A digital human interaction management system based on big data analysis

Abstract

The present application belongs to the field of digital human interaction management, and relates to data analysis technology, and is used for solving the problem that the prior art cannot monitor and analyze the multi-round interaction state of digital human, in particular to a digital human interaction management system based on big data analysis, which comprises a voice interaction support module, a multi-round dialogue supervision module and an effectiveness optimization analysis module connected in sequence; the voice interaction support module is used for recognizing the voice information received by the digital human and converting it into text information, inputting the text information into a general big model and calling corresponding output instructions, and sending the output instructions to an execution module; the present application can reasonably extend and supplement the answers according to the user's follow-up questions, and effectively remember the interaction history and intention information of the user in the multi-round dialogue, establish a dialogue context association model, and ensure that the platform can accurately understand the evolution and progression of the user's intention in the multi-round dialogue process.

Description

Technical Field

[0001] This invention belongs to the field of digital human interaction management and involves data analysis technology, specifically a digital human interaction management system based on big data analysis. Background Technology

[0002] Digital humans are digital human figures created using digital technology that closely resemble human appearances. In a broad sense, digital humans refer to the penetration of digital technology into all levels and stages of human anatomy, physics, physiology, and intelligence. In a narrow sense, digital humans are a product of the integration of information science and life science, using information science methods to virtually simulate the human body's form and function at different levels.

[0003] The invention patent with announcement number CN118708702A discloses an artificial intelligence-based digital human interaction management system. This system converts the acquired dialogue information into digital information and transmits the digital information to the interaction information pre-storage module. It avoids bypassing privacy word blocking by replacing keywords, thereby obtaining the privacy data of different users. However, the system cannot monitor and analyze the multi-turn interaction state of the digital human, resulting in the inability to automatically identify and optimize repetitive dialogues in multi-turn interactions, which affects the user experience.

[0004] To address the aforementioned technical problems, this application proposes a solution. Summary of the Invention

[0005] The purpose of this invention is to provide a digital human interaction management system based on big data analysis, which solves the problem that existing technologies cannot monitor and analyze the multi-round interaction states of digital humans;

[0006] The technical problem to be solved by this invention is: how to provide a digital human interaction management system based on big data analysis that can monitor and analyze the multi-round interaction states of digital humans.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A digital human interaction management system based on big data analysis includes a voice interaction support module, a multi-turn dialogue monitoring module, and an effectiveness optimization analysis module connected in sequence.

[0009] The voice interaction support module is used to recognize the voice information received by the digital human and convert it into text information, input the text information into the general large model and retrieve the corresponding output instructions, and send the output instructions to the execution module;

[0010] The multi-turn dialogue monitoring module is used to perform effectiveness monitoring analysis on the multi-turn dialogues between the digital human and the user: It generates a monitoring period, marks the multi-turn dialogue process between the digital human and the user within the monitoring period as a monitoring process, extracts all voice information extracted by the user during the monitoring process, and extracts the basic parameters of the voice information, including information type, information domain, and keyword set; it marks voice information with identical information type, information domain, and keyword set during the monitoring process as associated information, and marks the monitoring process as a repeated process or related process based on the associated information; it marks the ratio of the number of repeated processes to the number of monitoring processes within the monitoring period as the repetition coefficient of the monitoring period, and uses the repetition coefficient to determine whether the effectiveness of the multi-turn dialogues of the digital human within the monitoring period meets the requirements;

[0011] The effectiveness optimization analysis module is used to perform effectiveness optimization analysis on multi-turn dialogues between digital humans and users.

[0012] Furthermore, the output instructions include execution signals and question-and-answer signals; the execution signals include limb execution electrical signals corresponding to the text information, and the question-and-answer signals include voice broadcast electrical signals corresponding to the text information; after receiving the execution signal, the execution module performs limb control according to the execution signal, and after receiving the question-and-answer signals, the execution module performs voice broadcast control according to the question-and-answer signals.

