Financial digital human system

Abstract

The invention provides a financial digital human system, and relates to the field of financial intelligent consultation. The system comprises an AI financial digital human image management module, a character interaction module, a voice interaction module, an AI digital human animation effect module, a digital human scene switching module, a digital human interaction module, a user interface module, a knowledge base module and an access mode module. The digital person can intelligently and automatically operate for 24 hours, and smooth self-service business handling service is provided for a user. The method has the beneficial effects that the emotion analysis algorithm is introduced, so that the answer style and tone can be adjusted according to the emotional state of the user, and more intimate and humanized services are provided; three interpolation algorithms are introduced to optimize the actions of the digital human aiming at different conditions, so that the suitability of action optimization is improved, the digital human performs scene switching according to the use environment and time of the user, and the use comfort degree of the user can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of financial intelligence consultation, in particular to a financial digital human system. BACKGROUND

[0002] With the accelerated development of global digital economy, enterprise financial management is undergoing a deep transformation from traditional manual operation to intelligent and scenario-based direction. The traditional financial service mode is facing multiple challenges: Existing financial systems mainly use static interfaces such as forms and buttons, lacking emotional interaction, and users may feel bored during use, especially when querying complex business such as reimbursement policy consultation and budget analysis, it is difficult to obtain intuitive and efficient feedback. Financial consultation and approval processes rely on manual responses, with long processing cycles and high error rates. For example, employees need to repeatedly submit materials and wait for review during cross-border travel reimbursement, resulting in low process efficiency. Enterprise financial departments often act as "backstage support" and lack emotional connection with users, making it difficult to deliver brand value through services.

[0003] Under this background, AI financial digital human becomes the key to breaking through. By integrating natural language processing, multi-modal interaction, 3D visualization and other technologies, AI digital human can reshape the form of financial services. With a tangible digital image replacing traditional interfaces, expressions, actions and voices can enhance the sense of reality of interaction; 7x24 hour service: breaking through the time limit of manual service, responding to user needs in real time; brand value carrier: delivering enterprise culture and professional image through customized image design.

[0004] In the current AI financial digital human industry, there are still problems such as poor user interaction experience, insufficient brand image shaping, and backward technology application in the aspect of intelligent interaction from a financial professional perspective. SUMMARY

[0005] The main purpose of the present application is to provide a financial digital human system to solve the problems in the background art.

[0006] To solve the above technical problems, the technical solution adopted by the present application is: a financial digital human system, comprising: An AI financial digital human image management module, which strengthens the emotional connection between users and the brand through scenario-based and diversified image design, and realizes flexible switching of digital human image between 2D / 3D forms; A text interaction module connected with the AI financial digital human image management module, which accurately understands and professionally responds to user text input, and simultaneously realizes emotional interaction through emotional analysis; A voice interaction module connected with the AI financial digital human image management module, which realizes accurate conversion and natural response of voice commands through voice recognition and synthesis technology, and makes voice interaction more humanized through emotional analysis; The AI digital human animation effect module is connected with the text interaction module and the voice interaction module, generates an adaptive motion sequence according to the emotion and semantics of the interaction content, and optimizes the fluency of the motion; The digital human scene switching module is connected with the AI digital human image management module and the AI digital human animation effect module, and is used to realize dynamic adaptation of the digital human image and the financial service scene, and enhance the scene substitution of the user; The digital human interaction module is connected with the AI financial digital human image management module, the text interaction module, the voice interaction module, the AI digital human animation effect module, and the digital human scene switching module, and is used to increase the interactive form and trigger personalized interactive actions; The user interface module is connected with the AI financial digital human image management module, the text interaction module, the voice interaction module, the AI digital human animation effect module, the digital human scene switching module, and the digital human interaction module, is responsible for the front-end rendering of the 2D / 3D image, expression motion, and scene picture of the financial digital human, converts the visual design results of the AI digital human image management module and the digital human scene switching module into interface effects visible to the user, and displays information such as financial question and answer results and business process guidance; The knowledge base module is connected with the text interaction module, the voice interaction module, the user interface module, and the digital human interaction module, and is used to include structured content The access mode module is connected with other modules, and is used to provide different access modes and ensure smooth access experience of the user on different device ends.

[0007] Further, the AI financial digital human image management module includes: A core image design unit is used to design a formal core image integrated with brand elements, and focuses on the creation of a basic digital human image; An image expansion unit is used to develop differentiated images for various scenes such as business trips and meetings; An image presentation unit realizes switching and rendering of 2D / 3D images; A dynamic effect unit constructs and calls an expression library and a motion library.

