Data interaction method and system based on artificial intelligence

Abstract

The invention provides a data interaction method and system based on artificial intelligence, and belongs to the technical field of artificial intelligence. User hidden emotion features are accurately analyzed by fusing user interaction problem information and physiological signals, the future multi-round dialogue emotion trend is predicted accordingly, corresponding advanced planning and buffer pool management are achieved, and user experience is improved. Meanwhile, by means of correlation analysis and candidate grade division of the multi-segment information to be output, it is ensured that the most relevant and high-quality interaction content is preferentially presented, when user feedback does not accord with prediction, analysis can be conducted again in time, an output strategy can be dynamically adjusted, accurate and flexible interaction experience is provided for users, and the user experience is improved. And the data interaction efficiency and the user satisfaction are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to an artificial-intelligence-based data interaction method and system. Background Art

[0002] In today's digital interaction era, especially with the development of artificial intelligence, the depth and breadth of human-computer interaction have been continuously expanded, and users' expectations for intelligent interaction systems have also been increasing day by day. However, traditional human-computer data interaction methods have many limitations when dealing with complex and changing user needs and emotional states, and it is difficult to provide accurate, flexible, and efficient interaction experiences.

[0003] On the one hand, traditional interaction methods mainly rely on the problem information clearly expressed by users to generate responses, ignoring the potential emotional factors of users. In fact, users often have various subtle emotional changes during the interaction process, and these hidden emotional features are crucial for understanding users' true intentions and needs. Due to the lack of effective capture and analysis of users' hidden emotions, the system is difficult to accurately grasp the emotional trend of users, resulting in a deviation between the generated response and users' actual expectations, and unable to meet users' deep-level interaction needs.

[0004] On the other hand, traditional interaction systems lack foresight and flexibility in response generation and output. They usually retrieve information and generate responses after receiving the current problem of users, lacking anticipation of the future dialogue trend, and it is difficult to achieve advance planning and efficient management of responses. In addition, when facing multi-segment information to be output, traditional methods cannot reasonably sort and filter according to its relevance to users' potential emotions and needs, resulting in the content presented to users may not be the most relevant and highest quality, reducing the interaction efficiency.

[0005] Therefore, it is necessary to provide an artificial-intelligence-based data interaction method and system to solve the above technical problems. Summary of the Invention

[0006] To solve the above technical problems, the present invention provides an artificial-intelligence-based data interaction method and system for solving the existing artificial-intelligence data interaction technology problems, such as a large deviation between the generated response and users' actual expectations, and difficulty in achieving advance planning and efficient management of responses.

[0007] An artificial-intelligence-based data interaction method provided by the present invention, the interaction method includes: S1. Obtain user interaction problem information, and at the same time obtain user physiological signals through wearable devices, and analyze the hidden emotional features of user interaction based on the user interaction problem information and user physiological signals; S2. Predict the emotional categories of the next 3 rounds of conversations based on the hidden emotional features of user interaction; S3. Analyze and obtain semantic keywords from the user interaction problem information, retrieve and match the corresponding output information from the database, and form multi-segment output information to be output; S4. Based on the predicted emotional categories of the next three rounds of conversations, conduct a relevance analysis and sorting on the multi-segment output information to be output, and obtain the relevance sorting result; S5. According to the relevance sorting result, divide the candidate levels of the multi-segment output information to be output, where the candidate levels include first-level candidates, second-level candidates, and third-level candidates; S6. Based on the candidate levels of the multi-segment output information to be output, select the first-level candidate output information to be the current primary output, and preset a candidate output information buffer pool based on the predicted emotional categories of the next three rounds of conversations. Obtain the current user interaction hidden emotional characteristics by analyzing the current user interaction problem information in real time. If the current user interaction hidden emotional characteristics match the predicted emotional categories of the next three rounds of conversations, retrieve the response from the buffer pool. If they do not match, re-analyze the user feedback information and physiological signals and then execute the operations in steps S2 - S5 to dynamically adjust the output.

[0008] Preferably, the specific steps of step S1 include the following steps: S101. Receive the user's conversation content through voice or text input, and record it as the user interaction problem information; S102. Synchronously obtain the user's physiological signals through a wearable device, including heart rate, galvanic skin response, or body temperature; S103. Input the user interaction problem information into a pre-trained natural language processing model to analyze the implicit emotion in the text, and generate the user's current interaction hidden emotional characteristics through a pre-trained convolutional neural network model in combination with the user's physiological signals.