[0013] Furthermore, the specific process of marking the regulatory process as a repetitive or related process includes: determining whether the output instructions retrieved from the general large model through the association information are the same; if so, the regulatory process is marked as a repetitive process; if not, the regulatory process is marked as a related process.

[0014] Furthermore, the specific process for determining whether the effectiveness of the digital human's multi-turn dialogue within the regulatory period meets the requirements includes comparing the repetition coefficient with a preset repetition threshold: if the repetition coefficient is greater than or equal to the repetition threshold, it is determined that the effectiveness of the digital human's multi-turn dialogue within the regulatory period does not meet the requirements, an effectiveness optimization analysis signal is generated, and the effectiveness optimization analysis signal is sent to the effectiveness optimization analysis module; if the repetition coefficient is less than the repetition threshold, it is determined that the effectiveness of the digital human's multi-turn dialogue within the regulatory period meets the requirements, and a comprehensive analysis of the regulatory period is conducted.

[0015] Furthermore, the specific process for conducting a comprehensive analysis of the regulatory cycle includes: marking the number of dialogue rounds in relevant processes as a persistence value; marking relevant processes with persistence values ​​not less than a preset persistence threshold as fully optimized processes; marking the number of fully optimized processes as a fully optimized value; comparing the fully optimized value with a preset fully optimized threshold: if the fully optimized value is less than the fully optimized threshold, it is determined that the multi-turn dialogue of the digital human within the regulatory cycle does not have fully optimized characteristics; if the fully optimized value is greater than or equal to the fully optimized threshold, it is determined that the multi-turn dialogue of the digital human within the regulatory cycle has fully optimized characteristics; the information types, information domains, and all keywords in the keyword set of all fully optimized processes constitute a type set, a domain set, and a key set, respectively; the information type, information domain, and keywords with the most corresponding elements in the information type, information domain, and keyword set constitute a fully optimized dataset; generating a fully optimized signal; and sending the fully optimized signal and the fully optimized dataset to the mobile terminal of the management personnel.

[0016] Furthermore, the effectiveness optimization analysis module is used to perform effectiveness optimization analysis on the multi-turn dialogue between the digital human and the user: the repeated processes corresponding to the output instructions within the regulatory period are marked as the central tendency of the output instructions, the variance of all central tendency of the output instructions is calculated to obtain the central tendency coefficient, and the effectiveness optimization measures of the regulatory period are marked by the central tendency coefficient.

[0017] Furthermore, the specific process of marking the effectiveness optimization measures of the regulatory cycle includes: comparing the concentration coefficient with a preset concentration threshold; if the concentration coefficient is less than the concentration threshold, the effectiveness optimization measures of the regulatory cycle are marked as voice recognition optimization, a voice recognition optimization signal is generated and sent to the mobile terminal of the management personnel; if the concentration coefficient is greater than or equal to the concentration threshold, the effectiveness optimization measures of the regulatory cycle are marked as information integration optimization, the text information corresponding to the K1 repeated processes with the largest concentration values ​​is marked as integration optimization objects, an information integration optimization signal is generated and the information integration optimization signal and integration optimization objects are sent to the mobile terminal of the management personnel.

[0018] Furthermore, the voice interaction support module communicates with the general large model, which is generated by the large model integration module. The large model integration module is used to generate the general large model, FAQs based on knowledge base documents, design multi-turn dialogue management logic, and construct a dialogue instruction intent recognition model.

[0019] The present invention has the following beneficial effects:

[0020] 1. Integrate a general large model through the large model integration module, make reasonable logical extensions and supplementary answers based on user follow-up questions, and effectively remember the user's interaction history and intent information in multi-turn dialogues to establish a dialogue context association model, ensuring that the platform can accurately understand the evolution and progression of user intent in multi-turn dialogues.

[0021] 2. The voice interaction support module can recognize and convert the voice information received by the digital human into text information. Combined with the general large model, the corresponding output commands can be directly retrieved for digital human control. The output commands include multiple types, supporting multimodal control of the digital human.

