Human-machine interaction processing method, human-machine interaction processing system, electronic device, and program product

Abstract

The present disclosure discloses a human-computer interaction processing method and a human-computer interaction processing system, the human-computer interaction processing method comprising: generating, by a first intelligent agent, structured question bank data according to knowledge content and a plurality of virtual characters, the structured question bank data comprising question data corresponding to the plurality of virtual characters respectively and standard answer data corresponding to the question data; outputting, by a second intelligent agent, target question data corresponding to a target virtual character according to the structured question bank data, the target virtual character being selected from the plurality of virtual characters; and in response to obtaining user input data, performing, by a third intelligent agent, multi-dimensional analysis and processing on the user input data based on the target virtual character and target standard answer data corresponding to the target question data, to generate evaluation data. The human-computer interaction processing method according to the embodiments of the present disclosure can perform multi-dimensional dynamic evaluation on user input based on virtual characters.

Description

Technical Field

[0001] This disclosure relates to the field of computer application technology, specifically to a human-computer interaction processing method, a human-computer interaction processing system, an electronic device, and a program product. Background Technology

[0002] Currently, with the development of artificial intelligence technology, intelligent training systems that apply artificial intelligence technology to the training field have been proposed.

[0003] However, with the diversification of application needs, people expect that intelligent training systems can be further improved and perfected in terms of training effectiveness and intelligence level.

[0004] The background description is provided for the purpose of understanding the relevant technologies in this field and is not intended as an admission of prior art. Summary of the Invention

[0005] The embodiments disclosed herein aim to provide a human-computer interaction processing method, a human-computer interaction processing system, an electronic device, and a program product, which can improve the accuracy of evaluation results and the dynamic adaptability of dialogue generation in human-computer interaction processing, thereby improving the training effect of intelligent training systems.

[0006] In a first aspect, embodiments of this disclosure provide a human-computer interaction processing method, comprising:

[0007] Based on the knowledge content and multiple virtual characters, a structured question bank data is generated by a first intelligent agent. The structured question bank data includes question data corresponding to each of the multiple virtual characters and standard answer data corresponding to the question data.

[0008] Based on the structured question bank data, the second intelligent agent outputs the target question data corresponding to the target virtual character, wherein the target virtual character is selected from the plurality of virtual characters;

[0009] In response to acquiring user input data, a third intelligent agent performs multi-dimensional analysis and processing on the user input data based on the target virtual character and the target standard answer data corresponding to the target question data, and generates evaluation data.

[0010] In some embodiments, the human-computer interaction processing method further includes:

[0011] The multimodal raw data is parsed and processed to extract the knowledge content, wherein the multimodal raw data includes any one or more combinations of document data, image data, audio data, and video data; and the knowledge content includes any one or more combinations of text semantic features, image content features, and audio text features.

[0012] In some embodiments, the parsing and processing of the multimodal raw data to extract the knowledge content includes any one or a combination of the following processes:

[0013] The document data is subjected to text semantic analysis to obtain text semantic features;

[0014] Image recognition processing is performed on the image data and / or video data to obtain image content features;

[0015] The audio data and / or video data are processed by speech recognition to obtain audio-text features.

[0016] In some embodiments, before outputting the target question data, the method further includes:

[0017] In some embodiments, generating structured question bank data through a first intelligent agent based on knowledge content and multiple virtual characters includes:

[0018] Obtain preset generation parameters, wherein the preset generation parameters include a number of questions parameter and a difficulty level parameter;

[0019] The first intelligent agent generates structured question bank data that conforms to the preset generation parameters.

[0020] In some embodiments, generating structured question bank data through a first intelligent agent based on knowledge content and multiple virtual characters includes:

[0021] The first intelligent agent generates candidate question bank data;

[0022] The candidate question bank data is subjected to compliance verification processing based on preset compliance rules to obtain compliant structured question bank data.

[0023] In some embodiments, the virtual character is a virtual practice character;

[0024] The process of generating structured question bank data through a first intelligent agent based on knowledge content and multiple virtual roles includes:

[0025] Determine the target customer group corresponding to the virtual training characters;

[0026] Based on the target customer characteristic data corresponding to the target customer group, generate question data corresponding to the virtual training character and standard answer data corresponding to the question data.

[0027] In some embodiments, generating question data corresponding to the virtual practice character and standard answer data corresponding to the question data based on target customer characteristic data corresponding to the target customer group includes:

[0028] Determine the question style parameters based on the target customer characteristic data;

[0029] The style characteristics of the question data are controlled according to the question style parameters.

[0030] In some embodiments, the method further includes:

[0031] Obtain the user's voice input data;

[0032] The voice input data is processed by speech recognition to obtain transcribed text data;

[0033] The voice input data is subjected to voice feature extraction processing to obtain voice feature data;

[0034] The user input data includes the transcribed text data and the speech feature data;

[0035] The step of generating evaluation data by having a third intelligent agent perform multi-dimensional analysis and processing of the user input data based on the target virtual character and the target standard answer data corresponding to the target question data, includes:

[0036] Content dimension analysis is performed on the transcribed text data to obtain content analysis results;

[0037] The speech feature data is subjected to speech dimension analysis to obtain speech analysis results.

[0038] In some embodiments, the method further includes:

[0039] Identify a first dimension in the evaluation data where the user's performance score is higher than a preset first threshold, and generate first feedback data based on the first dimension;

[0040] Identify a second dimension in the evaluation data where the user's performance score is lower than a preset second threshold, and generate second feedback data based on the second dimension;

[0041] Personalized suggestions are generated for the user based on the first feedback data and / or the second feedback data.

[0042] In a second aspect, embodiments of this disclosure provide a human-computer interaction processing system, comprising:

[0043] The generation module is configured to generate structured question bank data through a first intelligent agent based on knowledge content and multiple virtual roles. The structured question bank data includes question data corresponding to each of the multiple virtual roles and standard answer data corresponding to the question data.

[0044] The output module is configured to output target question data corresponding to the target virtual character through a second intelligent agent based on the structured question bank data, wherein the target virtual character is selected from the plurality of virtual characters;

[0045] The evaluation module is configured to, in response to receiving user input data, perform multi-dimensional analysis and processing of the user input data by a third intelligent agent based on the target virtual character and the target standard answer data corresponding to the target question data, and generate evaluation data.

[0046] In a third aspect, embodiments of this disclosure provide an electronic device, including: a processor and a memory storing a computer program, the processor being configured to implement the method as described in the first aspect when the computer program is executed.

[0047] In a fourth aspect, embodiments of this disclosure provide a program product including a computer program, wherein the computer program, when executed by a processor, implements the method as described in the first aspect.

[0048] The human-computer interaction processing method of this disclosure generates a structured question bank through a first intelligent agent, comprising question data corresponding to multiple virtual characters and standard answer data corresponding to the question data. A second intelligent agent outputs target question data corresponding to a target virtual character. A third intelligent agent then performs multi-dimensional analysis on the acquired user input data based on the target virtual character and the target standard answer data corresponding to the target question data. This solution solves the technical problems of known solutions, such as question bank content being unrelated to the target object type, lack of precise reference in evaluation, and the inability of the evaluation process to adapt to different character characteristics. This solution establishes a correspondence between question data and virtual characters during the question bank generation stage and simultaneously introduces the standard answer data corresponding to the target virtual character and the target question data as evaluation criteria during the evaluation stage. This allows the evaluation process to be dynamically adjusted based on the current target virtual character, in addition to scoring based on precise question-level standard answers, thereby improving the training targeting and evaluation accuracy in the human-computer interaction process.

[0049] In a further embodiment of this disclosure, a role-based question bank is constructed by configuring virtual characters as virtual practice characters and by determining the target customer group corresponding to the virtual practice characters, generating question data corresponding to the virtual practice characters and standard answer data corresponding to the question data based on the target customer characteristic data corresponding to the target customer group. This further solution solves the technical problems that general question banks cannot reflect the differences between different customer types and that training scenarios lack specificity. By basing the generation of question data and standard answer data on the characteristic data of the target customer group, this solution enables the questions and standard answers in the question bank for different virtual practice characters to present differentiated content that matches the characteristics of the corresponding customer group, thereby supporting users to conduct targeted training for different types of practice characters.

[0050] In a further embodiment of this disclosure, question style parameters are determined based on target customer characteristic data, and the style characteristics of the question data are controlled through these parameters. This solution further addresses the technical problem of indistinguishable question styles among different types of virtual training roles. As an explanation, and not a limitation, different customer groups have different communication tendencies and expression habits. The question style parameters quantify these group differences into controllable parameters, thereby constraining the style performance during question generation. Thus, questions generated under different role configurations within the same system exhibit identifiable differences in style, enabling the simulation of diverse real-world interaction scenarios, avoiding homogenization of virtual character behavior, and enhancing the comprehensiveness and practicality of training.

[0051] Other optional features and technical effects of the embodiments of this disclosure are described in part below, and in part will be apparent from reading this document. Attached Figure Description

[0052] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings. The elements shown are not limited to the scale shown in the drawings, and the same or similar reference numerals in the drawings denote the same or similar elements, wherein:

[0053] Figure 1 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0054] Figure 2 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0055] Figure 3 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0056] Figure 4 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0057] Figure 5 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0058] Figure 6 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0059] Figure 7 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0060] Figure 8 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0061] Figure 9 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0062] Figure 10 An exemplary flowchart of a human-computer interaction processing method according to an embodiment of the present disclosure is shown;

[0063] Figure 11 An exemplary block diagram of a human-computer interaction processing system according to an embodiment of the present disclosure is shown;

[0064] Figure 12 An exemplary architecture diagram of a human-computer interaction processing system according to embodiments of the present disclosure is shown; and

[0065] Figure 13 An exemplary structural diagram of an electronic device that can implement a human-computer interaction processing method according to embodiments of the present disclosure is shown. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this disclosure clearer, the disclosure will be further described in detail below with reference to specific embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this disclosure are used to explain this disclosure, but are not intended to limit this disclosure.

[0067] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. 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.

[0068] The term "comprising" and its variations as used herein signify open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least partially based on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc., may refer to different or the same objects.