[0009] Preferably, the specific steps of step S2 include the following steps: S201. Collect historical conversation data, including the user's historical interaction problem information, physiological signal data, and the corresponding conversation emotional category labels as the training set, and obtain an emotion prediction model after training with a time series analysis model; S202. Input the current interaction hidden emotional characteristics into the emotion prediction model to predict the emotional categories that the user will show in the next three rounds of conversations, and obtain the emotional categories of the next three rounds of conversations; Preferably, the specific steps of step S3 include the following steps: S301. Analyze and obtain semantic keywords from the user interaction problem information through a natural language processing model, and organize and classify the semantic keywords; S302. According to the extracted semantic keywords, retrieve the output information related to the semantic keywords from the pre-constructed database; S303. Organize and combine the retrieved relevant output information to form multi-paragraph output information to be output. Each paragraph of the output information to be output corresponds to one or more semantic keywords.

[0010] Preferably, the specific steps of step S4 include the following steps: S401. Construct an emotional keyword library, extract keywords from the multi-paragraph output information to be output, and obtain the number of single-paragraph keywords for multiple paragraphs. S402. Statistically analyze the coincidence degree, that is, the correlation degree, between the multi-paragraph output information to be output and the emotional keyword library, and obtain the correlation degree of each paragraph of the output information to be output in the multi-paragraph output information to be output. S403. According to the numerical size of the correlation degree of each paragraph of the output information to be output, sort each paragraph of the output information to be output in descending order to generate a correlation degree sorting result.

[0011] Preferably, the specific steps of step S5 include the following steps: S501. According to the correlation degree sorting result, set the candidate level division criteria, including: classifying the output information to be output with the top 30% of the correlation degree rankings as first-level candidates, those ranked 30%-60% as second-level candidates, and those ranked 60% and later as third-level candidates. S502. Based on the set candidate level division criteria, classify each paragraph of the output information to be output in the multi-paragraph output information to be output, that is, mark each paragraph of the output information to be output with the corresponding candidate level.

[0012] Preferably, the specific steps of step S6 include the following steps: S601. At the beginning of the interaction, according to the candidate level of the multi-paragraph output information to be output, select the first-level candidate output information to be output as the current primary output information and display it to the user through the corresponding output device. S602. Based on the predicted emotional categories of the next 3 rounds of conversations, preset a candidate output information buffer pool, where the candidate output information buffer pool includes the output information to be output of the second-level candidates and the output information to be output of the third-level candidates, and is used to supplement the current primary output information. S603. During the interaction process, monitor and obtain the current user interaction problem information in real time, and analyze the current hidden emotional characteristics of the user interaction by the same method as in step S1, identify its emotional category, and compare the current hidden emotional characteristics of the user interaction with the predicted emotional categories of the next 3 rounds of conversations in step S2 to determine whether they match. S604. If it is determined that the currently monitored current user interaction hidden emotion feature matches the predicted future 3-round conversation emotion category, the corresponding response information is retrieved from the candidate output information buffer pool in a preset order for information supplementation; if the currently monitored current user interaction hidden emotion feature does not match the predicted future 3-round conversation emotion category, the user interaction problem information and physiological signals are re-analyzed to obtain new user interaction problem information and physiological signals, and the operations in steps S2 - S5 are performed again.

[0013] An artificial intelligence-based data interaction system, comprising: A data acquisition module, configured to obtain user interaction problem information, and at the same time obtain user physiological signals through a wearable device, and analyze user interaction hidden emotion features based on the user interaction problem information and user physiological signals; A feature analysis module, configured to predict the future 3-round conversation emotion category based on the user interaction hidden emotion features; A feature matching module, configured to analyze and obtain semantic keywords from the user interaction problem information, retrieve and match the corresponding output information from the database to form multi-segmented information to be output; An association analysis module, configured to perform association analysis and sorting on the multi-segmented information to be output based on the predicted future 3-round conversation emotion category to obtain an association degree sorting result; A level division module, configured to divide the candidate levels of the multi-segmented information to be output according to the association degree sorting result, where the candidate levels include first-level candidates, second-level candidates, and third-level candidates.