[0022] 3. The multi-turn dialogue monitoring module can perform effectiveness monitoring analysis on the multi-turn dialogue between the digital human and the user. Based on the basic parameters of the voice information, the monitoring process is differentiated and marked. The effectiveness of the digital human's multi-turn dialogue within the monitoring period is evaluated by the proportion of marked repetitive processes.

[0023] 4. The effectiveness optimization analysis module can perform effectiveness optimization analysis on the multi-turn dialogue between the digital human and the user. It can calculate the concentration coefficient based on the number of repeated processes corresponding to the output instructions, and then mark the effectiveness optimization measures of the regulatory cycle through the concentration coefficient to improve the efficiency of effectiveness optimization of the digital human's multi-turn dialogue. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a system block diagram of Embodiment 1 of the present invention;

[0026] Figure 2 This is a flowchart of the method in Embodiment 2 of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1: As Figure 1As shown, a digital human interaction management system based on big data analysis includes a voice interaction support module, a multi-turn dialogue monitoring module, and an effectiveness optimization analysis module connected in sequence. The voice interaction support module is communicatively connected to a general large model, which is generated by the large model integration module.

[0029] The large model integration module is used to integrate general large models, ensuring that the platform can intelligently understand and accurately answer a wide range of questions raised by users, covering common fields such as history, technology, culture, business, and life, and can provide reasonable logical extensions and supplementary answers based on user follow-up questions;

[0030] FAQ generation based on knowledge base documents: It can automatically identify and extract key information, core concepts and common questions from the knowledge base documents provided by customers, and automatically generate a structured list of dialogue FAQs based on these contents. The coverage of the generated FAQs should reach more than 80% of the effective information in the documents, and the answers to each FAQ are concise and clear, with an average word count between 100 and 300 words, making them easy for users to understand and interact with.

[0031] A well-designed multi-turn dialogue management logic can effectively remember the user's interaction history and intent information in multiple rounds of dialogue, establish a dialogue context association model, and ensure that the platform can accurately understand the evolution and progression of the user's intent during multiple rounds of dialogue, avoiding situations where the topic deviates or repeated answers occur.

[0032] A high-precision dialogue command intent recognition model can be built to quickly and accurately extract command intents from user dialogue text, such as opening a specific website (e.g., official website), playing a specific video (e.g., company introduction video), or querying specific data (e.g., company operating data from the previous year), with an intent recognition accuracy rate of over 90%.

[0033] We develop a complete front-end software development kit (SDK), encapsulating core functional interfaces such as digital human interaction, speech recognition, and large-scale dialogue models, enabling customers to quickly integrate the system's functionality into their own applications or websites. We also provide detailed interface documentation, including interface function descriptions, parameter descriptions, return value descriptions, and sample code, ensuring that customer developers can easily understand and use the SDK for integration development.

[0034] The business knowledge Q&A platform and large model capability call functions in the digital human system need to provide interfaces for the company's OA system or other systems to call these functions. Interface forms include, but are not limited to: web integration, JS embedding, WeChat official account, API interface, etc. This function call is not limited by the number of users. The usage permissions of users in the OA system or other systems should be integrated with the digital human system's permission management module to achieve different application scenarios based on different role permissions.

[0035] Based on the identified instruction intent, the platform can generate and send corresponding instructions or scripts to relevant systems (such as BI systems or digital human display systems) within 500 milliseconds, ensuring the timeliness and accuracy of instruction execution; it can also provide reasonable logical extensions and supplementary answers based on the user's follow-up questions, and effectively remember the user's interaction history and intent information in multi-turn dialogues, establishing a dialogue context association model to ensure that the platform can accurately understand the evolution and progression of the user's intent during multi-turn dialogues.