[0008] Further, the text interaction module includes: A semantic analysis unit is used to clean, tokenize, extract semantics, and identify the intent of the user's text input; An emotion analysis unit is used to screen emotional words in the user's text, determine the emotional tendency in combination with an algorithm, and realize emotion quantification through a formula; The text emotion quantification uses a weighted algorithm, and the expression is: (1); Wherein, S 1 is the text emotion score, N1 is the total number of text sentiment words, is the weight of the i th text sentiment word, is the base score of the i th text sentiment word; Action matching unit, according to the results of sentiment analysis, drive digital people to show corresponding expressions and actions; Text feedback unit, retrieve information from financial knowledge base, generate professional and easy-to-understand text answers and output.

[0009] Further, the voice interaction module includes: Voice recognition unit, convert user's voice instructions into text input; Speech synthesis unit, generate natural and emotional voice answers; Voice emotion analysis unit, judge and quantify the emotional tendency of the content after voice-to-text, and capture the user's voice emotion; Voice emotion quantification also uses a weighting algorithm, the expression is: (2); Where, S 2 is the voice emotion score, N 2 is the total number of voice emotion words, is the weight of the i th voice emotion word, is the base score of the i th voice emotion word; Voice-action mapping unit, according to the results of voice emotion analysis, drive digital people to generate matching actions.

[0010] Further, the digital human interaction module includes: Behavior monitoring unit, real-time monitoring of user login, click digital people, long time no operation and other behaviors, capture interaction trigger signal; Action triggering unit, according to the monitoring results, call the corresponding interactive action from the action library; Random action unit, combined with randomization algorithm, show diversified actions in specific scenarios; The determination of random action depends on random number generation. In the scenario where random action needs to be shown, a random number is generated; according to the range of random number and the number of actions in the action library, the action to be shown is determined; When the user has not operated for a long time, some lively actions are added.

[0011] Further, the AI digital human animation effect module includes: Content emotion analysis unit, analyze the emotion type and intensity of the content after text or voice conversion, and provide emotional basis for response action generation; Response sentiment quantification also uses a weighted algorithm, expressed as: (3); in, S 3 is the response emotion score, N 3 is the total number of emotional words in response. For the first i The weight of each emotional response word For the first i The basic score of each response sentiment vocabulary; this quantifies the sentiment tendency in the system's output response content; The action sequence generation unit retrieves and combines action sequences from the action library based on the sentiment analysis results and semantic content to generate suitable action sequences. The motion optimization unit uses motion interpolation technology to optimize the smooth transition of digital human motion and avoid abrupt and stiff movements. Three different interpolation methods are used for motion interpolation: linear interpolation, which is suitable for simple uniform motion; Bezier interpolation, which is suitable for natural motion that requires acceleration changes; and spherical linear interpolation, which is suitable for rotational motion of the head and joints. The principle of linear interpolation is: (4); in, P ( t (The percentage of time is) t The corresponding linear interpolation position, P i The starting frame coordinates, P f The coordinates of the end frame. t This represents the percentage of time spent, and its value ranges from [0,1]. The principle of Bezier interpolation is: (5); in, B ( t (The percentage of time is) t The corresponding Bezier interpolation position, P 1 represents a single control point of the Bezier curve, used to determine the direction of curvature of the motion curve; The principle of spherical linear interpolation is as follows: (6); in, The time percentage is t The corresponding spherical linear interpolation position, , These are the starting rotation angle and the ending rotation angle. for and The included angle.

[0012] Furthermore, the digital human scene switching module includes: The scene image design unit designs exclusive digital human images for various financial service scenarios; The resource library management unit stores and manages the resource materials for various scene images; The switching control unit enables dynamic and smooth switching of the digital human's image; the system's preset rules use a multi-dimensional triggering rule, which is divided into two dimensions: user operation triggering and time / environment triggering. The switching effect evaluation unit monitors the response time and smoothness of image switching, collects user feedback, and optimizes the switching experience.

[0013] Furthermore, the user interface module includes: The interface rendering unit is responsible for the visual rendering of the digital human avatar and interactive interface; The operation capture unit captures various user operations through event listening technology, and then standardizes the operation signals and passes them to the corresponding backend modules for processing. The device adaptation unit enables the interface to adapt and adjust itself on PC and mobile devices; at the same time, it optimizes the interface loading speed and rendering accuracy to take into account the performance differences of different devices. The feedback display unit is responsible for transforming the processing results of the backend modules into user-perceptible interface feedback.