[0014] An adjusted output module, configured to select the first-level candidate information to be output as the current primary output based on the candidate levels of the multi-segmented information to be output, and preset a candidate output information buffer pool based on the predicted future 3-round conversation emotion category, obtain the current user interaction problem information in real time and analyze the current user interaction hidden emotion feature. If the current user interaction hidden emotion feature matches the predicted future 3-round conversation emotion category, the response is retrieved from the buffer pool; if not, the user feedback information and physiological signals are re-analyzed and the operations in steps S2 - S5 are performed again to dynamically adjust the output.

[0015] Compared with related technologies, an artificial intelligence-based data interaction method and system provided by the present invention have the following beneficial effects: The present invention accurately analyzes the hidden emotional characteristics of users by integrating user interaction problem information and physiological signals, predicts the emotional trends of future multi-round conversations based on this, realizes corresponding advance planning and buffer pool management, and at the same time uses the correlation analysis and candidate level division of multi-segmented information to be output to ensure that the most relevant and high-quality interaction content is presented first. When the user feedback does not match the prediction, it can immediately re-analyze and dynamically adjust the output strategy, providing users with an accurate and flexible interaction experience, greatly improving the efficiency of data interaction and user satisfaction. Brief Description of the Drawings

[0016] Figure 1 It is a schematic flowchart of a data interaction method based on artificial intelligence according to the present invention; Figure 2 It is a system block diagram of a data interaction system based on artificial intelligence according to the present invention. Detailed Embodiments

[0017] The present invention will be further described below in conjunction with the drawings and embodiments.

[0018] Embodiment 1 As Figure 1 shown, a data interaction method based on artificial intelligence includes the following steps: S1. Obtain user interaction problem information, and at the same time obtain user physiological signals through a wearable device, and analyze the hidden emotional characteristics of user interaction based on the user interaction problem information and the user physiological signals; S2. Predict the emotional categories of the next 3 rounds of conversations based on the hidden emotional characteristics of user interaction; S3. Analyze and obtain semantic keywords from the user interaction problem information, retrieve and match the corresponding output information from the database to form multi-segmented information to be output; S4. Conduct a correlation analysis and sorting on the multi-segmented information to be output based on the predicted emotional categories of the next 3 rounds of conversations to obtain a correlation sorting result; S5. Divide the candidate levels of the multi-segmented information to be output according to the correlation sorting result, where the candidate levels include first-level candidates, second-level candidates, and third-level candidates; S6. According to the candidate levels of the multi-segmented information to be output, select the first-level candidate information to be output as the current primary output, and preset a candidate output information buffer pool based on the predicted emotional categories of the next 3 rounds of conversations. Real-time obtain the current user interaction problem information and analyze the current hidden emotional characteristics of user interaction. If the current hidden emotional characteristics of user interaction match the predicted emotional categories of the next 3 rounds of conversations, retrieve the response from the buffer pool; if not, re-analyze the user feedback information and physiological signals and then execute the operations of steps S2 - S5 to dynamically adjust the output.

[0019] In the specific implementation process, step S1 specifically includes the following steps: S101. Receive the user's conversation content through voice or text input, and record it as user interaction problem information; S102. Synchronously obtain the user's physiological signals through a wearable device, including heart rate, galvanic skin response, or body temperature.

[0020] Specifically, establish a communication connection with a wearable device (such as a smart bracelet, smart watch, etc.), and connect using wireless communication methods such as Bluetooth, Wi-Fi, etc. The wearable device is built with multiple sensors for real-time collection of the user's physiological signals. Exemplarily, a heart rate sensor measures the user's heart rate through the principle of photoelectric reflection; a galvanic skin response sensor reflects the user's skin electrical activity by measuring changes in the conductivity of the skin surface; a body temperature sensor measures the user's body temperature using components such as a thermistor.

[0021] S103. Input the user interaction problem information into a pre-trained natural language processing model, analyze the implicit emotion in the text, and generate the user's current interaction hidden emotion feature through a pre-trained convolutional neural network model in combination with the user's physiological signals.

[0022] Specifically, obtain historical interaction data and use it as a training set. After training a deep learning framework, obtain a natural language processing model. Input the user interaction problem information into the pre-trained natural language processing model, perform lexical analysis, syntactic analysis, etc. on the input text, extract the semantic features of the text, and then judge the implicit emotion category (such as positive, negative, neutral, etc.) of the text through an emotion classification layer; then, perform feature extraction and normalization processing on the user's physiological signals (heart rate, galvanic skin response, body temperature, etc.) obtained in step S102 to make them match the text emotion features output by the natural language processing model in terms of dimension and numerical range. Then splice or fuse the processed physiological signal features and text emotion features as the input of the convolutional neural network model. The convolutional neural network model extracts features from the input features through structures such as convolutional layers and pooling layers, and finally generates the user's current interaction hidden emotion feature.