[0069] The user data, data acquisition, and / or use involved in the embodiments of this disclosure strictly comply with the laws, regulations, and industry standards of relevant countries and regions. The collection and acquisition of data involved in the embodiments of this disclosure are all done in advance by actively prompting or prominently displaying information to inform users and obtaining authorization, or by obtaining full authorization from all parties. The processing, manipulation, forwarding, and use of data involved in the embodiments of this disclosure are all carried out on the premise that the user or relevant party is fully informed and authorized. When implementing the embodiments of this disclosure, the types of data or information, scope of use, and usage scenarios that may be involved are informed to users or relevant parties and authorization is obtained through appropriate means. The specific methods of notification and authorization may vary according to actual circumstances, and this disclosure is not limited in this regard. The processing of personal information involved in the embodiments of this disclosure is carried out under the premise of having a legal basis (such as obtaining the consent of the personal information subject or being necessary for the performance of a contract), and is only processed within the prescribed or agreed scope. Sensitive personal information such as biometric information, medical and health information, financial account information, and precise location information involved in the embodiments of this disclosure are all processed under the premise of having a specific purpose and sufficient necessity, and with the separate authorization and consent of the user or relevant party. In some embodiments of this disclosure, if a user or related party refuses to process personal information other than the information necessary for the basic functions, it will not affect the use of the basic functions of the embodiments of this disclosure.

[0070] In the embodiments of this disclosure, the "target customer characteristic data" and other data mentioned are all anonymized group profile data obtained after statistical analysis and processing of historical business data. This data only reflects the statistical characteristic distribution of the customer group and does not contain any personal information that can identify a specific natural person, nor does it correspond to any specific natural person. In the embodiments of this disclosure, the attribute information and behavioral characteristics of the virtual training characters are all virtual data for simulating training scenarios, synthesized based on group statistical characteristics, and do not represent or correspond to any real natural person.

[0071] As mentioned earlier, solutions for applying artificial intelligence (AI) technology to intelligent training systems have been proposed in the training field. Therefore, how to leverage human-computer interaction (HCI) technology, especially by using intelligent agents to drive virtual characters to interact with users and analyze and evaluate user performance, is a technological direction with broad application value. However, in known HCI solutions, significant shortcomings remain regarding the dialogue generation process, the differentiation of virtual character behavior, the depth of evaluation and analysis, and the accuracy of feedback.

[0072] First, known solutions have shortcomings in the relevance of question bank content to the types of practice partners, as well as in the reference criteria and role adaptation of evaluation processing. Specifically, on the one hand, the generation of question bank content in known solutions often does not consider the differences in client types of practice partners, making the same question bank applicable to all practice scenarios, thus limiting the diversity and relevance of practice training. On the other hand, the evaluation process of known solutions is often based on general scoring standards or vague reference criteria, lacking a standard answer that precisely corresponds to the current question as an evaluation benchmark, making it difficult for the evaluation results to accurately reflect the specific differences between the user's answer and the standard answer. In addition, the evaluation processing of known solutions does not consider the impact of the characteristics of the current practice role on the evaluation, making it impossible for the evaluation results to present differences adapted to different roles. For example, known solutions may give a general evaluation such as "good answer" or "answer needs improvement" for a user's answer, but they cannot compare it with precise standards for the current role and question, nor can they adjust the evaluation processing according to the role characteristics.

[0073] In response, some existing solutions have proposed the concept of virtual characters. However, these solutions suffer from significant homogenization of their virtual characters. Specifically, some existing solutions simply mechanically divide a generic question bank for multiple virtual characters, resulting in a largely uniform interaction pattern across these characters. Understandably, in real-world interaction scenarios, different types of users have different communication styles, focuses, and decision-making tendencies—conservative users may be more concerned with risk factors, exhibiting a cautious and hesitant dialogue style; while proactive users may be more focused on expected returns, with a more direct and rapid dialogue pace. However, the virtual characters in existing solutions fail to reflect these differences, significantly diminishing the training value of the interaction process. Therefore, this disclosure recognizes that how to enable virtual characters to exhibit differentiated behaviors at the dialogue level that match the characteristics of the group they simulate, thereby helping users to more effectively adjust their expression strategies, is a problem that current intelligent training systems urgently need to solve.

[0074] Secondly, known solutions have significant limitations in analyzing user input data dimensions, primarily relying on speech recognition technology to convert speech into text for analysis. While speech recognition processing can convert speech to text, and the transcribed text can be used for content-level analysis (such as accuracy of expression and logical structure), this disclosure notes that speech signals also contain multiple dimensions of information, including timbre, intonation, speech rate, and emotion. The same literal information can express opposite tendencies under different intonations, and known speech recognition processing lacks the ability to analyze these dimensions of data. Known solutions evaluate based solely on transcribed text, resulting in evaluation results lacking analysis of the speech expression dimension and failing to comprehensively reflect user interaction performance.

[0075] To address this, this disclosure proposes a human-computer interaction processing method and a corresponding human-computer interaction processing system. The human-computer interaction processing method of this disclosure can be applied to intelligent training scenarios, such as human-computer interaction training and training assessment scenarios in the financial field. By way of explanation and not limitation, in such scenarios, the training subjects may include product sales personnel, account managers, or financial advisors, etc. This method can be used to simulate multi-round dialogue interactions between virtual customers and training subjects, and generate multi-dimensional evaluation data and feedback information for the training subjects based on the dialogue process, in order to assist in improving training effectiveness and assessing capabilities.

[0076] In some embodiments, the human-computer interaction processing method of this disclosure may be executed by human-computer interaction processing systems including but not limited to intelligent training systems, intelligent training evaluation systems, and virtual customer interaction training systems. These human-computer interaction processing systems may be deployed on servers, cloud computing platforms, or computing nodes within enterprise intranets and work in collaboration with user terminals.

[0077] The following will describe in detail the specific implementation methods of the human-computer interaction processing method and human-computer interaction processing system disclosed herein, with reference to the accompanying drawings.

[0078] In this embodiment of the disclosure, reference is made to Figure 1 A human-computer interaction processing method is provided, which may include the following optional steps 110, 120, 130, 140 and 150.

[0079] 110: Parse and process the multimodal raw data to extract knowledge content.

[0080] In some embodiments, the multimodal raw data may include any one or more combinations of document data, image data, audio data, and video data.

[0081] In some embodiments, document data may include, but is not limited to, PDF documents, word processing documents, presentation documents, etc.; image data may include, but is not limited to, product illustration images, flowchart images, table screenshots, etc.; audio data may include, but is not limited to, training recordings, product introduction audio, etc.; video data may include, but is not limited to, training videos, product explanation videos, and other audio and video content.

[0082] In some embodiments, multimodal raw data can be uploaded to the system through the management configuration interface. The system can automatically identify the type of the uploaded file and call the corresponding parsing and processing method for processing.

[0083] In some embodiments, the raw multimodal data may be parsed to extract knowledge content. In some embodiments, the knowledge content may include any one or more combinations of textual semantic features, image content features, and audio text features.

[0084] In some embodiments, the parsing process may include any one or a combination of the following processes: performing text semantic analysis on document data to obtain text semantic features; performing image recognition on image data or video data to obtain image content features; and performing speech recognition on audio data or video data to obtain audio text features.

[0085] In some embodiments of this disclosure, reference is made to Figure 2 Step 110 may include the following steps 210, 220 and 230.

[0086] 210: Perform text semantic analysis on the document data to obtain text semantic features.

[0087] In some embodiments, text semantic analysis processing may include semantic parsing and understanding of the text content in document data. In some embodiments, text semantic analysis processing may include automatically extracting product knowledge points, key information, and structured content from document data.

[0088] In some embodiments, text semantic analysis processing may include operations such as segmenting document data, extracting keywords, identifying topics, and summarizing knowledge points to obtain structured text semantic features. In some embodiments, text semantic features may include product knowledge points, descriptions of core concepts, definitions of key terms, and related wording information.

[0089] 220: Perform image recognition processing on image or video data to obtain image content features.

[0090] In some embodiments, image recognition processing may include recognizing and extracting visual information contained in image data. In some embodiments, image recognition processing may include recognizing text content, chart information, flowchart structures, and other visual elements in image data.

[0091] In some embodiments, for video data, image recognition processing can identify the image content in video frames and extract the knowledge content presented in the video. In some embodiments, image content features may include text information, chart data, structured content descriptions, etc., identified from images or video frames. In some embodiments, image recognition processing can be based on optical character recognition technology and image content understanding technology to extract and structure the various types of visual information in image data.

[0092] 230: Perform speech recognition processing on audio or video data to obtain audio-text features.

[0093] In some embodiments, speech recognition processing may include converting speech content in audio or video data into text data. In some embodiments, audio-text features may include transcribed text information identified from audio or video speech content.

[0094] In some embodiments, speech recognition processing includes extracting audio channel data from the video and performing speech recognition processing on the extracted audio data to obtain text data corresponding to the speech content in the video.

[0095] In this embodiment of the disclosure, knowledge content can be extracted from multimodal raw data through at least one of the above steps 210, 220, and 230. In some embodiments, feature information from different modalities can be integrated to form unified structured knowledge content. As an explanation and not a limitation, the parsing and processing of multimodal raw data can achieve unified processing and knowledge extraction of learning materials in various formats, thereby providing comprehensive knowledge content data for subsequent structured question bank data.

[0096] 120: Based on the knowledge content and multiple virtual characters, generate structured question bank data through the first intelligent agent.

[0097] In some embodiments, such as step 120 above, the first agent may include a question bank generation agent.

[0098] In some embodiments, the structured question bank data may include question data corresponding to multiple virtual characters and standard answer data corresponding to the question data. By way of explanation and not limitation, the question data may be used to output questions to the user as questions to be answered by a second intelligent agent, and the standard answer data may be used as a scoring benchmark in subsequent evaluation steps.

[0099] In some embodiments, multiple virtual characters may be preset. In some embodiments, a virtual character involves a data structure including a character identifier and preset character characteristic data associated with the character identifier. In this embodiment, the character identifier can be used to distinguish different virtual characters, and the preset character characteristic data can be used to characterize the attribute characteristics and / or behavioral characteristics of the object simulated by the corresponding virtual character.

[0100] In some embodiments, each virtual character corresponds to a preset character characteristic data, and the preset character characteristic data corresponding to multiple virtual characters are at least partially different. By way of explanation and not limitation, the preset character characteristic data may include at least one of risk preference, focus, expression habits, and communication style.

[0101] In some embodiments, a first intelligent agent may generate question data corresponding to a virtual character and standard answer data corresponding to the question data based on preset character feature data corresponding to the virtual character. For explanation, the question data in the structured question bank data can be organized according to the corresponding virtual character, so that a second intelligent agent can subsequently determine and output target question data corresponding to the target virtual character from the structured question bank data based on the determined target virtual character.

[0102] In some embodiments, the correspondence between question data and virtual characters can be represented by association information. In this embodiment, the association information may include, for example, character identifiers recorded in the question data, pre-established mapping records between question data and character identifiers, or index relationships. It is understood that the preset character feature data and structured question bank data can be stored separately or stored together. This disclosure does not limit the specific storage format of the two, as long as the corresponding virtual character can be determined based on the question data. In the above embodiments of this disclosure, the specific methods for determining virtual characters and their corresponding customer groups, as well as the specific generation mechanism and style control of question data, will be described in detail in steps 310 and 320, and steps 410 and 420 below.