[0014] Furthermore, the knowledge base module includes: The knowledge storage unit collects structured content and synchronizes new rules in real time through the knowledge update unit; The intelligent retrieval unit quickly searches based on the content of the user's question and extracts the answer from the knowledge storage unit.

[0015] Furthermore, the access mode module includes: The client-side optimization unit optimizes page loading speed, interaction logic, and function layout for PC and mobile devices, respectively, based on user habits and performance requirements. The efficiency assessment unit monitors and evaluates the business processing efficiency of different devices, providing data support for edge optimization.

[0016] Beneficial effects: (1) By introducing sentiment analysis algorithms, the style and tone of the response can be adjusted according to the user's emotional state, providing more considerate and humanized services; (2) Three interpolation algorithms are introduced to optimize the movements of the digital human under different conditions, thereby improving the adaptability of the movement optimization. (3) Digital humans can switch scenes according to the user's environment and time, which can improve the user's comfort. Attached Figure Description

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram of the system connection of the present invention. Detailed Implementation

[0018] Example 1 like Figure 1 As shown, a financial digital human system includes: The AI-powered financial digital avatar image management module, serving as the system's "visual core," is responsible for creating digital avatar IP images with characteristics specific to the financial industry and strong corporate brand recognition. Through scenario-based and diversified image design, it strengthens the emotional connection between users and the brand, addressing the issues of traditional financial services lacking a unique image and having weak brand communication. It also enables flexible switching between 2D and 3D forms of the digital avatar image, adapting to the visual needs of different interactive scenarios. The module includes: The core image design unit is used to design a formal core image that incorporates brand elements. It focuses on creating a basic image for digital humans, combining the professional and rigorous characteristics of the financial field to design a formal core image, and incorporating brand elements such as the company logo and theme colors to establish the visual tone of digital humans. The Image Expansion Unit is used to develop differentiated images for various scenarios such as business trips and meetings. It develops differentiated digital human images for high-frequency financial service scenarios such as business trips, meetings, and daily office work to meet the visual adaptation needs of diverse scenarios; for example, using a casual business style for business trip scenarios and a formal dress style for meeting scenarios. The image presentation unit enables the switching and rendering of 2D / 3D images. This unit can switch between 2D flat images and 3D high-precision modeled images and render them in real time. The 2D images ensure simple interaction, while the 3D images enhance visual realism. The Dynamic Effects Unit builds and calls upon the expression and action libraries. This unit can build a rich expression and action library to provide digital humans with natural dynamic expression materials, ensuring that the image is vivid and lively in interaction.

[0019] The text interaction module, connected to the AI-powered financial digital avatar management module, serves as the core channel for text communication between users and the digital avatar. Leveraging natural language processing technology, it achieves accurate understanding and professional responses to user text input. Simultaneously, it incorporates sentiment analysis to enable emotional interaction, addressing the issues of misunderstanding, monotonous format, and lack of emotion in traditional financial system text interactions. The module includes: The semantic parsing unit cleans, segments, extracts semantics, and identifies intent from user text input, accurately capturing the user's financial needs and true intent. The sentiment analysis unit filters the sentiment words in the user's text, judges the emotional tendency by combining algorithms, and realizes sentiment quantification through formulas; The sentiment quantification of the text uses a weighted algorithm, and the expression is: (1); Among them, S $S$ is the text sentiment score, N $n$ is the total number of text sentiment words, $w_i$ is the weight of the i $i$-th text sentiment word, $s_i$ is the basic score of the i $i$-th text sentiment word; The quantification here is for the sentiment words and semantics of this passage of text, capturing the user's text emotions; The action matching unit, according to the sentiment analysis result, drives the digital human to display the corresponding expressions and actions, realizing the linkage of text and visual interaction; The text feedback unit retrieves information from the financial knowledge base, generates professional and easy-to-understand text answers and outputs them, forming a text interaction closed loop.