[0023] In the specific implementation process, step S2 specifically includes the following steps: S201. Collect historical conversation data, including the user's historical interaction problem information, physiological signal data, and the corresponding conversation emotion category labels as a training set, and obtain an emotion prediction model after training using a time series analysis model; S202. Input the current interaction hidden emotion feature into the emotion prediction model to predict the emotion category that the user will show in the next 3 rounds of conversations, and obtain the emotion categories of the next 3 rounds of conversations.

[0024] In the specific implementation process, step S3 specifically includes the following steps: S301. Analyze and obtain semantic keywords from the user interaction problem information through a natural language processing model, and organize and classify the semantic keywords.

[0025] Specifically, use a pre-trained natural language processing model to analyze the user interaction problem information, identify the semantic keywords in the text, organize the extracted semantic keywords, remove duplicate keywords, and perform preliminary classification; the classification can be carried out according to the semantics, part of speech, field of the keywords, etc. Exemplarily, for example, classify noun keywords into categories such as people, places, things, etc.; classify verb keywords into categories such as actions, behaviors, etc.

[0026] S302. Retrieve the output information related to the semantic keywords from the pre-constructed database according to the extracted semantic keywords.

[0027] Specifically, pre-construct a database containing a large amount of information covering knowledge in multiple fields such as history, culture, science and technology, life, etc. Among them, the information in the database is stored in a structured or unstructured form. According to the extracted semantic keywords, use full-text retrieval technology to retrieve in the database, match according to the keywords in the text content of the database, and find the information related to the keywords.

[0028] S303. Organize and combine the retrieved relevant output information to form multi-paragraph output information to be output, where each paragraph of output information to be output corresponds to one or more semantic keywords.

[0029] Specifically, according to the classification and relevance of the semantic keywords, organize the sorted information to form multi-paragraph output information to be output. Each paragraph of output information to be output is centered around one or more semantic keywords to ensure the coherence and logic of the information; mark the corresponding semantic keywords for each paragraph of output information to be output for subsequent operations such as correlation analysis.

[0030] In the specific implementation process, step S4 specifically includes the following steps: S401. Construct an emotional keyword library, extract keywords from the multi-paragraph output information to be output, and obtain the number of single-paragraph keywords.

[0031] Specifically, common emotion categories (happiness, sadness, anger, calmness) are classified. Then, a large amount of text data related to these emotion categories is collected, and through manual annotation or the use of sentiment analysis tools, the words closely related to each emotion category are marked, and these words are organized into a sentiment keyword library. For each piece of information in the multi-paragraph information to be output, natural language processing techniques (such as TF-IDF algorithm, TextRank algorithm, etc.) are used to extract keywords. Exemplarily, there is a piece of information to be output: "This mobile phone has a fashionable appearance, great photo-taking effect, and fast running speed." Through the keyword extraction algorithm, keywords such as "mobile phone", "appearance", "fashionable", "photo-taking effect", and "running speed" will be extracted. Count the number of keywords extracted from each piece of information to be output. For example, the above-mentioned piece of information extracted 5 keywords, so the keyword count of this piece of information is 5.

[0032] S402. Statistically analyze the overlap degree, that is, the correlation degree, between the multi-paragraph information to be output and the sentiment keyword library to obtain the correlation degree of each piece of information to be output in the multi-paragraph information to be output.

[0033] Specifically, for each piece of information to be output, match the extracted keywords with the sentiment keyword library, calculate the ratio of the number of keywords in each piece of information to be output to the number of sentiment keywords in the matching sentiment keyword library. The calculation formula is: ; This ratio is the correlation degree of this piece of information to be output.

[0034] S403. According to the numerical size of the correlation degree of each piece of information to be output, sort each piece of information to be output in descending order to generate a correlation degree sorting result.

[0035] Specifically, compare the calculated correlation degree values of each piece of information to be output, sort the multi-paragraph information to be output in descending order, and then adjust the order of the multi-paragraph information to be output according to the sorting result.