[0103] In some embodiments, the first agent may be configured to receive the knowledge content extracted in step 110 above, and automatically generate structured question bank data based on the knowledge content and multiple virtual characters. For explanation purposes, and not as a limitation, the first agent may extract core knowledge points and key information from the knowledge content, and then generate questions and corresponding standard answers around these knowledge points to assess the user's mastery of the corresponding knowledge, and associate these questions and corresponding standard answers with one of the multiple virtual characters respectively.

[0104] In some embodiments, the standard answer data may include the standard answer and answer key points. In this embodiment, the standard answer may include the complete answer text corresponding to the question, and the answer key point list may include a set of key knowledge points extracted from the standard answer content. For explanation, combining the standard answer and answer key points helps to perform a more granular key point matching evaluation of user answers during subsequent multi-dimensional analysis and processing.

[0105] In some embodiments, the structured question bank data may further include difficulty tags and question type tags. In this embodiment, difficulty tags can be used to identify the difficulty level of each question, and question type tags can be used to identify the type category to which the question belongs.

[0106] In some embodiments, the difficulty label may be determined based on the complexity of the knowledge points and the expression requirements. By way of explanation and not limitation, the complexity of the knowledge points may reflect the depth and breadth of product knowledge tested by the question, and the expression requirements may reflect the organizational and expressive abilities required by the user to answer the question.

[0107] In some embodiments, question type labels may include product introduction type, objection handling type, and scenario response type. In this embodiment, the product introduction type can be used to assess a user's ability to describe product functions and features, the objection handling type can be used to assess a user's communication skills when responding to customer inquiries, and the scenario response type can be used to assess a user's comprehensive adaptability in specific business scenarios.

[0108] In some embodiments, the virtual character is a virtual practice character. In this embodiment, reference... Figure 3 Step 120 may include steps 310 and 320 below.

[0109] 310: Determine the target customer group corresponding to the virtual practice characters.

[0110] In some embodiments, a mapping relationship can be established between virtual training roles and target customer groups (types). In embodiments of this disclosure, target customer characteristic data can be generated based on authorized group-level statistical data and analytical data. It is understood that target customer characteristic data can be used to characterize the statistical characteristic distribution of the corresponding target customer group across multiple attribute dimensions, but does not include personally identifiable information that can identify a specific natural person, nor does it establish a correspondence with any specific natural person. By way of explanation and not limitation, by associating virtual training roles with target customer groups, the behavioral patterns, expressions, and concerns of the virtual training roles can reflect the characteristics of the real customer group, thereby improving the authenticity and relevance of dialogue training.

[0111] In some embodiments, the target customer group can be determined based on parameter values ​​and / or parameter ranges corresponding to multiple customer attribute dimensions. In some embodiments, customer attribute dimensions may include any one or more combinations of information familiarity, focus preference, expression habits, communication style, and information confirmation preference. In this embodiment, different target customer groups have different parameter values ​​or parameter ranges corresponding to at least one customer attribute dimension.

[0112] In some embodiments, the target customer group can be determined based on at least one of the following parameters: business participation duration, role background type, and virtual resource holdings. In some embodiments, business participation duration can be divided into short-term, medium-term, and long-term intervals; role background type can include management, technical, and operational backgrounds; and virtual resource holdings can be divided into high, medium, and basic holdings. As an explanation, and not a limitation, different target customer groups typically differ in their product feature requirements, virtual resource allocation preferences, and communication styles. Determining the target customer group based on the above attribute dimensions allows for the coverage of diverse customer types.

[0113] In some embodiments, target customer characteristic data can be obtained based on statistical analysis of historical business data. In some embodiments, existing business record data can be aggregated and statistically analyzed to extract typical characteristic distributions of various customer groups (types), thereby providing group profile data for the construction of virtual practice roles. In this embodiment, the characteristic distribution includes, for example, average risk preference and common concerns distribution. In some embodiments, the statistical analysis process may include steps such as data aggregation, characteristic distribution calculation, and group profile generation. By way of explanation and not limitation, the statistical analysis process may output target customer characteristic data that does not contain any personally identifiable information.

[0114] In some embodiments, a preset customer group database may be maintained, which comprises a structured dataset including multiple customer group configuration records. In some embodiments, the preset customer group database may include various predefined customer groups (types). In this embodiment, the target customer group may be randomly selected from the preset customer group database or selected according to a preset strategy. In some embodiments, random selection may be used to increase the diversity of dialogue scenarios, or selection according to a preset strategy may be used to conduct targeted practice for specific training objectives. In one example, the preset strategy may prioritize customer groups that the user has had less contact with based on the user's historical practice history.

[0115] In some embodiments, target customer characteristic data can be generated based on a combination of parameter values ​​from multiple attribute dimensions, each of which includes multiple optional parameter values. By way of explanation and not limitation, combining parameter values ​​from multiple attribute dimensions such as business participation duration, role background type, and virtual resource holdings can generate a rich variety of target customer characteristic data. In other words, the target customer characteristic data can characterize the attribute parameter combinations of a corresponding target customer group. In a specific example, virtual practice role 1 is set as a customer group combination of "long-term business participation, management background, and high virtual resource holdings," and virtual practice role 2 is set as a customer group combination of "medium-term business participation, technical background, and medium virtual resource holdings."

[0116] 320: Based on the target customer characteristic data corresponding to the target customer group, generate question data corresponding to the virtual practice role and standard answer data corresponding to the question data.

[0117] In some embodiments, such as in step 320 above, the first intelligent agent can be driven to generate question data and corresponding standard answer data that conform to the characteristics of the customer group based on the target customer feature data determined in step 310. In some embodiments, the first intelligent agent can generate question data that matches the characteristics of the customer group and standard answer data corresponding to the question data (question content) based on the customer attribute information and behavioral characteristics contained in the target customer feature data. In this embodiment, customer attribute information and behavioral characteristics may include, but are not limited to, the questioning style, focus of attention, and expression habits of the virtual practice character.

[0118] In some embodiments, reference Figure 4 Step 320 may include steps 410 and 420 below.

[0119] 410: Determine the question style parameters based on the target customer characteristic data.

[0120] In some embodiments, style parameters can be extracted and determined from target customer characteristic data as question style parameters. In this embodiment, question style parameters can be used to define the expression style and interaction characteristics of question data corresponding to virtual practice roles. As an explanation, and not a limitation, different customer groups typically exhibit different style characteristics in communication due to differences in attributes such as business participation duration, role background type, and virtual resource holdings. Therefore, question style parameters that reflect these differences can be determined by analyzing target customer characteristic data.

[0121] In some embodiments, the question style parameter may include a style type identifier and a style intensity value. In some embodiments, the style type identifier can be used to indicate the specific type of question data style; the style intensity value can be used to control the degree to which that style type is expressed in the question data. For explanation, the combination of style type identifier and style intensity value allows for fine-grained control over the question data style.

[0122] 420: Control the style characteristics of the question data according to the question style parameters.

[0123] In some embodiments, such as in step 420 above, the style characteristics of the question data output by the first agent can be controlled based on the question style parameters determined in step 410. In some embodiments, the first agent can adjust the style characteristics of the question data (question content) according to the question style parameters when outputting question data corresponding to virtual practice roles. As an explanation, this scheme allows the output questions to reflect the communication characteristics of real customer groups.

[0124] In some embodiments, style features may include at least one of content features, expression features, phonetic features, and tone features. In some embodiments, content features may include focus features and word choice features; tone features may include attitude features and emotion features; expression features may include sentence structure features and rhythm features; and phonetic features may include intonation features, timbre features, and speech rate features.

[0125] In some embodiments, the first intelligent agent can determine the specific parameter values ​​of each of the above style features based on the group attribute information in the target customer feature data, and generate question content that matches the group-level behavioral characteristics of the corresponding target customer group based on the combination of the values ​​of the above style feature parameters.

[0126] As an explanation, and not a limitation, in product service training scenarios, for target customer groups with a high amount of virtual resources, the focus parameters in the content features can lean towards the combination, configuration, and security of virtual resource usage. For target customer groups with a basic level of virtual resources, the focus parameters can lean towards the efficiency of acquiring virtual resources and the conditions for participation. For target customer groups with a technical background, the word choice feature parameters can include a tendency towards technical terminology. For target customer groups with a non-technical background, the word choice feature parameters can include a tendency towards colloquial language.

[0127] In some embodiments, in terms of tone characteristics, attitude characteristics can be reflected as a cautious interaction tendency corresponding to a more conservative feedback style, while a quick decision-making interaction tendency corresponds to a feedback style that makes decisions quickly.

[0128] In some embodiments, in terms of emotional characteristics, a low-certainty expression tendency may manifest as strong uncertainty in a conversation, while a high-certainty expression tendency may correspond to clear and explicit expression.

[0129] In some embodiments, in terms of expression features, sentence structure features can be reflected in different sentence structure preference parameters corresponding to the use of long sentences for detailed description and the use of short sentences for concise expression.

[0130] In some embodiments, in terms of rhythmic features, different interaction rhythm parameters can correspond to a relatively gentle and a relatively rapid dialogue rhythm, respectively.

[0131] In some embodiments, in terms of speech features, the preset intonation parameters, timbre parameters, and speech rate parameters corresponding to different target customer groups may be different, thereby making the corresponding speech differ in terms of intonation, timbre, and speech rate.

[0132] In the embodiments described above, by determining the question style parameters based on target customer characteristic data and controlling the style characteristics of the question data through these parameters, different types of virtual characters can correspond to different (dialogue) styles. As an explanation, and not a limitation, different real customer groups have different communication tendencies and expression habits. The style parameters quantify these group differences into controllable parameters, thereby constraining the style performance during dialogue generation. Therefore, questions corresponding to different character configurations within the same system exhibit identifiable differences in style, enabling more realistic and diverse dialogue performances of virtual practice characters.

[0133] In some embodiments, step 120 may further include a parameterization generation process. In this embodiment, reference... Figure 5 Step 120 may include steps 510 and 520 below.

[0134] 510: Get the preset generation parameters.

[0135] In the embodiments of this disclosure, preset generation parameters can be used to control the specific output specifications when the first intelligent agent generates question bank data.

[0136] In some embodiments, the preset generation parameters may include a question quantity parameter and a difficulty level parameter. In this embodiment, the question quantity parameter may specify the total number of questions the first agent needs to generate for the current knowledge content; the difficulty level parameter may specify the overall difficulty level of the generated questions. As an explanation, and not a limitation, by setting different preset generation parameters, the same batch of knowledge content can generate structured question bank data of different specifications, thereby adapting to different training scenarios and user groups.