[0020] The voice interaction module builds a voice communication bridge between the user and the digital human, realizes the accurate conversion and natural response of voice commands through voice recognition and synthesis technologies, and combines sentiment analysis to make voice interaction more user-friendly, solving the problems of lack of voice interaction channels and low communication efficiency in traditional financial services. The module includes: The voice recognition unit converts the user's voice commands into text input, providing a basis for subsequent parsing and response; The voice synthesis unit, through the TTS system driven by deep learning, generates natural and emotional voice answers, ensuring the fluency and affinity of voice output; The TTS system has 4 functions: text preprocessing, acoustic feature modeling, voice synthesis, and post-processing; The text preprocessing filters special symbols in the text, such as the summation symbol, percentage, etc. in financial formulas, redundant spaces, and at the same time converts financial professional abbreviations, such as "ERP", "OA", into full names that can be read aloud by voice; performs voice adaptation processing on numbers, for example, converts "¥5000" to "Five thousand yuan in RMB"; In acoustic feature modeling, through the Transformer encoder in the deep learning model, the preprocessed text sequence is converted into text features in the form of high-dimensional vectors, capturing the semantic, syntactic, and emotional label information of the text. For example, the text features of "I'm sorry for the inconvenience" will incorporate the emotional dimension of "apology"; The LSTM decoder predicts the corresponding acoustic feature parameters according to the text features; The acoustic feature parameters include: Duration characteristics determine the pronunciation duration of each syllable. For example, the pronunciation duration of the financial term "budget preparation" is slightly longer than that of ordinary words to ensure clear distinction. The fundamental frequency feature corresponds to the pitch of the speech. The fundamental frequency is adjusted to match the emotion label. For example, "positive" emotion corresponds to a higher fundamental frequency, while "apology" emotion corresponds to a lower and flatter fundamental frequency. Spectral characteristics determine the timbre of the voice. Based on the professional attributes of financial services, a "professional and calm" basic timbre is preset, while fine-tuning is made according to emotional tags, such as making the timbre softer with the emotion of "patience". The voice sentiment analysis unit judges and quantifies the emotional tendency of the speech-to-text content, capturing the user's voice emotions; the voice sentiment quantification also uses a weighted algorithm, the expression of which is: (2); in, S 2 represents the voice emotion score. N 2 represents the total number of words expressing emotional sentiment. For the first i The weight of each emotional word in speech. For the first i The basic score for each voice emotion vocabulary; this quantifies the emotional tendency in the text content converted from the user's voice. The voice-action mapping unit drives the digital human to generate matching actions based on the voice emotion analysis results, thereby achieving synchronization between voice and visual actions.

[0021] The AI ​​digital human animation effects module connects with the text and voice interaction modules, giving digital humans dynamic and natural animation capabilities. It generates adaptive action sequences based on the emotion and semantics of the interactive content, optimizing motion smoothness and solving the problems of stiff digital human movements and disconnect from content. This enhances the fun and immersion of the interaction. The module includes: The content sentiment analysis unit analyzes the sentiment type and intensity of the text or speech-converted content to provide emotional basis for generating response actions. Response sentiment quantification also uses a weighted algorithm, expressed as: (3); in, S 3 is the response emotion score, N 3 is the total number of emotional words in response. For the first i The weight of each emotional response word For the first i The basic score of each response sentiment vocabulary; this quantifies the sentiment tendency in the system's output response content; The action sequence generation unit retrieves and combines action sequences from the action library based on the sentiment analysis results and semantic content to generate suitable action sequences. The motion optimization unit uses motion interpolation technology to optimize the smooth transition of digital human motion and avoid abrupt and stiff movements. Three different interpolation methods are used for motion interpolation: linear interpolation, which is suitable for simple uniform motion; Bezier interpolation, which is suitable for natural motion that requires acceleration changes; and spherical linear interpolation, which is suitable for rotational motion of the head and joints. The principle of linear interpolation is: (4); in, P ( t (The percentage of time is) t The corresponding linear interpolation position, P i The starting frame coordinates, P f The coordinates of the end frame. t This represents the percentage of time spent, and its value ranges from [0,1]. The principle of Bezier interpolation is: (5); in, B ( t (The percentage of time is) t The corresponding Bezier interpolation position, P 1 represents a single control point of the Bezier curve, used to determine the direction of curvature of the motion curve; The principle of spherical linear interpolation is as follows: (6); in, The time percentage is t The corresponding spherical linear interpolation position, , These are the starting rotation angle and the ending rotation angle. for and The included angle.