[0036] In the specific implementation process, step S5 specifically includes the following steps: S501. According to the correlation degree sorting result, set the candidate level division criteria, including: classifying the information to be output with the top 30% of the correlation degree rankings as first-level candidates, those ranked 30%-60% as second-level candidates, and those ranked 60% and later as third-level candidates.

[0037] S502. Based on the set candidate level division criteria, conduct candidate level division on each piece of information to be output in the multi-paragraph information to be output, that is, mark each piece of information to be output with the corresponding candidate level.

[0038] Specifically, according to the relevance ranking results obtained in step S403, candidate levels are assigned to each segment of the information to be output in descending order, that is, the information to be output with the top 30% of the relevance rankings is classified as first-level candidates, the information with rankings from 30% to 60% is classified as second-level candidates, and the information with rankings of 60% and later is classified as third-level candidates.

[0039] Exemplarily, if there are 10 segments of multi-segment information to be output, then the number of first-level candidate segments = 10 × 30% = 3 segments (rounded down), the number of second-level candidate segments = 10 × 30% = 3 segments, and the number of third-level candidate segments = 10 - 3 - 3 = 4 segments.

[0040] In the specific implementation process, step S6 specifically includes the following steps: S601. At the beginning of the interaction, according to the candidate levels of the multi-segment information to be output, first-level candidate information to be output is preferentially selected as the current primary output information and displayed to the user through the corresponding output device.

[0041] Specifically, obtain the multi-segment information to be output with the candidate levels already divided from step S5, traverse the multi-segment information to be output, and screen out all the information to be output marked as first-level candidates; among the screened first-level candidate information to be output, select one segment as the current primary output information according to the order of the first-level candidate information in the relevance ranking results, and display the selected primary output information to the user through the corresponding output device (such as screen display, voice broadcast, etc.).

[0042] S602. Based on the predicted emotional categories of the next 3 rounds of conversations, preset a candidate output information buffer pool, where the candidate output information buffer pool includes the information to be output of second-level candidates and third-level candidates, and is used to supplement the current primary output information.

[0043] Specifically, based on the predicted emotional categories of the next 3 rounds of conversations, preset a candidate output information buffer pool, which contains the information to be output of second-level candidates and third-level candidates.

[0044] S603. During the interaction, real-time monitor and obtain the current user interaction problem information, and analyze the current hidden emotional characteristics of the user interaction through the same method as in step S1, identify its emotional category, and compare the current hidden emotional characteristics of the user interaction with the predicted emotional categories of the next 3 rounds of conversations in step S2 to determine whether they match.

[0045] Specifically, the problem information currently proposed by the user is monitored in real time. According to the same method in step S1, the current problem information of the user is analyzed to identify the current emotional category of the user. After the identification is completed, the currently obtained hidden emotional features of user interaction are compared with the emotional categories predicted in the next 3 rounds of conversations in step S2. If the emotional categories match, it is considered a match; otherwise, it is considered a mismatch.

[0046] S604. If it is determined that the currently monitored hidden emotional features of user interaction match the emotional categories predicted in the next 3 rounds of conversations, the corresponding response information is retrieved from the candidate output information buffer pool in a preset order for information supplementation; if the currently monitored hidden emotional features of user interaction do not match the emotional categories predicted in the next 3 rounds of conversations, the user interaction problem information and physiological signals are re-analyzed to obtain new user interaction problem information and physiological signals, and the operations in steps S2 - S5 are performed again.

[0047] Specifically, if it is determined that the currently monitored hidden emotional features of user interaction match the emotional categories predicted in the next 3 rounds of conversations, according to the current conversation turn, the corresponding secondary candidate or tertiary candidate information to be output is retrieved from the candidate output information buffer pool in a preset order, and the retrieved secondary candidate or tertiary candidate information to be output is displayed through the corresponding output device in a way of annotation or different colors as a supplement to the primary output information for the user to view; if it is determined that they do not match, the user interaction problem information is received again through voice or text input, and at the same time the user physiological signals are obtained through the wearable device as new user interaction problem information and physiological signals. Using the new user interaction problem information and physiological signals, the operations in step S2 (predicting the emotional categories of the next 3 rounds of conversations), step S3 (analyzing and obtaining semantic keywords from the user interaction problem information and retrieving output information to form multi - segment information to be output), step S4 (performing relevance analysis and sorting), and step S5 (dividing candidate levels) are performed again to adjust the output content.