[0137] In some embodiments, the number of questions parameter can take values ​​including, but is not limited to, 10, 20, and 30 questions; the difficulty level parameter can take values ​​including, but is not limited to, high difficulty, medium difficulty, and low difficulty. As an explanation, and not a limitation, different numbers of questions can be adapted to training tasks of different durations; for example, 10 questions are suitable for rapid testing scenarios, while 30 questions are suitable for comprehensive assessment scenarios. Similarly, different difficulty levels can be adapted to user groups with different ability levels; for example, high difficulty is suitable for advanced assessments for experienced users, while low difficulty is suitable for beginners to familiarize themselves with the material.

[0138] In some embodiments, preset generation parameters can be obtained through the management terminal configuration interface. By way of explanation and not limitation, the management terminal user can select or input the required number of questions and difficulty level parameters through the management terminal configuration interface. The system can then pass the selected parameters to the first intelligent agent, which generates structured question bank data that meets the parameter requirements.

[0139] 520: Generate structured question bank data that conforms to preset generation parameters through the first intelligent agent.

[0140] In this embodiment, the first intelligent agent, after receiving preset generation parameters, can automatically generate corresponding structured question bank data by combining the aforementioned knowledge content and according to the number of questions specified by the number of questions parameter and the difficulty level specified by the difficulty level parameter. In this embodiment, the preset generation parameters are, for example, the preset generation parameters obtained through step 510. As an explanation, and not a limitation, the first intelligent agent can adjust the depth of knowledge point coverage and the complexity of expression requirements of the questions according to the difficulty level parameter, so that the overall difficulty of the generated questions matches the preset difficulty level. Simultaneously, the number of questions generated by the first intelligent agent strictly adheres to the set value of the number of questions parameter.

[0141] In some embodiments, step 120 may further include a compliance verification process. In this embodiment, refer to... Figure 6 Step 120 may include steps 610 and 620 below.

[0142] 610: Generate candidate question bank data through the first intelligent agent.

[0143] In this embodiment of the disclosure, the candidate question bank data includes initial version question bank data that has not yet undergone compliance verification. In this embodiment, the candidate question bank data includes, for example, all question data and standard answer data automatically generated by the first intelligent agent based on knowledge content. By way of explanation and not limitation, the generation process of the candidate question bank data focuses on the coverage of knowledge points and the quality of questions. At this stage, compliance of the content may not be constrained to ensure the diversity and comprehensiveness of the generated questions.

[0144] 620: Based on preset compliance rules, the candidate question bank data is subjected to compliance verification processing to obtain compliant structured question bank data.

[0145] In some embodiments, the candidate question bank data may undergo compliance verification. In this embodiment, the candidate question bank data is, for example, the candidate question bank data generated in step 610 above.

[0146] In some embodiments, compliance verification processing can be performed based on preset compliance rules. For explanation, compliance verification processing can detect whether there are non-compliant expressions, inappropriate terms, or sensitive content in the candidate question bank data, and process the detected non-compliant content accordingly, thereby ensuring the compliance and security of the question bank data.

[0147] In some embodiments, preset compliance rules may include a list of prohibited words, a list of sensitive words, and a standardized expression mapping table. In some embodiments, the standardized expression mapping table may include a mapping relationship between sensitive words and compliant alternative words. For explanation, the standardized expression mapping table may specify a corresponding compliant alternative expression for each sensitive word, so that non-compliant sensitive words are replaced with compliant expressions during the verification process.

[0148] In some embodiments, compliance verification processing may include at least one of lexical matching verification and semantic compliance verification. In some embodiments, semantic compliance verification may perform in-depth analysis of the semantic content of candidate question bank data to identify expressions that, although not directly using prohibited or sensitive words, pose compliance risks at the semantic level.

[0149] In some embodiments, if the compliance verification process indicates the presence of non-compliant content, the non-compliant content can be adjusted. In this embodiment, the adjustment process may include replacing sensitive words based on a canonical expression mapping table.

[0150] In some embodiments, before outputting the target question data, the human-computer interaction processing method may further include the following steps: in response to a received optimization instruction, optimizing and adjusting the structured question bank data to obtain optimized structured question bank data.

[0151] In some embodiments, the optimization and adjustment process can be performed before or during the output of the target question data. In one example, the optimization and adjustment process adjusts the output question data during the dialogue in which the target question data is output.

[0152] In some embodiments, the optimization process can be performed by a fourth agent. In some embodiments, the fourth agent may include a question bank optimization agent. By way of explanation and not limitation, the fourth agent may be configured to fine-tune the structured question bank data, which is used to understand the semantic intent of the optimization instructions and perform targeted optimization processing on the structured question bank data accordingly.

[0153] In some embodiments, the optimization instructions include preset optimization instructions.

[0154] In some embodiments, a preset set of optimization instructions may be maintained, and the preset optimization instructions can be selected from this set. In some embodiments, the preset set of optimization instructions may include at least one of answer simplification instructions, difficulty adjustment instructions, and description optimization instructions. In this embodiment, the answer simplification instruction can be used to instruct the fourth agent to simplify the content of the standard answer data; the difficulty adjustment instruction can be used to instruct the fourth agent to adjust the difficulty level of the question data; and the description optimization instruction can be used to instruct the fourth agent to refine and optimize the textual description of the question data or standard answer data. As an explanation, and not a limitation, the management user can select one or more preset optimization instructions from the preset set of optimization instructions and send them to the fourth agent. The fourth agent can then perform corresponding optimization operations on the corresponding structured question bank data according to the received preset optimization instructions.

[0155] In some embodiments, the optimization instructions also include manual editing instructions. In some embodiments, manual editing instructions can be generated based on the user's input editing intent, and the fourth intelligent agent can intelligently optimize and adjust the structured question bank data based on the manual editing instructions. By way of explanation and not limitation, unlike the aforementioned preset optimization instructions, manual editing instructions include personalized optimization requests input by the management user in natural language. In one specific embodiment, the manual editing instruction is "make the answer more concise."

[0156] In some embodiments, the processing of manual editing instructions may include, but is not limited to, the following steps: receiving editing intent text; performing semantic understanding of the editing intent text through a fourth intelligent agent to determine the user's optimization needs; and having the fourth intelligent agent optimize and adjust the structured question bank data according to the optimization needs. As an explanation, in this embodiment, the management end user can flexibly express various personalized optimization needs through manual editing instructions, without being limited by a preset set of optimization instructions.

[0157] In some embodiments, the optimization and adjustment process may include at least one of the following: adjustment of question (data) content, adjustment of standard answer (data) content, and adjustment of question (data) difficulty level. In this embodiment, question content adjustment may include modifying the way the question is phrased, the angle of examination, or the wording of the question; answer content adjustment may include modifying the level of detail, the scope of key points covered, or the expression of the standard answer; difficulty level adjustment may include adjusting the difficulty label value corresponding to the question and the complexity of the question content and the standard answer. As an explanation, and not a limitation, the management user may trigger one or more of the above optimization and adjustment processes through preset optimization instructions or manual editing instructions. The fourth intelligent agent may determine the specific type of optimization and adjustment to be performed based on the instruction content and execute the corresponding adjustment operation on the corresponding structured question bank data.

[0158] 130: Based on the structured question bank data, the second intelligent agent outputs the target question data corresponding to the target virtual character.

[0159] In some embodiments, such as in step 130 above, the structured question bank data is, for example, the structured question bank generated in step 120 above.

[0160] In some embodiments, the target virtual character can be selected from multiple virtual characters. In this embodiment, the target virtual character can be determined from multiple virtual characters corresponding to structured question bank data by a second intelligent agent. In this embodiment, the second intelligent agent can determine the target virtual character by means including but not limited to random selection, selection according to a preset strategy, or determination based on user selection. In other embodiments, the target virtual character can also be determined from preset (default) virtual characters. This disclosure does not limit the specific method of determining the target virtual character.

[0161] In some embodiments, such as when or after the second agent determines the target virtual character, the second agent may obtain one or more question data corresponding to the target virtual character and standard answer data corresponding to each of the one or more question data from a structured question bank, and select one of the question data as the target question data and output the target question data. In an optional embodiment, the second agent further optimizes the target question data to generate initial dialogue content corresponding to the target question data, so as to present the dialogue scene to the user.

[0162] In some embodiments, user input data may be continuously acquired during the conversation. In this embodiment, user input data may include, but is not limited to, text input data and / or voice input data.

[0163] In an optional embodiment, upon receiving user input data, the second intelligent agent can dynamically determine subsequent target question data for the target virtual character based on the user input data, thereby achieving a multi-turn interactive dialogue experience. As an explanation, and not a limitation, the second intelligent agent can dynamically adjust the subsequently output target question data based on the content of the user input data and the correlation between the target question data and / or standard answer data in the structured question bank data, making the dialogue process closer to a real-world interaction scenario.

[0164] In some embodiments, reference Figure 7 Step 130 may further include steps 710 and 720.

[0165] 710: Generate character attribute display information for virtual practice characters based on target customer characteristic data.

[0166] In some embodiments, character attribute information for displaying to users can be generated based on target customer characteristic data, such as target customer characteristic data corresponding to the target virtual character. By way of explanation and not limitation, displaying the attribute information of the virtual role-playing character to users can help them quickly understand the type of customer they are about to face, thereby allowing them to adjust their communication strategies and responses accordingly.

[0167] In some embodiments, the role attribute display information may include at least one of the following: role avatar, role title, business participation duration tag, role background type tag, and virtual resource holding amount tag. In one example, the role attribute display information includes a virtual avatar representing the characteristics of the customer group, a title identifier "Mr," a corresponding business participation duration tag "long-term participant," a job type tag "management," and an asset size tag "high holding amount."

[0168] 720: Before and / or during the output of the target question data corresponding to the target virtual character, display the character attribute information to the user.

[0169] In some embodiments, the generated character attribute display information may be shown to the user at an appropriate time. In this embodiment, the character attribute display information is, for example, the character attribute display information generated through the aforementioned step 710.

[0170] In some embodiments, character attribute display information may be shown before the target question data is output. In other embodiments, character attribute display information may be continuously displayed during the output of the target question data. For explanation, continuously displaying character attribute information allows users to easily refer to the attribute information of the virtual practice character at any time during the dialogue interaction.

[0171] In some embodiments, the step of displaying character attribute information to the user may include displaying character attribute cards in the front-end dialog interface. By way of explanation and not limitation, character attribute cards can be used to visually present various attribute information of the virtual training character. In a specific example, a card-style component containing information such as avatar, title, age, occupation, and asset tags may be displayed in the sidebar or top area of ​​the dialog interface.