[0036] The voice interaction support module is used to recognize and convert the voice information received by the digital human into text information. The text information is input into a general model and the corresponding output commands are retrieved. The output commands include execution signals and question-and-answer signals. The execution signals include limb execution electrical signals corresponding to the text information, and the question-and-answer signals include voice broadcast electrical signals corresponding to the text information. The output commands are sent to the execution module. After receiving the execution signals, the execution module performs limb control according to the execution signals. After receiving the question-and-answer signals, the execution module performs voice broadcast control according to the question-and-answer signals. The digital human is controlled by directly retrieving the corresponding output commands in combination with the general model. The output commands include multiple types, supporting multimodal control of the digital human.

[0037] The multi-turn dialogue monitoring module is used to perform effectiveness monitoring analysis on the multi-turn dialogue between the digital human and the user: it generates a monitoring period, marks the multi-turn dialogue program between the digital human and the user within the monitoring period as the monitoring process, extracts all voice information extracted by the user in the monitoring process, and extracts the basic parameters of the voice information, including information type, information domain and keyword set;

[0038] Voice information with identical information type, information domain, and keyword set during the regulatory process is marked as associated information. It is then determined whether the output instructions retrieved from the general model through the associated information are identical. If so, the regulatory process is marked as a repeated process; otherwise, it is marked as a related process. The ratio of the number of repeated processes to the total number of regulatory processes within a regulatory period is marked as the repetition coefficient of the regulatory period. This repetition coefficient is then compared with a preset repetition threshold.

[0039] If the repetition coefficient is greater than or equal to the repetition threshold, it is determined that the effectiveness of the digital human's multi-turn dialogue within the regulatory period does not meet the requirements, and an effectiveness optimization analysis signal is generated and sent to the effectiveness optimization analysis module.

[0040] If the repetition coefficient is less than the repetition threshold, the validity of the digital human's multi-turn dialogue within the regulatory period is deemed to meet the requirements. A comprehensive analysis of the regulatory period is then conducted: the number of dialogue rounds in the relevant process is marked as a persistence value; processes with persistence values ​​not less than a preset persistence threshold are marked as fully optimized processes; the number of fully optimized processes is marked as a fully optimized value; and the fully optimized value is compared with the preset fully optimized threshold.

[0041] If the overall optimization value is less than the overall optimization threshold, it is determined that the digital human's multi-round dialogues within the regulatory period do not have the characteristics of overall optimization.

[0042] If the overall optimization value is greater than or equal to the overall optimization threshold, the multi-turn dialogue of the digital human within the regulatory period is determined to have overall optimization characteristics. The information types, information domains, and all keywords in the keyword set of all overall optimization processes constitute the type set, domain set, and key set, respectively. The information type, information domain, and keyword with the most corresponding elements in the information type, information domain, and keyword set constitute the overall optimization dataset. An overall optimization signal is generated and the overall optimization signal and the overall optimization dataset are sent to the mobile terminal of the management personnel. Effectiveness regulatory analysis is performed on the multi-turn dialogue between the digital human and the user. The regulatory process is differentiated and marked according to the basic parameters of the voice information. The effectiveness of the multi-turn dialogue of the digital human within the regulatory period is evaluated by the proportion of marked repeated processes.

[0043] The effectiveness optimization analysis module is used to perform effectiveness optimization analysis on multi-turn dialogues between the digital human and the user: Repeated processes corresponding to output instructions within the monitoring period are marked as the central tendency values ​​of the output instructions; variance is calculated for all central tendency values ​​to obtain the central tendency coefficients; and these coefficients are compared with preset central tendency thresholds.

[0044] If the concentration coefficient is less than the concentration threshold, the effectiveness optimization measures of the regulatory cycle will be marked as voice recognition optimization, a voice recognition optimization signal will be generated and sent to the mobile terminal of the management personnel;

[0045] If the concentration coefficient is greater than or equal to the concentration threshold, the effectiveness optimization measures of the regulatory cycle are marked as information integration optimization. The text information corresponding to the K1 repetitive processes with the largest concentration values ​​is marked as integration optimization objects. An information integration optimization signal is generated and sent to the mobile terminal of the management personnel along with the integration optimization objects. The effectiveness optimization analysis of the multi-turn dialogue between the digital human and the user is performed. The concentration coefficient is calculated by comprehensively based on the number of repetitive processes corresponding to the output instructions. The effectiveness optimization measures of the regulatory cycle are then marked by the concentration coefficient to improve the efficiency of the effectiveness optimization of the multi-turn dialogue between the digital human.