[0048] Embodiment 2 As Figure 2 shown, an artificial - intelligence - based data interaction system includes: A data acquisition module, which is used to obtain user interaction problem information, and at the same time obtain user physiological signals through a wearable device, and analyze the hidden emotional features of user interaction based on the user interaction problem information and user physiological signals; A feature analysis module, which is used to predict the emotional categories of the next 3 rounds of conversations based on the hidden emotional features of user interaction; A feature matching module, which is used to analyze and obtain semantic keywords from the user interaction problem information, retrieve and match the corresponding output information from the database, and form multi - segment information to be output; The association analysis module is used to perform association degree analysis and sorting on multi-segment to-be-output information based on the predicted future 3-round dialogue sentiment categories, and obtain the association degree sorting result; The level division module is used to divide the candidate levels of the multi-segment to-be-output information according to the association degree sorting result, where the candidate levels include first-level candidates, second-level candidates, and third-level candidates; The adjustment output module is used to, if it is determined that the currently monitored current user interaction hidden emotion feature matches the predicted future 3-round dialogue sentiment category, retrieve the corresponding response information from the candidate output information buffer pool in a preset order for information supplementation; if the currently monitored current user interaction hidden emotion feature does not match the predicted future 3-round dialogue sentiment category, re-analyze the user interaction problem information and physiological signals, obtain new user interaction problem information and physiological signals, and execute the operations in steps S2 - S5 again.

[0049] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0050] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disc memories, tape memories, or any other medium that can be used to carry or store data and is computer-readable.

[0051] It should also be noted that the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, commodity or device including the element.

Claims

1. A data interaction method based on artificial intelligence, characterized in that, The steps of the interaction method include: S1. Obtain user interaction problem information, and at the same time obtain the user's physiological signals through a wearable device. Based on the user interaction problem information and the user's physiological signals, analyze the hidden emotional characteristics of the user interaction; S2. Based on the hidden emotional characteristics of the user interaction, predict the emotional categories of the next 3 rounds of conversations; S3. Analyze and obtain semantic keywords from the user interaction problem information, retrieve and match the corresponding output information from the database, and form multi-segment output information to be output; S4. Based on the predicted emotional categories of the next 3 rounds of conversations, perform a relevance analysis and sorting on the multi-segment output information to be output, and obtain a relevance sorting result; S5. According to the relevance sorting result, divide the candidate levels of the multi-segment output information to be output. Among them, the candidate levels include first-level candidates, second-level candidates, and third-level candidates; S6. According to the candidate levels of the multi-segment output information to be output, select the first-level candidate output information as the current primary output, and preset a candidate output information buffer pool based on the predicted emotional categories of the next 3 rounds of conversations. Real-time obtain the current user interaction problem information and analyze the current hidden emotional characteristics of the user interaction. If the current hidden emotional characteristics of the user interaction match the predicted emotional categories of the next 3 rounds of conversations, retrieve the response from the buffer pool. If they do not match, re-analyze the user feedback information and physiological signals and then execute the operations of steps S2 - S5 to dynamically adjust the output.

2. The data interaction method based on artificial intelligence according to claim 1, characterized in that, The specific steps of step S1 include the following steps: S101. Receive the user's conversation content through voice or text input, and record it as user interaction problem information; S102. Synchronously obtain the user's physiological signals through a wearable device, including heart rate, galvanic skin response, or body temperature; S103. Input the user interaction problem information into a pre-trained natural language processing model to analyze the implicit emotions in the text, and combine the user's physiological signals to generate the current hidden emotional characteristics of the user interaction through a pre-trained convolutional neural network model.

3. The data interaction method based on artificial intelligence according to claim 1, wherein The specific steps of step S2 include the following steps: S201. Collect historical conversation data, including the user's historical interaction problem information, physiological signal data, and the corresponding conversation emotional category labels as a training set, and obtain an emotion prediction model after training with a time series analysis model; S202. Input the current hidden emotional characteristics into the emotion prediction model to predict the emotional categories that the user will show in the next 3 rounds of conversations, and obtain the emotional categories of the next 3 rounds of conversations.