[0172] In some embodiments, such as in step 130 above, while outputting the target question data corresponding to the virtual character through the second intelligent agent, user input data during the dialogue process can also be acquired.

[0173] In some embodiments, the step of acquiring user input data during a conversation may include: acquiring the user's voice input data; processing the voice input data to obtain data to be analyzed for multi-dimensional analysis.

[0174] In some embodiments, when the user selects a voice input method, refer to Figure 8 The step of obtaining user input data during the dialogue process may include the following steps 810, 820 and 830.

[0175] 810: Obtain the user's voice input data.

[0176] In some embodiments, when a user chooses to answer questions from a virtual practice character via voice, the user's voice input data can be acquired through the audio acquisition device of the terminal device.

[0177] 820: Perform speech recognition processing on the voice input data to obtain transcribed text data.

[0178] In some embodiments, speech recognition processing can be performed on the acquired speech input data to convert the speech signal into corresponding text content. In this embodiment, automatic speech recognition technology can be used to extract acoustic features and decode language models from the speech signal in the speech input data, thereby converting the continuous speech signal into a discrete text sequence to obtain transcribed text data. In some embodiments, the speech input data can be processed in real time. Real-time processing enables immediate feedback and a smooth interactive experience during dialogue.

[0179] In some embodiments, after generating the transcribed text data, the human-computer interaction processing method may further include a step of displaying the transcribed text data (preview) to the user. In optional embodiments, editable transcribed text data may be provided to the user. As an explanation, and not a limitation, since speech recognition processing may contain recognition errors, editable transcribed text data allows users to verify and correct the transcription results and ensure the accuracy of the transcribed text data, thereby improving the reliability of subsequent analysis and processing.

[0180] In some embodiments, step 820 described above can identify and transcribe the language content in the speech signal into text form. By way of explanation and not limitation, the output of speech recognition processing is transcribed text data suitable for content-dimensional analysis.

[0181] In a further embodiment of this disclosure, the acoustic features of the speech signal itself (such as timbre, intonation, and speech rate) can be further analyzed. Accordingly, the method may also include the following step 830.

[0182] 830: Perform speech feature extraction processing on the speech input data to obtain speech feature data.

[0183] In some embodiments, feature information related to speech expression can be extracted from the speech input data. For explanation, unlike step 820 described above, step 830 can extract the acoustic and prosodic features of the speech signal itself, rather than the linguistic content carried in the speech. In some embodiments, speech feature extraction processing may include, but is not limited to, acoustic parameter analysis, prosodic parameter extraction, and audio feature modeling of the speech signal.

[0184] In some embodiments, speech feature data may include one or more of timbre features, intonation features, speech rate features, and emotion features. For explanation, timbre features reflect the sound quality characteristics of a user's speech; intonation features reflect the pattern of tone changes in a user's speech; speech rate features reflect the speed and rhythm of a user's speech; and emotion features reflect the emotional state exhibited by the user in a conversation.

[0185] In some embodiments, speech feature data can be used for speech dimension analysis, such as timbre analysis, intonation analysis, speech rate analysis, and emotion analysis, thereby enabling a quantitative assessment of a user's speech expression ability.

[0186] In some embodiments, the user data used for analysis, i.e. the aforementioned data to be analyzed, may include transcribed text data and speech feature data.

[0187] In the above embodiments of this disclosure, by using two different technical means, step 820 (speech recognition processing) and step 830 (speech feature extraction), language content information and speech feature data can be extracted from the speech input data, respectively. These data can provide input data for subsequent content dimension analysis and speech dimension analysis, respectively.

[0188] 140: In response to the acquisition of user input data, the third intelligent agent performs multi-dimensional analysis and processing of the user input data based on the target virtual character and the target standard answer data corresponding to the target question data, and generates evaluation data.

[0189] In some embodiments, such as after obtaining user input data in step 130, the third agent can perform multi-dimensional analysis and processing on the user input data based on the target virtual character and the target standard answer data corresponding to the target question data. In this embodiment, the third agent can obtain target character feature data associated with the target virtual character according to the character identifier of the target virtual character, and determine the analysis dimensions, evaluation benchmarks, and / or weights of each analysis dimension in the multi-dimensional analysis and processing based on the target character feature data. As an explanation, but not a limitation, the third agent can obtain the target standard answer data corresponding to the target question data as a scoring benchmark, and refer to the character feature information corresponding to the target virtual character to comprehensively evaluate the user's performance from multiple dimensions, thereby generating evaluation data. In some embodiments, the character feature information corresponding to the target virtual character can be used to determine the focus, evaluation benchmark, or weight allocation of each analysis dimension during the evaluation process, so that the evaluation process can present an evaluation focus that is appropriate for different virtual characters.

[0190] In some embodiments, multi-dimensional analysis processing may include content-dimensional analysis and voice-dimensional analysis. In some embodiments, reference... Figure 9 Step 140 may include steps 910 and 920 below.

[0191] 910: Perform content dimension analysis on the transcribed text data to obtain the content analysis results.

[0192] In some embodiments, a third agent can perform content-level analysis on the transcribed text data obtained from the speech input data. In some embodiments, the third agent can compare and analyze the transcribed text data with target standard answer data in a structured question bank to obtain content analysis results. As an explanation, and not a limitation, content-level analysis can evaluate the performance level of the user's text content in terms of knowledge accuracy, logical expression, and sales skills.

[0193] In some embodiments, content dimension analysis may include at least one of product knowledge accuracy analysis, sales skills analysis, logical structure analysis, and language expression analysis. By way of explanation and not limitation, product knowledge accuracy analysis can verify the product information involved in the user's answer to determine the user's mastery of key product knowledge points; sales skills analysis can assess the user's ability to use persuasive techniques and strategies when dealing with customer questions; logical structure analysis can evaluate the logicality, hierarchy, and argumentation process of the user's answer; and language expression analysis can assess the user's word choice standardization, fluency, and professionalism.

[0194] 920: Perform speech dimension analysis on the speech feature data to obtain the speech analysis results.

[0195] In some embodiments, a third-party intelligent agent can perform voice dimension analysis on the voice feature data extracted from the voice input data. By way of explanation and not limitation, voice dimension analysis can evaluate a user's communication performance at the level of voice expression, focusing on the user's acoustic performance characteristics during voice interaction.

[0196] In some embodiments, voice dimension analysis may include at least one of timbre analysis, intonation analysis, speech rate analysis, and emotion analysis. For example, timbre analysis can assess the clarity, resonance, and stability of a user's voice; intonation analysis can analyze the rise and fall, stress distribution, and intonation variations of a user's voice; speech rate analysis can determine whether the user's speaking rhythm and speed are appropriate; and emotion analysis can identify a user's emotional state through voice features, such as whether they exhibit confident, nervous, or hesitant emotional characteristics.

[0197] In some embodiments, the evaluation data may include content dimension scores corresponding to the content analysis results and speech dimension scores corresponding to the speech analysis results. In some embodiments, the third agent may quantify the various analysis results of the content dimension analysis into corresponding content dimension scores and the various analysis results of the speech dimension analysis into corresponding speech dimension scores, thereby generating structured evaluation data.

[0198] In some embodiments, multi-dimensional analysis processing may include quantifying and scoring each analysis dimension to generate multi-dimensional scoring data. In some embodiments, for each sub-dimension in content dimension analysis and each sub-dimension in voice dimension analysis, a third-party intelligent agent may quantify and score the user's performance based on preset scoring criteria, generating quantified scoring data covering multiple dimensions. For explanation, this multi-dimensional scoring data can serve as the basis for generating subsequent feedback information.

[0199] In some embodiments, the method steps according to the above embodiments of this disclosure can be executed cyclically. Accordingly, in some embodiments, after generating evaluation data in step 140, the method provided by the embodiments of this disclosure may further include the following steps A1, A2, and A3:

[0200] A1: In response to the generation of evaluation data, determine the next target question data corresponding to the target virtual character from the structured question bank data.

[0201] In some embodiments, after completing the evaluation process for the current target question data, question data that corresponds to the target virtual character but has not yet been output can be found from the structured question bank data and used as the next target question data.

[0202] A2: Take the next target question data as the new target question data, and output the new target question data through the second agent.

[0203] In some embodiments, for example, the next target question data determined in step A1 can be used as the new target question data, and the second agent can output the new target question data and obtain the user's input data for the new target question data.

[0204] A3: In response to the acquisition of user input data, the third intelligent agent performs multi-dimensional analysis and processing of the user input data based on the target virtual role and the target standard answer data corresponding to the new target question data, and generates evaluation data.

[0205] In some embodiments, the third agent may perform multi-dimensional analysis processing on the new user input data based on the target virtual character and the target standard answer data corresponding to the new target question data, thereby generating evaluation data for the current question.

[0206] In some embodiments, the target virtual character may remain unchanged during the loop, with only the target question data being replaced by the next target question data. In some embodiments, the loop process can continue until all question data corresponding to the target virtual character has been processed, or until a termination instruction is received from the user. As an explanation, and not a limitation, the solution according to the above embodiments enables a multi-round question-answering-evaluation practice process, allowing users to complete multiple rounds of practice questions against the same target virtual character.

[0207] 150: Generate feedback information based on the evaluation data.

[0208] In some embodiments, after obtaining evaluation data, feedback information for the user can be further generated based on the evaluation data. In some embodiments, the step of generating feedback information based on the evaluation data can be performed by a third intelligent agent. In some embodiments, the step of generating feedback information based on the evaluation data can be performed by an artificial intelligence big data model. In some embodiments, the third intelligent agent includes an intelligent agent built based on an artificial intelligence big data model. It is understood that other intelligent agents in the embodiments of this disclosure may also be built by or have their functions implemented by an artificial intelligence big data model, and this disclosure does not limit this.

[0209] In some embodiments, step 150 described above may be performed after each step of generating evaluation data, and feedback information for the current target question data may be generated based on the generated evaluation data. By way of explanation and not limitation, this means that the user receives corresponding feedback information each time they complete an answer to a target question, and the feedback information is immediate and question-specific.

[0210] In optional embodiments, feedback information can be generated by comprehensively analyzing multiple evaluation data accumulated from multiple executions after a predetermined number of evaluation data runs or after a predetermined number of evaluation data runs have been generated. By way of explanation and not limitation, this feedback information may include statistical feedback data derived from the aggregation of multiple evaluation data, such as overall performance score, average scores for each dimension, and comprehensive improvement suggestions across multiple questions, thereby providing more summative feedback information. However, it is understood that in other embodiments, the above-described embodiments for generating feedback information can be used in combination, i.e., generating single-question feedback information after each evaluation, and then comprehensively generating pass-level feedback information covering multiple questions after multiple iterations, which falls within the protection scope of this disclosure.