[0046] Example 2: Figure 2As shown, a digital human interaction management method based on big data analysis includes the following steps:

[0047] Step 1: Generate a general large model, a FAQ based on knowledge base documents, design a complete multi-turn dialogue management logic, and build a high-precision dialogue instruction intent recognition model;

[0048] Step 2: Recognize the voice information received by the digital human and convert it into text information. Input the text information into the general large model and retrieve the corresponding output instructions. Send the output instructions to the execution module.

[0049] Step 3: Conduct effectiveness monitoring analysis on multi-turn dialogues between digital humans and users: generate monitoring cycles, mark the multi-turn dialogue procedures between digital humans and users within the monitoring cycle as monitoring processes, and mark the monitoring processes as related processes or repeated processes;

[0050] Step 4: Mark the ratio of the number of repeated processes to the number of regulatory processes within the regulatory period as the repetition coefficient of the regulatory period. Use the repetition coefficient to determine whether the effectiveness of the digital human's multi-turn dialogue within the regulatory period meets the requirements.

[0051] Step 5: Conduct effectiveness optimization analysis on the multi-turn dialogue between the digital human and the user: Mark the repeated processes corresponding to the output instructions within the regulatory period as the central tendency of the output instructions, calculate the variance of all central tendency of the output instructions to obtain the central tendency coefficient, and mark the effectiveness optimization measures of the regulatory period through the central tendency coefficient.

[0052] A digital human interaction management system based on big data analysis generates a general large model and FAQs based on knowledge base documents during operation, designs a comprehensive multi-turn dialogue management logic, and constructs a high-precision dialogue instruction intent recognition model.

[0053] Step Two: Recognize and convert the voice information received by the digital human into text information. Input the text information into the general large model and retrieve the corresponding output instructions. Send the output instructions to the execution module. Generate a monitoring period. Mark the multi-turn dialogue program between the digital human and the user within the monitoring period as a monitoring process. Mark the monitoring process as a related process or a repeated process. Mark the ratio of the number of repeated processes to the number of monitoring processes within the monitoring period as the repetition coefficient of the monitoring period. Use the repetition coefficient to determine whether the effectiveness of the multi-turn dialogue of the digital human within the monitoring period meets the requirements. Mark the repeated processes corresponding to the output instructions within the monitoring period as the set value of the output instructions. Calculate the variance of the set values ​​of all output instructions to obtain the set coefficient. Use the set coefficient to mark the effectiveness optimization measures of the monitoring period.