4. A data interaction method based on artificial intelligence according to claim 1, characterized in that The specific steps of step S3 include the following steps: S301. Analyze and obtain semantic keywords from the user interaction problem information through a natural language processing model, and organize and classify the semantic keywords; S302. According to the extracted semantic keywords, retrieve the output information related to the semantic keywords from a pre-constructed database; S303. Organize and combine the retrieved relevant output information to form multi-segment output information to be output. Among them, each segment of output information to be output corresponds to one or more semantic keywords.

5. A data interaction method based on artificial intelligence according to claim 1, characterized in that The specific steps of step S4 include the following steps: S401. Build an emotion keyword library, extract keywords from the multi-segment output information to be output, and obtain the number of single-segment keywords for multiple segments; S402. Statistically analyze the overlap degree, i.e., the correlation degree, between the multi-segment information to be output and the emotion keyword library, and obtain the correlation degree of each segment of the information to be output in the multi-segment information to be output. S403. According to the numerical values of the correlation degrees of each segment of the information to be output, sort each segment of the information to be output in descending order to generate a correlation degree sorting result.

6. The data interaction method based on artificial intelligence according to claim 1, wherein The specific steps of step S5 include the following steps: S501. According to the correlation degree sorting result, set the candidate level division criteria, including: classifying the information to be output with the top 30% of the correlation degree rankings as first-level candidates, those ranked 30%-60% as second-level candidates, and those ranked 60% and later as third-level candidates. S502. Based on the set candidate level division criteria, classify each segment of the multi-segment information to be output into candidate levels, that is, mark each segment of the information to be output with the corresponding candidate level.

7. A data interaction method based on artificial intelligence according to claim 1, characterized in that, The specific steps of step S6 include the following steps: S601. At the beginning of the interaction, according to the candidate levels of the multi-segment information to be output, select the first-level candidate information to be output as the current primary output information and display it to the user through the corresponding output device. S602. Based on the predicted emotion categories of the next 3 rounds of conversations, preset a candidate output information buffer pool, where the candidate output information buffer pool includes the information to be output of the second-level candidates and the information to be output of the third-level candidates, and is used to supplement the current primary output information. S603. During the interaction process, real-time monitor and obtain the current user interaction problem information, and analyze the current hidden emotion features of the user interaction by the same method as in step S1, identify its emotion category, and compare the current hidden emotion features of the user interaction with the predicted emotion categories of the next 3 rounds of conversations in step S2 to determine whether they match. S604. If it is determined that the current hidden emotion features of the user interaction monitored in real time match the predicted emotion categories of the next 3 rounds of conversations, retrieve the corresponding response information from the candidate output information buffer pool in the preset order for information supplementation; if the current hidden emotion features of the user interaction monitored in real time do not match the predicted emotion categories of the next 3 rounds of conversations, re-analyze the user interaction problem information and physiological signals to obtain new user interaction problem information and physiological signals, and then execute the operations of steps S2-S5 again.

8. An artificial intelligence-based data interaction system, which is applied to an artificial intelligence-based data interaction method according to any one of claims 1-7, characterized in that, The interactive system includes: A data collection module, which is used to obtain user interaction problem information, and at the same time obtain user physiological signals through wearable devices, and analyze the hidden emotion features of the user interaction based on the user interaction problem information and user physiological signals. A feature analysis module, which is used to predict the emotion categories of the next 3 rounds of conversations based on the hidden emotion features of the user interaction. A feature matching module, which is used to analyze and obtain semantic keywords from the user interaction problem information, retrieve and match the corresponding output information from the database, and form multi-segment information to be output. An association analysis module, which is used to perform association analysis and sorting on the multi-segment information to be output based on the predicted emotion categories of the next 3 rounds of conversations, and obtain an association degree sorting result. A grading module, which is used to divide the candidate grades of multi-segment information to be output according to the relevance sorting result. Among them, the candidate grades include first-level candidates, second-level candidates, and third-level candidates; An adjustment output module, which is used to select the first-level candidate information to be output as the current primary output according to the candidate grades of the multi-segment information to be output, and preset a candidate output information buffer pool based on the predicted emotional categories of the next 3 rounds of conversations. It obtains the current user interaction problem information in real time and analyzes the current hidden emotional characteristics of the user interaction. If the current hidden emotional characteristics of the user interaction match the predicted emotional categories of the next 3 rounds of conversations, it retrieves a response from the buffer pool. If they do not match, it re-analyzes the user feedback information and physiological signals and then executes the operations in steps S2 - S5 to dynamically adjust the output.

Images