[0211] In some embodiments, reference Figure 10 Step 150 may include the following steps 1010, 1020 and 1030.

[0212] 1010: Identify the first dimension in the evaluation data where the user's performance score is higher than a preset first threshold, and generate the first feedback data based on the first dimension.

[0213] In some embodiments, the scores of user performance in each dimension of the evaluation data can be analyzed, and the dimensions with scores higher than a preset first threshold can be identified as the first dimension.

[0214] In some embodiments, the first dimension represents aspects that can be used to characterize a user's superior performance. In this embodiment, the first feedback data includes strengths data. In some embodiments, the first feedback data includes highlight information. In some embodiments, the preset first threshold is a preset strength threshold. For explanation, this threshold can be used to define the criteria for determining whether a user's performance in a certain dimension reaches a level of superiority.

[0215] 1020: Identify the second dimension in the evaluation data where the user's performance score is lower than the preset second threshold, and generate second feedback data based on the second dimension.

[0216] In some embodiments, the scores of user performance in each dimension of the evaluation data can be analyzed, and dimensions with scores below a preset second threshold can be identified as second dimensions.

[0217] In some embodiments, the second dimension can be used to characterize aspects of user performance that need improvement. In some embodiments, the second feedback data includes improvement point data. For explanation, the second feedback data can be used to characterize aspects of the user's inadequate performance during the conversation, and may include a specific description of the user's problems and directions for improvement in the corresponding dimension. In some embodiments, the second feedback data includes improvement point information. For explanation, the second feedback data can be presented in the form of improvement points, indicating the user's shortcomings and room for improvement in certain dimensions. In some embodiments, the preset second threshold is a preset improvement threshold.

[0218] In some embodiments, the first feedback data may include high-performing content snippets and corresponding positive evaluation descriptions; the second feedback data may include content snippets to be improved, improvement suggestions, and reference scripts. In some embodiments, the first feedback data may also locate specific high-performing content snippets in the user's response and provide corresponding positive evaluation descriptions; the second feedback data may also locate specific content snippets to be improved and provide targeted improvement suggestions and reference script examples.

[0219] 1030: Generate personalized suggestions for users based on the first feedback data and / or the second feedback data.

[0220] In some embodiments, the content of the first feedback data and / or the second feedback data may be combined to generate personalized recommendations for the user.

[0221] In some embodiments, step 1030 may include: matching a corresponding improvement suggestion template from a preset suggestion library based on the second dimension corresponding to the second feedback data, and generating personalized suggestions in conjunction with the user's specific performance. In some embodiments, a preset suggestion library may be maintained, which stores improvement suggestion templates for different dimensions and different types of problems. In some embodiments, after identifying the user's second dimension, an improvement suggestion template corresponding to the second dimension may be matched from the preset suggestion library, and the template may be personalized and adjusted according to the user's specific performance during the dialogue, thereby generating targeted personalized suggestions.

[0222] In some embodiments, feedback information may include multi-dimensional rating data from the evaluation data, first feedback data, second feedback data, and personalized suggestions. For explanation, the method of this disclosure embodiment can provide users with a comprehensive feedback report, which includes quantitative rating data for each dimension to visually demonstrate the user's overall performance level, first feedback data to affirm the user's strengths, second feedback data to point out areas for improvement, and personalized suggestions to provide specific directions for improvement.

[0223] In an optional embodiment, the human-computer interaction processing method may further include: switching between a first dialogue mode and a second dialogue mode in response to a user's mode selection instruction.

[0224] In some embodiments, the first dialogue mode is a practice mode, and the second dialogue mode is a practical mode. By way of explanation and not limitation, the practice mode can be designed for users' daily learning and skill practice scenarios, allowing users to practice repeatedly in a relatively relaxed environment; the practical mode can be designed for users' assessment and competency verification scenarios, simulating a real customer interaction environment to comprehensively evaluate users.

[0225] In some embodiments, the first dialogue mode can support viewing standard answer data, re-answering, and skipping the current question. Specifically, in this embodiment, users can view standard answer data in the structured question bank after answering to compare the differences between their own answers and the standard answers; users can also choose to re-answer the same question to reinforce their understanding of the knowledge points; in addition, users can skip the current question and go directly to the next question to flexibly arrange their practice pace.

[0226] In some embodiments, viewing standard answer data can be disabled in the second dialogue mode, and the answer results data are included in the evaluation data. Specifically, in this embodiment, the system may not display standard answer data to the user to ensure the authenticity and validity of the evaluation results; at the same time, the user's answer results data in the practical mode can be recorded and included in the evaluation data as a basis for assessing the user's ability level.

[0227] In some embodiments, the first dialogue mode may employ a random question generation strategy, while the second dialogue mode may employ a full question generation strategy.

[0228] In some embodiments, in practice mode, questions may be randomly selected from a structured question bank and presented to the user. This random selection allows the user to encounter different combinations of questions each time they practice, increasing the diversity of the practice. In some embodiments, in real-world practice mode, all questions from the structured question bank may be presented to the user sequentially.

[0229] In an optional embodiment, the human-computer interaction processing method may further include data analysis and query functions. In this embodiment, historical evaluation data may be recorded, such as historical evaluation data generated during multiple executions of the aforementioned step 140, thereby recording the user's performance history during multiple rounds of practice.

[0230] In some embodiments, the human-computer interaction processing method may further include the step of accessing practice record data, wherein the practice record data may include at least one of a first indicator data, a second indicator data, and a third indicator data. In some embodiments, the first indicator data is the pass score data, the second indicator data is the answering time data, and the third data is the pass rate data. By accessing the aforementioned practice record data, administrators can gain a comprehensive understanding of the user's performance and overall completion status during the human-computer interaction practice process.

[0231] In some embodiments, the human-computer interaction processing method may further include the following steps: receiving an unstructured query request; generating structured query logic based on the query request; and executing the structured query logic to query historical evaluation data to obtain query result data. In some embodiments, the unstructured query request can be received through a business management terminal, wherein the unstructured query request may include a query request in natural language form. In a specific example, the user inputs a query request of "the three questions with the lowest pass rate this week". As an explanation, the user does not need to master professional database query syntax, but can achieve flexible querying of data by inputting a natural language description. In some embodiments, the unstructured query request can be processed by a fifth intelligent agent. In this embodiment, the fifth intelligent agent includes a data analysis intelligent agent, which is configured to understand the user's query intent and automatically generate corresponding query logic. As an explanation and not a limitation, the business management terminal may be an interface for business management personnel, through which business management personnel can input query requests to query and analyze historical evaluation data.

[0232] In some embodiments, the human-computer interaction processing method may further include the step of outputting query result data in a visualized form. As an explanation, visualization allows users to more intuitively understand data distribution and trend changes. In an optional embodiment, the human-computer interaction processing method may further include the step of exporting the query result data into a preset format file in response to an export command. As an explanation, the export function facilitates users using the query results for subsequent report generation or further analysis.

[0233] In some embodiments, the objects of data analysis and querying may include evaluation data and feedback data. For example, by analyzing and querying multi-dimensional scoring data and feedback information obtained by users during practice sessions, managers can understand the training effectiveness from multiple perspectives.

[0234] In some embodiments, statistical analysis of training record data may include aggregate statistical processing of first indicator data and trend analysis processing of third indicator data.

[0235] In some embodiments, aggregated statistical processing can be used to obtain summary indicator data, including but not limited to average scores, highest scores, and lowest scores. For explanation, trend analysis can be used to obtain trend data on pass rates over time, thereby helping managers identify trends of improvement or decline in training effectiveness and adjust training strategies in a timely manner.

[0236] In an optional embodiment, the human-computer interaction processing method can be implemented based on a pipelined architecture. In this embodiment, the pipelined architecture may include multiple processing units connected in sequence. As an explanation, and not a limitation, a pipelined architecture is a technical architecture that decomposes complex business processes into multiple independent processing stages, where the output of one processing unit can serve as the input of the next, thereby achieving modularization and standardization of business logic.

[0237] In some embodiments, the multiple processing units may include a first processing unit, a second processing unit, a third processing unit, a fourth processing unit, a fifth processing unit, a sixth processing unit, a seventh processing unit, and an eighth processing unit. In some embodiments, the first processing unit is a material access unit, used to receive and identify multimodal raw data; the second processing unit is a knowledge parsing unit, used to parse and process the received multimodal raw data to extract knowledge content; the third processing unit is a question bank generation unit, used to generate structured question bank data through a first intelligent agent based on the extracted knowledge content; the fourth processing unit is a conversation initialization unit, used to determine the dialogue mode and virtual role configuration; the fifth processing unit is a question presentation unit, used to select and present questions from the structured question bank data according to the question-generating strategy; the sixth processing unit is a response collection unit, used to acquire user input data during the dialogue process; the seventh processing unit is a scoring and evaluation unit, used to perform multi-dimensional analysis and processing of user input data through a third intelligent agent and generate evaluation data and feedback information; and the eighth processing unit is a data analysis unit, used to perform statistical analysis on various types of data generated during the practice process.

[0238] In some embodiments, the pipeline architecture can define business processes through configuration data. In this embodiment, the configuration data can specify the execution order and parameter configuration of each processing unit. For explanation, a configuration-driven approach allows for flexible adjustment of the execution order and specific parameters of each stage of the business process without modifying the logic of the processing units themselves, thereby improving the system's business adaptability.

[0239] In some embodiments, the pipeline architecture can support dynamic loading and unloading of processing units. In some embodiments, the human-computer interaction processing method may further include a step of version control of the processing units. In some embodiments, in the event of an anomaly in a new version of the processing unit, a version rollback operation is supported, and a rapid recovery to a stable version is possible.

[0240] In an optional embodiment, the human-computer interaction processing method can be executed collaboratively by a server and a client. In some embodiments, the server can be used for configuration and management, and the client can be used for interactive practice.

[0241] In some embodiments, the server is a management interface and the client is a user interface. By way of explanation and not limitation, the management interface is an interface for administrators, used to perform administrative operations such as system configuration, rule setting, and data management; the user interface is an interactive interface for practice users, used to perform practice operations such as dialogue interaction, voice input, and result viewing. In some embodiments, the server is a web application and the client is a mobile application.

[0242] In some embodiments, the server is configured to perform at least one of the following operations: question bank configuration processing, rule setting processing, access control processing, and data statistics processing.