[0022] The digital human scene switching module, connected to the AI ​​digital human image management module and the AI ​​digital human animation effects module, is used to achieve dynamic adaptation of the digital human image to financial service scenarios, enhancing the user's sense of immersion. The module includes: The scene image design unit designs exclusive digital human images for financial service scenarios such as work, business trips, and meetings, enriching the visual expression. The material library management unit stores and manages the material resources for various scene images, ensuring the supply of materials for image switching; The switching control unit enables dynamic and smooth switching of the digital human's image based on user selection or system preset rules. The system preset rules use a multi-dimensional triggering rule, which is divided into two dimensions: user operation triggering and time / environment triggering. User operation trigger dimension: When a user makes a triggering operation or has the intention to switch, the digital human image is switched. This includes: when a user actively selects scene tags such as "business trip expense reimbursement", "meeting budget" and "daily office consultation" in the user interface module, the system immediately triggers scene switching. When a user clicks the "Scene Change" button next to the digital human avatar, or says "Switch to the business trip scene" or "I want to see the digital human in the meeting scene" via text / voice commands, the switching command is triggered. When a user continuously inputs / speaks keywords that are strongly related to a certain scenario during the interaction, and the keywords appear more than twice, the system will automatically trigger the corresponding scenario switch. Time / Environment Trigger Dimension: This means that the digital human displays different appearances depending on the time and location. During regular office hours from 9:00 to 18:00 on weekdays, the default display is "Daily Office Scene"; During non-office hours, the "Nighttime Rest Scene" will be displayed. If a user initiates a financial inquiry at this time, the system will be linked to the user's action trigger and switch to the "Nighttime Service Scene". When the system detects the user's device location information and it shows that the user is in a foreign location, it automatically switches to the "business trip scenario"; if the location is within the company, it maintains the "daily office scenario", and the location information needs to be authorized by the user. The switching effect evaluation unit monitors the response time and smoothness of image switching, collects user feedback, and optimizes the switching experience.

[0023] The digital human interaction module connects with the AI ​​financial digital human image management module, text interaction module, voice interaction module, AI digital human animation effects module, and digital human scene switching module. It is used to increase interactive formats, trigger personalized interactive actions, and assess and mitigate compliance risks during interaction. The module includes: The behavior monitoring unit monitors user behaviors such as login, clicking on the digital human, and long periods of inactivity in real time, and captures interaction trigger signals. The action triggering unit calls the corresponding interactive action from the action library based on the listening results, such as greeting or giving a thumbs up. Random action units, combined with randomization algorithms, can display diverse actions in specific scenarios, avoiding the monotony of interactive actions; The determination of random actions relies on random number generation. In scenarios where random actions need to be displayed, the JavaScript `Math.random()` method is used to generate random numbers. Based on the range of the random number and the number of actions in the action library, the action to be displayed is determined. For example, if there are 10 actions in the action library, each action corresponds to a number between 0 and 9. Then, a random number between 0 and 9 is generated, and the corresponding action is selected based on the value of the random number. When users are inactive for a long time, increase the probability of some lively actions, such as jumping, smiling and swaying, to attract their attention.

[0024] The user interface module connects with the AI ​​financial digital human image management module, text interaction module, voice interaction module, AI digital human animation effects module, digital human scene switching module, and digital human interaction module. It is responsible for the front-end rendering of the financial digital human's 2D / 3D image, expressions, actions, and scene visuals. It transforms the visual design results of the AI ​​digital human image management module and digital human scene switching module into user-visible interface effects, while also displaying financial Q&A results, business process guidance, and other information, making the system functions visible. The module includes: The interface rendering unit is responsible for the visual rendering of the digital human image and interactive interface. Specifically, it includes front-end loading and real-time rendering of the digital human 2D / 3D model, frame animation display of facial expressions and actions, and drawing and switching of scene backgrounds. At the same time, it renders interactive controls such as text input boxes, voice buttons, and function panels to ensure the visual presentation effect of interface elements. The operation capture unit captures various user operations through event listening technology, such as text input, voice recording trigger, scene tag click, digital human interaction, etc., and transmits the operation signals to the corresponding backend modules for processing after standardization. The device adaptation unit uses front-end frameworks such as Bootstrap to achieve adaptive adjustments of the interface on PC and mobile devices; at the same time, it optimizes interface loading speed and rendering accuracy to ensure a consistent experience across different devices, taking into account the performance differences of different devices. The feedback display unit is responsible for transforming the processing results of the backend modules into user-perceptible interface feedback, such as displaying text replies from the text interaction module, playing synthesized speech from the voice interaction module, and simultaneously presenting action commands from the AI ​​digital human animation effect module. It can also display progress prompts and result notifications for financial transactions.

