a structure as shown in fig. 4, and when the personality testing device 30 based on the man-machine conversation shown in fig. 3 has a structure as shown in fig. 4, the processor and the transceiver in fig. 4 can implement the same or similar functions as the processing module 302 and the transceiver module 301 provided in the device embodiment corresponding to the device, and the memory in fig. 4 stores a computer program that needs to be invoked when the processor executes the personality testing method based on the man-machine conversation.

Fig. 5 is a schematic diagram of a server structure provided in an embodiment of the present application, where the server 1100 may vary considerably in configuration or performance, and may include one or more central processing units (central processing units, CPU) 1122 (e.g., one or more processors) and memory 1132, one or more storage media 1130 (e.g., one or more mass storage devices) storing applications 1142 or data 1144. Wherein the memory 1132 and the storage medium 1130 may be transitory or persistent. The program stored on the storage medium 1130 may include one or more modules (not shown), each of which may include a series of instruction operations on a server. Still further, the central processor 1122 may be provided in communication with a storage medium 1130, executing a series of instruction operations in the storage medium 1130 on the server 1100.

The Server 1100 may also include one or more power supplies 1126, one or more wired or wireless network interfaces 1150, one or more input-output interfaces 1158, and/or one or more operating systems 1141, such as Windows Server, mac OS X, unix, linux, freeBSD, and the like.

The steps performed by the server in the above embodiments may be based on the structure of the server 1100 shown in fig. 5. For example, the steps performed by the human-machine-dialog-based personality testing device 80 shown in fig. 5 in the above-described embodiments may be based on the server structure shown in fig. 5. For example, the CPU 1122 may perform the following operations by calling instructions in the memory 1132:

receiving and acquiring dialogue data between the target user and the chat robot, which meets the preset test conditions, through the input-output interface 1158; obtaining a scale test result of the target user; the scale test result is obtained based on psychology scale feedback;

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to related descriptions of other embodiments.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the systems, apparatuses and modules described above may refer to the corresponding processes in the foregoing method embodiments, which are not repeated herein.

In the several embodiments provided in the embodiments of the present application, it should be understood that the disclosed systems, apparatuses, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, and for example, the division of the modules is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple modules or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or modules, which may be in electrical, mechanical, or other forms.

The modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium.

In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product.

The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, the flow or functions described in accordance with embodiments of the present application are fully or partially produced. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid State Disk (SSD)), etc.

The foregoing describes in detail the technical solution provided by the embodiments of the present application, in which specific examples are applied to illustrate the principles and implementations of the embodiments of the present application, where the foregoing description of the embodiments is only used to help understand the methods and core ideas of the embodiments of the present application; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope according to the ideas of the embodiments of the present application, the present disclosure should not be construed as limiting the embodiments of the present application in view of the above.

Claims

1. A character testing method based on man-machine conversation is characterized by comprising the following steps:

acquiring a scale test result fed back by the target user based on a psychology scale;

inputting the character bias feature vector and the semantic tendency feature vector into a test model to obtain a character test result of the target user;

the character bias feature vector and the semantic bias feature vector are input into a test model to obtain character test results of the target user, and the character test results comprise:

2. The method of claim 1, wherein predicting, by the test model, the psychological-characteristic image of the target user based on the diversity semantic-characteristic vector, as a result of the personality test of the target user, further comprises:

3. The method of claim 1, wherein after obtaining the scale test result or the semantic-predisposed feature vector, further comprising:

and adopting an incremental learning algorithm to update the test model locally based on the scale test result or the dialogue data so as to optimize network parameters in the test model.

4. The method of claim 1, wherein the inputting the character bias feature vector and the semantic bias feature vector into a test model to obtain character test results for the target user further comprises:

acquiring probability distribution trend of the test model in a preset period;

And updating and training the test model of the attenuation trend.

5. The method as set forth in claim 1, wherein the acquiring dialogue data between the target user and the chat robot satisfying a preset test condition includes:

6. The method of any one of claims 1 to 5, wherein the test model is a visual model with interpretability;

the visual model shows the prediction process from the diversified semantic feature vectors to the psychological feature representation.

7. A personality testing device based on a human-machine conversation, the device comprising:

the processing module is used for inputting the character bias characteristic vector and the semantic tendency characteristic vector into a test model to obtain character test results of the target user;

the processing module inputs the character bias feature vector and the semantic tendency feature vector into a test model, and when obtaining character test results of the target user, the processing module is configured to:

8. An electronic device comprising a memory and one or more processors; wherein the memory is for storing computer program code, the computer program code comprising computer instructions; the computer instructions, when executed by the processor, cause the electronic device to perform the personality test method based on human-machine interaction of any of claims 1-6.

9. A computer readable storage medium comprising a computer program which, when run on a computer, causes the computer to perform the character testing method based on man-machine conversation as claimed in any one of claims 1 to 6.