[0243] In some embodiments, question bank configuration processing may include setting one or more of the following: the number of questions, difficulty level, and the versions of the question bank that are added and removed from the platform. In some embodiments, rule setting processing may include configuring one or more of the following: compliance rules, question generation strategies, and scoring criteria. In some embodiments, access control processing may include managing user access permissions and setting a pass list. In some embodiments, data statistics processing may include viewing practice record data and generating statistical reports.

[0244] In some embodiments, the client is configured to perform at least one of the following operations: dialogue mode selection processing, voice interaction processing, rating display processing, and feedback information presentation processing. In this embodiment, voice interaction processing may include acquiring and processing the user's voice input data; rating display processing may include displaying multi-dimensional rating data to the user; and feedback information presentation processing may include displaying first feedback data, second feedback data, and personalized suggestions to the user.

[0245] In some embodiments, the human-computer interaction processing method may further include feeding back practice data generated by the client to the server, performing statistical analysis on the practice data to generate analysis report data, and optimizing the structured question bank data based on the analysis report data. As an explanation, and not a limitation, the above-described data closed-loop mechanism enables the practice data generated by the client (user end) to continuously feed back to the server (management end). The management end can then perform in-depth statistical analysis on the fed-back practice data to generate analysis report data containing information such as user performance trends, question difficulty distribution, and knowledge weaknesses. Based on this analysis report data, the management end can then perform targeted optimization of the structured question bank data, such as adding questions in areas where users are generally weak or adjusting the question difficulty distribution, thereby forming a continuous improvement closed loop of "practice-analysis-optimization".

[0246] In some embodiments, the configuration data published by the server may include question bank version data, practice rules data, and permission configuration data. The client can execute the practice process based on the configuration data, thereby enabling the management end to uniformly control the practice behavior of the user end.

[0247] Based on the technical solutions of the above embodiments, the human-computer interaction processing method provided by the present disclosure can achieve at least the following technical effects.

[0248] The human-computer interaction processing method of this disclosure generates a structured question bank through a first intelligent agent, comprising question data corresponding to multiple virtual characters and standard answer data corresponding to the question data. A second intelligent agent outputs target question data corresponding to a target virtual character. A third intelligent agent then performs multi-dimensional analysis on the acquired user input data based on the target virtual character and the target standard answer data corresponding to the target question data. This solution solves the technical problems of known solutions, such as question bank content being unrelated to the target object type, lack of precise reference in evaluation, and the inability of the evaluation process to adapt to different character characteristics. This solution establishes a correspondence between question data and virtual characters during the question bank generation stage and simultaneously introduces the standard answer data corresponding to the target virtual character and the target question data as evaluation criteria during the evaluation stage. This allows the evaluation process to be dynamically adjusted based on the current target virtual character, in addition to scoring based on precise question-level standard answers, thereby improving the training targeting and evaluation accuracy in the human-computer interaction process.

[0249] In a further embodiment of this disclosure, a role-based question bank is constructed by configuring virtual characters as virtual practice characters and by determining the target customer group corresponding to the virtual practice characters, generating question data corresponding to the virtual practice characters and standard answer data corresponding to the question data based on the target customer characteristic data corresponding to the target customer group. This further solution solves the technical problems that general question banks cannot reflect the differences between different customer types and that training scenarios lack specificity. By basing the generation of question data and standard answer data on the characteristic data of the target customer group, this solution enables the questions and standard answers in the question bank for different virtual practice characters to present differentiated content that matches the characteristics of the corresponding customer group, thereby supporting users to conduct targeted training for different types of practice characters.

[0250] In a further embodiment of this disclosure, question style parameters are determined based on target customer characteristic data, and the style characteristics of the question data are controlled through these parameters. This solution further addresses the technical problem of indistinguishable question styles among different types of virtual training roles. As an explanation, and not a limitation, different customer groups have different communication tendencies and expression habits. The question style parameters quantify these group differences into controllable parameters, thereby constraining the style performance during question generation. Thus, questions generated under different role configurations within the same system exhibit identifiable differences in style, enabling the simulation of diverse real-world interaction scenarios, avoiding homogenization of virtual character behavior, and enhancing the comprehensiveness and practicality of training.

[0251] In this embodiment of the disclosure, a third intelligent agent can also perform multi-dimensional analysis and processing on user input data to generate evaluation data during the analysis and feedback stage. Based on the scores of each dimension in the evaluation data, the user's strengths and areas for improvement can be identified, and corresponding feedback data and personalized suggestions can be generated respectively. This method can provide users with targeted improvement guidance, enabling users to clearly understand their own strengths and weaknesses and make targeted improvements.

[0252] In a further embodiment of this disclosure, speech recognition and speech feature extraction processes can be performed simultaneously on the user's voice input data to obtain transcribed text data and speech feature data, respectively. Content-dimensional analysis and speech-dimensional analysis are then performed separately, enabling a comprehensive evaluation of user performance from both content and expression perspectives. This approach provides a more comprehensive and objective evaluation result than single text analysis. Specifically, the solution in this embodiment can generate first and second feedback data based on the comparison results of the scores of each dimension in the evaluation data with preset thresholds, and generate personalized suggestions accordingly. This allows the feedback information to have both positive incentive and improvement guidance functions, helping users maintain their strengths while specifically addressing their weaknesses.

[0253] In some embodiments of this disclosure, reference is made to Figure 11 Furthermore, a human-computer interaction processing system 1100 is provided, which can be used to implement the human-computer interaction processing methods described in the foregoing embodiments. In some embodiments, the human-computer interaction processing system 1100 may include a generation module 1120, an output module 1130, and an evaluation module 1140.

[0254] In some embodiments, the generation module 1120 is configured to generate structured question bank data through a first intelligent agent based on knowledge content and multiple virtual characters. The structured question bank data includes question data corresponding to each of the multiple virtual characters and standard answer data corresponding to the question data. In this embodiment, the generation module 1120 may include a determining submodule and a first generation submodule. In some embodiments, the determining submodule is configured to determine a target customer group corresponding to the virtual practice character; the first generation submodule is configured to generate question data corresponding to the virtual practice character and standard answer data corresponding to the question data based on target customer characteristic data corresponding to the target customer group. In this embodiment, the specific implementation of the generation module 1120 can be found in the detailed description of step 120 in the foregoing method embodiments, and will not be repeated here.

[0255] In some embodiments, the output module 1130 is configured to output target question data corresponding to the target virtual character through a second intelligent agent based on the structured question bank data, wherein the target virtual character is selected from the plurality of virtual characters. In some embodiments, the virtual character is a virtual practice character.

[0256] In some embodiments, the dialogue processing module 1130 may further include a parameter determination submodule and a style control submodule. In some embodiments, the parameter determination submodule is configured to determine topic style parameters based on target customer feature data; the style control submodule is configured to control the style features of the dialogue based on the topic style parameters, wherein the style features include at least one of content features, expression features, voice features, and tone features.

[0257] In some embodiments, the dialogue processing module 1130 may further include a second generation submodule and a display submodule. In some embodiments, the second generation submodule is configured to generate character attribute display information of the virtual practice role based on the target customer feature data; the display submodule is configured to display the character attribute display information to the user during the output of target question data corresponding to the target virtual role and / or the process. In this embodiment, the specific implementation of the dialogue processing module 1130 can be referred to the detailed description of step 130 in the foregoing method embodiments, and will not be repeated here.

[0258] In some embodiments, the evaluation module 1140 is configured to, in response to acquiring user input data, perform multi-dimensional analysis on the user input data using a third intelligent agent based on the target virtual character and the target standard answer data corresponding to the target question data, thereby generating evaluation data. In some embodiments, the evaluation module 1140 may perform content dimension analysis on the transcribed text data to obtain content analysis results, and perform speech dimension analysis on the speech feature data to obtain speech analysis results. In this embodiment, the specific implementation of the evaluation module 1140 can be referred to the detailed description of step 140 in the foregoing method embodiments, and will not be repeated here.

[0259] In some embodiments, refer to Figure 11 The human-computer interaction processing system 1100 may also include a parsing module 1110 and a feedback module 1150.

[0260] In some embodiments, the parsing module 1110 is configured to parse and process multimodal raw data to extract knowledge content. The multimodal raw data includes at least one of document data, image data, audio data, and video data. In this embodiment, the specific implementation of the parsing module 1110 can be found in the detailed description of step 110 in the foregoing method embodiments, and will not be repeated here.

[0261] In some embodiments, the feedback module 1150 is configured to generate feedback information based on evaluation data. In some embodiments, the feedback module 1150 may include a first identification submodule, a second identification submodule, and a suggestion submodule. In some embodiments, the first identification submodule is configured to identify a first dimension in the evaluation data where the user's performance score is higher than a preset first threshold, and generate first feedback data based on the first dimension; the second identification submodule is configured to identify a second dimension in the evaluation data where the user's performance score is lower than a preset second threshold, and generate second feedback data based on the second dimension; the suggestion submodule is configured to generate personalized suggestions for the user based on the first feedback data and / or the second feedback data. In this embodiment, the specific implementation of the feedback module 1150 can be found in the detailed description of step 150 in the foregoing method embodiments, and will not be repeated here.

[0262] In some embodiments, the human-computer interaction processing system may adopt a layered architecture consisting of a front-end layer, an access layer, and a back-end layer. In some embodiments, the front-end layer may be responsible for the presentation of the user interface and the collection of interactive operations, the access layer may be responsible for request routing and protocol conversion, and the back-end layer may be responsible for the processing of core business logic and the storage and management of data. As an explanation, and not a limitation, the responsibilities of each layer in the above-described layered architecture are clearly defined and decoupled, which not only facilitates system expansion and maintenance but also enables the implementation of the human-computer interaction processing methods described above. For example, the practice data generated by the client can be fed back to the server, thereby achieving the aforementioned continuous improvement closed loop of "practice-analysis-optimization".

[0263] In some embodiments, reference Figure 12 The system also provides a human-computer interaction processing system 1200, which may include a management subsystem 1210, a client subsystem 1220, and a unified access layer 1230.

[0264] In some embodiments, the management subsystem 1210 is configured to perform system-level configuration and management operations, including question bank configuration processing, rule setting processing, access control processing, and data statistics processing. In this embodiment, the management subsystem can set parameters for the number of questions and difficulty levels to control the generation behavior of the generation module 120, configure preset compliance rules to standardize the question bank content, set question generation strategies and scoring criteria for dialogue modes, and manage user access permissions.

[0265] In some embodiments, the client subsystem 1220 is configured to perform user-oriented interactive training operations, including dialogue mode selection processing, voice interaction processing, score display processing, and feedback information presentation processing. In this embodiment, the client subsystem 1220 can receive and process the user's voice input data, display the character attribute information of the virtual training role to the user, and present multi-dimensional score data and feedback information. In some embodiments, as described above, the human-computer interaction processing system 1200 can also implement a data closed-loop mechanism. Specifically, the training data generated by the client subsystem 1220 can be fed back to the management subsystem 1210. The management subsystem 1210 can perform statistical analysis on the fed-back training data to generate analysis report data, and optimize and adjust the structured question bank data based on the analysis report data, thereby forming a data closed loop for continuous improvement.