[0025] The knowledge base module, connected to the text interaction module, voice interaction module, user interface module, and digital human interaction module, is used to collect structured content such as national financial regulations, corporate internal systems, and tax policies; the module includes: The knowledge storage unit contains structured content such as national financial regulations, corporate internal systems, and tax policies, and synchronizes with new regulations in real time through the knowledge update unit to ensure the timeliness of knowledge. The intelligent retrieval unit quickly retrieves answers from the knowledge storage unit based on user queries using keywords and semantic analysis.

[0026] The access mode module, connected to other modules, ensures a smooth user experience across different devices, such as PCs and mobile devices. Through responsive design and client-side optimization, it adapts page layout and interaction performance, resolving the issues of poor cross-device access and slow loading in traditional financial systems. The module includes: The client-side optimization unit optimizes page loading speed, interaction logic, and function layout for PC and mobile devices, respectively, based on user habits and performance requirements. The efficiency assessment unit monitors and evaluates the business processing efficiency of different devices, providing data support for edge optimization.

[0027] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A financial digital human system, characterized in that, include: The AI-powered financial digital human image management module strengthens the emotional connection between users and brands through scenario-based and diversified image design, and enables flexible switching between 2D and 3D digital human images. The text interaction module connects with the AI ​​financial digital human image management module, providing accurate understanding and professional responses to user text input, while incorporating sentiment analysis to achieve emotional interaction; The voice interaction module connects with the AI ​​financial digital human image management module. It uses voice recognition and synthesis technology to achieve accurate conversion and natural response of voice commands, and combines emotion analysis to make the voice interaction more humanized. The AI ​​digital human animation effect module connects with the text interaction module and the voice interaction module to generate an appropriate action sequence based on the emotion and semantics of the interactive content, thereby optimizing the smoothness of the action. The digital human scene switching module is connected to the AI ​​digital human image management module and the AI ​​digital human animation effect module to realize the dynamic adaptation of the digital human image to the financial service scenario and enhance the user's sense of immersion in the scenario. The digital human interaction module connects with the AI ​​financial digital human image management module, text interaction module, voice interaction module, AI digital human animation effect module, and digital human scene switching module to increase interaction methods and trigger personalized interactive actions. The user interface module connects with the AI ​​financial digital human image management module, text interaction module, voice interaction module, AI digital human animation effect module, digital human scene switching module, and digital human interaction module. It is responsible for the front-end rendering of the 2D / 3D image, facial expressions, actions, and scene images of the financial digital human. It transforms the visual design results of the AI ​​digital human image management module and the digital human scene switching module into interface effects that are visible to users, while displaying information such as financial Q&A results and business process guidance. The knowledge base module connects to the text interaction module, voice interaction module, user interface module, and digital human interaction module, and is used to collect structured content. The access mode module connects with other modules to provide different access modes, ensuring a smooth access experience for users on different devices.

2. The financial digital human system according to claim 1, characterized in that, The AI-powered financial digital avatar management module includes: The core image design unit is used to design formal core images that incorporate brand elements, focusing on creating a basic digital human image; Image expansion unit, used to develop differentiated images for various scenarios such as business trips and meetings; The image presentation unit enables the switching and rendering of 2D / 3D images; Dynamic effects unit, which builds and calls emoji and action libraries.

3. The financial digital human system according to claim 1, characterized in that, The text interaction module includes: The semantic parsing unit cleans, segments, extracts semantics, and identifies intent from user text input. The sentiment analysis unit filters sentiment words in user text, combines algorithms to determine sentiment tendencies, and quantifies sentiment through formulas. Text sentiment quantification uses a weighted algorithm, expressed as: （1）； in, S 1 represents the text sentiment score. N 1 represents the total number of sentiment words in the text. For the first i The weight of each sentiment word in the text. For the first i The basic score of each text's sentiment vocabulary; The motion matching unit drives the digital human to display corresponding expressions and actions based on the sentiment analysis results; The text feedback unit retrieves information from the financial knowledge base, generates professional and easy-to-understand text answers, and outputs them.