[0054] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0055] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0056] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A digital human interaction management system based on big data analysis, characterized in that, This includes a voice interaction support module, a multi-turn dialogue monitoring module, and an effectiveness optimization analysis module, which are connected in sequence. The voice interaction support module recognizes the voice information received by the digital human and converts it into text information. It inputs the text information into a general large model and retrieves the corresponding output command, which is then sent to the execution module. The multi-turn dialogue monitoring module generates a monitoring period, marks the multi-turn dialogue process between the digital human and the user within the monitoring period as a monitoring process, extracts all voice information extracted by the user during the monitoring process, and extracts the basic parameters of the voice information, including information type, information domain, and keyword set; it marks voice information with the same information type, information domain, and keyword set during the monitoring process as associated information, and marks the monitoring process as a repeated process or related process through the associated information; it marks the ratio of the number of repeated processes to the number of monitoring processes within the monitoring period as the repetition coefficient of the monitoring period, and uses the repetition coefficient to determine whether the effectiveness of the multi-turn dialogue of the digital human within the monitoring period meets the requirements; The effectiveness optimization analysis module performs effectiveness optimization analysis on the multi-turn dialogue between the digital human and the user. The specific process for marking a regulatory process as a repetitive or related process includes: determining whether the output instructions retrieved from the general large model through the correlation information are the same; if so, the regulatory process is marked as a repetitive process; if not, the regulatory process is marked as a related process. The specific process for determining whether the effectiveness of the digital human's multi-turn dialogue within the regulatory period meets the requirements includes comparing the repetition coefficient with a preset repetition threshold: if the repetition coefficient is greater than or equal to the repetition threshold, it is determined that the effectiveness of the digital human's multi-turn dialogue within the regulatory period does not meet the requirements, an effectiveness optimization analysis signal is generated and sent to the effectiveness optimization analysis module; if the repetition coefficient is less than the repetition threshold, it is determined that the effectiveness of the digital human's multi-turn dialogue within the regulatory period meets the requirements, and a comprehensive analysis of the regulatory period is conducted. The effectiveness optimization analysis module is used to perform effectiveness optimization analysis on the multi-turn dialogue between the digital human and the user: the repeated process corresponding to the output instruction within the regulatory period is marked as the central value of the output instruction, the variance of all central values ​​of the output instruction is calculated to obtain the central coefficient, and the effectiveness optimization measures of the regulatory period are marked by the central coefficient; The specific process of marking the effectiveness optimization measures of the regulatory cycle includes: comparing the concentration coefficient with a preset concentration threshold; if the concentration coefficient is less than the concentration threshold, the effectiveness optimization measures of the regulatory cycle are marked as voice recognition optimization, a voice recognition optimization signal is generated and sent to the mobile terminal of the management personnel; if the concentration coefficient is greater than or equal to the concentration threshold, the effectiveness optimization measures of the regulatory cycle are marked as information integration optimization, the text information corresponding to the K1 repeated processes with the largest concentration values ​​is marked as integration optimization objects, an information integration optimization signal is generated and the information integration optimization signal and integration optimization objects are sent to the mobile terminal of the management personnel.

2. The digital human interaction management system based on big data analysis according to claim 1, characterized in that, The output instructions include execution signals and question-and-answer signals; the execution signals include limb execution electrical signals corresponding to the text information, and the question-and-answer signals include voice broadcast electrical signals corresponding to the text information; after receiving the execution signal, the execution module performs limb control according to the execution signal, and after receiving the question-and-answer signal, the execution module performs voice broadcast control according to the question-and-answer signal.

3. The digital human interaction management system based on big data analysis according to claim 2, characterized in that, The specific process for conducting a comprehensive analysis of the regulatory period includes: marking the number of dialogue rounds in relevant processes as a persistence value; marking relevant processes with persistence values ​​not less than a preset persistence threshold as fully optimized processes; marking the number of fully optimized processes as a fully optimized value; comparing the fully optimized value with a preset fully optimized threshold: if the fully optimized value is less than the fully optimized threshold, it is determined that the multi-turn dialogue of the digital human within the regulatory period does not have fully optimized characteristics; if the fully optimized value is greater than or equal to the fully optimized threshold, it is determined that the multi-turn dialogue of the digital human within the regulatory period has fully optimized characteristics; the information types, information domains, and all keywords in the keyword set of all fully optimized processes constitute a type set, a domain set, and a key set, respectively; the information type, information domain, and keyword with the most corresponding elements in the information type, information domain, and keyword set constitute a fully optimized dataset; generating a fully optimized signal; and sending the fully optimized signal and the fully optimized dataset to the mobile terminal of the management personnel.

4. The digital human interaction management system based on big data analysis according to claim 3, characterized in that, The voice interaction support module communicates with the general large model, which is generated by the large model integration module. The large model integration module is used to generate the general large model, FAQs based on knowledge base documents, design multi-turn dialogue management logic, and construct a dialogue command intent recognition model.

Images