[0266] In some embodiments, the human-computer interaction processing system may support a multi-tenant deployment mode. In this embodiment, the human-computer interaction processing system can provide services to multiple tenants simultaneously and achieve data isolation between different tenants. By way of explanation and not limitation, tenants may correspond to different enterprises, organizations, or business entities, and different tenants access the human-computer interaction processing system through their respective tenant identifiers.

[0267] In some embodiments, the management subsystem 1210 and the client subsystem 1220 may be associated with a tenant identifier. In some embodiments, the structured question bank data, practice data, evaluation data, and feedback data corresponding to different tenants may be physically isolated from each other.

[0268] In some embodiments, the management subsystem 1210 can configure question bank generation parameters, compliance rules, dialogue modes, and scoring standards for different tenants. In some embodiments, the client subsystem 1220 can carry the corresponding tenant identifier when performing practice interactions and uploading data, and the unified access layer 1230 can correctly return the practice data generated by the client to the management data space of the corresponding tenant based on the tenant identifier.

[0269] In this embodiment of the disclosure, the multi-tenant isolation mechanism described above can ensure system resource sharing while achieving data security isolation and independent business operation between different tenants, avoiding data commingling between different enterprises or organizations, thereby improving the security and scalability of the system in enterprise-level application scenarios.

[0270] The human-computer interaction processing system and its components, modules, units, and features described in the embodiments of this disclosure can be incorporated into the human-computer interaction processing method of the embodiments of this disclosure in a non-contradictory manner, and will not be elaborated here. Conversely, the human-computer interaction processing method and its steps, sub-steps, and features described in the embodiments of this disclosure can also be incorporated into the human-computer interaction processing system of the embodiments of this disclosure in a non-contradictory manner.

[0271] In this disclosure, an electronic device is also provided, including: a processor and a memory storing a computer program, the processor being configured to perform the method of any of the embodiments of this disclosure when running the computer program.

[0272] Figure 13 The illustration shows a method or electronic device 1300 that can implement embodiments of the present disclosure. In some embodiments, it may include more or fewer electronic devices than illustrated. In some embodiments, it may be implemented using a single or multiple electronic devices. In some embodiments, it may be implemented using cloud-based or distributed electronic devices.

[0273] like Figure 13 As shown, the electronic device 1300 includes a processor 1301, which can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) 1302 or programs and / or data loaded from storage portion 1308 into random access memory (RAM) 1303. The processor 1301 may be a multi-core processor or may contain multiple processors. In some embodiments, the processor 1301 may include a general-purpose main processor and one or more special coprocessors, such as a graphics processing unit (GPU), a neural network processor (NPU), a digital signal processor (DSP), etc. Various programs and data required for the operation of the electronic device 1300 are also stored in RAM 1303. The processor 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. An input / output (I / O) interface 1305 is also connected to bus 1304.

[0274] The processor and memory described above are used together to execute a program stored in the memory. When the program is executed by a computer, it can implement the steps or functions of the methods described in the above embodiments.

[0275] The following components are connected to I / O interface 1305: an input section 1306 including a keyboard, mouse, touchscreen, etc.; an output section 1307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1308 including a hard disk, etc.; and a communication section 1309 including a network interface card such as a LAN card, modem, etc. The communication section 1309 performs communication processing via a network such as the Internet. Drive 1310 is also connected to I / O interface 1305 as needed. Removable media 1311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 1310 as needed so that computer programs read from them can be installed into storage section 1308 as needed. Figure 13The diagram only shows a portion of the components and does not imply that the computer system 1300 only includes... Figure 13 The components shown.

[0276] The systems, devices, modules, or units described in the above embodiments can be implemented by a computer or its associated components. The computer may be, for example, a mobile terminal, smartphone, personal computer, laptop computer, in-vehicle human-machine interface device, personal digital assistant, media player, navigation device, game console, tablet computer, wearable device, smart TV, Internet of Things system, smart home, industrial computer, server, or a combination thereof.

[0277] Although not shown, in embodiments of this disclosure, a program product is provided, the program product comprising a computer program configured to be run to implement the methods of any embodiment of this disclosure.

[0278] Although not shown, in embodiments of this disclosure, a storage medium is provided storing a computer program configured to be executed to implement the methods of any of the embodiments of this disclosure.

[0279] Storage media in embodiments of this disclosure include articles that are permanent and non-permanent, removable and non-removable, capable of storing information by any method or technology. Examples of storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0280] The methods, programs, systems, apparatuses, etc., of the embodiments of this disclosure can be executed or implemented in a single or multiple networked computers, or practiced in a distributed computing environment. In the embodiments of this specification, in these distributed computing environments, tasks can be performed by remote processing devices connected via a communication network.

[0281] Those skilled in the art will understand that the embodiments described in this specification can be provided as methods, systems, or computer program products. Therefore, those skilled in the art will realize that the functional modules / units or controllers and related method steps described in the above embodiments can be implemented in software, hardware, or a combination of both.

[0282] Unless explicitly stated otherwise, the actions or steps of the methods or procedures described in the embodiments of this disclosure do not necessarily have to be performed in a specific order and may still achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0283] This document describes several embodiments of the present disclosure; however, for the sake of brevity, the descriptions of the embodiments are not exhaustive, and identical or similar features or portions between the embodiments may be omitted. In this document, "one embodiment," "some embodiments," "example," "specific example," or "some examples" refers to at least one embodiment or example applicable to the present disclosure, but not all embodiments. The above terms do not necessarily refer to the same embodiment or example. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of the different embodiments or examples.

[0284] The exemplary systems and methods of this disclosure have been specifically shown and described with reference to the foregoing embodiments, and are merely examples of the best mode for implementing the systems and methods. Those skilled in the art will understand that various changes can be made to the embodiments of the systems and methods described herein without departing from the spirit and scope of this disclosure as defined in the appended claims when implementing the systems and / or methods.

Claims

1. A human-computer interaction processing method, characterized in that, include: Based on the knowledge content and multiple virtual characters, a structured question bank data is generated by a first intelligent agent. The structured question bank data includes question data corresponding to each of the multiple virtual characters and standard answer data corresponding to the question data. Based on the structured question bank data, the second intelligent agent outputs the target question data corresponding to the target virtual character, wherein the target virtual character is selected from the plurality of virtual characters; In response to acquiring user input data, a third intelligent agent performs multi-dimensional analysis and processing on the user input data based on the target virtual character and the target standard answer data corresponding to the target question data, and generates evaluation data.

2. The human-computer interaction processing method according to claim 1, characterized in that, Also includes: The multimodal raw data is parsed and processed to extract the knowledge content, wherein the multimodal raw data includes any one or more combinations of document data, image data, audio data, and video data; and the knowledge content includes any one or more combinations of text semantic features, image content features, and audio text features.

3. The human-computer interaction processing method according to claim 2, characterized in that, The parsing and processing of the multimodal raw data to extract the knowledge content includes any one or a combination of the following processing methods: The document data is subjected to text semantic analysis to obtain text semantic features; Image recognition processing is performed on the image data and / or video data to obtain image content features; The audio data and / or video data are processed by speech recognition to obtain audio-text features.

4. The human-computer interaction processing method according to claim 1, characterized in that, The process of generating structured question bank data through a first intelligent agent based on knowledge content and multiple virtual roles includes: Obtain preset generation parameters, wherein the preset generation parameters include a number of questions parameter and a difficulty level parameter; The first intelligent agent generates structured question bank data that conforms to the preset generation parameters.

5. The human-computer interaction processing method according to claim 1, characterized in that, The process of generating structured question bank data through a first intelligent agent based on knowledge content and multiple virtual roles includes: The first intelligent agent generates candidate question bank data; The candidate question bank data is subjected to compliance verification processing based on preset compliance rules to obtain compliant structured question bank data.

6. The human-computer interaction processing method according to claim 1, characterized in that, The virtual character is a virtual practice character; The process of generating structured question bank data through a first intelligent agent based on knowledge content and multiple virtual roles includes: Determine the target customer group corresponding to the virtual training characters; Based on the target customer characteristic data corresponding to the target customer group, generate question data corresponding to the virtual training character and standard answer data corresponding to the question data.

7. The human-computer interaction processing method according to claim 6, characterized in that, The step of generating question data corresponding to the virtual practice character and standard answer data corresponding to the question data based on the target customer characteristic data corresponding to the target customer group includes: Determine the question style parameters based on the target customer characteristic data; The style characteristics of the question data are controlled according to the question style parameters.

8. The human-computer interaction processing method according to claim 1, characterized in that, The method further includes: Obtain the user's voice input data; The voice input data is processed by speech recognition to obtain transcribed text data; The voice input data is subjected to voice feature extraction processing to obtain voice feature data; The user input data includes the transcribed text data and the speech feature data; The step of generating evaluation data by having a third intelligent agent perform multi-dimensional analysis and processing of the user input data based on the target virtual character and the target standard answer data corresponding to the target question data, includes: The transcribed text data is subjected to content dimension analysis to obtain content analysis results; The speech feature data is subjected to speech dimension analysis to obtain speech analysis results.

9. The human-computer interaction processing method according to claim 1, characterized in that, The method further includes: Identify a first dimension in the evaluation data where the user's performance score is higher than a preset first threshold, and generate first feedback data based on the first dimension; Identify a second dimension in the evaluation data where the user's performance score is lower than a preset second threshold, and generate second feedback data based on the second dimension; Personalized suggestions are generated for the user based on the first feedback data and / or the second feedback data.

10. A human-computer interaction processing system, characterized in that, include: The generation module is configured to generate structured question bank data through a first intelligent agent based on knowledge content and multiple virtual roles. The structured question bank data includes question data corresponding to each of the multiple virtual roles and standard answer data corresponding to the question data. The output module is configured to output target question data corresponding to the target virtual character through a second intelligent agent based on the structured question bank data, wherein the target virtual character is selected from the plurality of virtual characters; The evaluation module is configured to, in response to receiving user input data, perform multi-dimensional analysis and processing of the user input data by a third intelligent agent based on the target virtual character and the target standard answer data corresponding to the target question data, and generate evaluation data.

11. An electronic device, characterized in that, include: A processor and a memory storing a computer program, the processor being configured to implement the method as described in any one of claims 1 to 9 when the computer program is executed.

12. A program product comprising a computer program, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 9.

Images