4. The financial digital human system according to claim 1, characterized in that, The voice interaction module includes: The voice recognition unit converts the user's voice commands into text input; The speech synthesis unit generates natural and emotional speech responses; The voice sentiment analysis unit judges and quantifies the emotional tendency of the speech-to-text content, capturing the user's voice emotions; the voice sentiment quantification also uses a weighted algorithm, the expression of which is: （2）； in, S 2 represents the voice emotion score. N 2 represents the total number of words expressing emotional sentiment. For the first i The weight of each emotional word in speech. For the first i The basic score of each voice emotion lexical; The voice-action mapping unit drives the digital human to generate matching actions based on the voice emotion analysis results.

5. The financial digital human system according to claim 1, characterized in that, The digital human interaction module includes: The behavior monitoring unit monitors user behaviors such as login, clicking on the digital human, and long periods of inactivity in real time, and captures interaction trigger signals. The action triggering unit calls the corresponding interactive action from the action library based on the listening results; Random action units, combined with randomization algorithms, can display diverse actions in specific scenarios; The determination of random actions relies on random number generation. When a scenario requires random actions to be displayed, random numbers are generated. Based on the range of the random numbers and the number of actions in the action library, the actions to be displayed are determined. Add some lively actions when users have not interacted with the system for a long time.

6. The financial digital human system according to claim 1, characterized in that, The AI ​​digital human animation effects module includes: The content sentiment analysis unit analyzes the sentiment type and intensity of the text or speech-converted content to provide emotional basis for generating response actions. Response sentiment quantification also uses a weighted algorithm, expressed as: （3）； in, S 3 is the response emotion score, N 3 is the total number of emotional words in response. For the first i The weight of each emotional response word For the first i The basic score of each response sentiment vocabulary; this quantifies the sentiment tendency in the system's output response content; The action sequence generation unit retrieves and combines action sequences from the action library based on the sentiment analysis results and semantic content to generate suitable action sequences. The motion optimization unit uses motion interpolation technology to optimize the smooth transition of digital human motion and avoid abrupt and stiff movements. Three different interpolation methods are used for motion interpolation: linear interpolation, which is suitable for simple uniform motion; Bezier interpolation, which is suitable for natural motion that requires acceleration changes; and spherical linear interpolation, which is suitable for rotational motion of the head and joints. The principle of linear interpolation is: （4）； in, P ( t (The percentage of time is) t The corresponding linear interpolation position, P i The starting frame coordinates, P f The coordinates of the end frame. t This represents the percentage of time spent, and its value ranges from [0,1]. The principle of Bezier interpolation is: （5）； in, B ( t (The percentage of time is) t The corresponding Bezier interpolation position, P 1 is a single control point of the Bezier curve, used to determine the direction of curvature of the motion curve; The principle of spherical linear interpolation is as follows: （6）； in, The time percentage is t The corresponding spherical linear interpolation position, , These are the starting rotation angle and the ending rotation angle. for and The included angle.

7. The financial digital human system according to claim 1, characterized in that, The digital human scene switching module includes: The scene image design unit designs exclusive digital human images for various financial service scenarios; The resource library management unit stores and manages the resource materials for various scene images; The switching control unit enables dynamic and smooth switching of the digital human's image; the system's preset rules use a multi-dimensional triggering rule, which is divided into two dimensions: user operation triggering and time / environment triggering. The switching effect evaluation unit monitors the response time and smoothness of image switching, collects user feedback, and optimizes the switching experience.

8. The financial digital human system according to claim 1, characterized in that, The user interface module includes: The interface rendering unit is responsible for the visual rendering of the digital human avatar and interactive interface; The operation capture unit captures various user operations through event listening technology, and then standardizes the operation signals and passes them to the corresponding backend modules for processing. The device adaptation unit enables the interface to adapt and adjust itself on PC and mobile devices; at the same time, it optimizes the interface loading speed and rendering accuracy to take into account the performance differences of different devices. The feedback display unit is responsible for transforming the processing results of the backend modules into user-perceptible interface feedback.

9. A financial digital human system according to claim 1, characterized in that, The knowledge base module includes: The knowledge storage unit collects structured content and synchronizes new rules in real time through the knowledge update unit; The intelligent retrieval unit quickly searches based on the content of the user's question and extracts the answer from the knowledge storage unit.

10. A financial digital human system according to claim 1, characterized in that, The access mode module includes: The client-side optimization unit optimizes page loading speed, interaction logic, and function layout for PC and mobile devices, respectively, based on user habits and performance requirements. The efficiency assessment unit monitors and evaluates the business processing efficiency of different devices, providing data support for edge optimization.

Images