AI digital human-based customer service platform, service method, equipment and medium

Abstract

The invention provides a customer service platform based on an AI digital human, a service method, equipment and a medium, and relates to the field of intelligent customer service, and the platform comprises a front-end display layer which is used for generating a chat interface for interaction between a user and the AI digital human; the digital human customer service interaction layer is used for performing sentiment analysis on the query content to obtain a sentiment category corresponding to the query content; matching a target virtual image and a target expression feature corresponding to the emotion category from a configured comparison table of different emotion categories, different virtual images and different expression features; generating a target virtual image of the AI digital human according to the target virtual image and the target expression feature; outputting response content by using the target virtual image voice; and the business processing layer is used for matching response content corresponding to the query content from the configured customer service knowledge base after receiving the query request. According to the application, the AI digital human image can be combined to realize multi-terminal efficient interaction intelligent customer service.

Description

Technical Field

[0001] This application relates to the field of intelligent customer service, and specifically, to a customer service platform, service method, equipment and medium based on AI digital people. Background Art

[0002] Existing intelligent customer service technology has the following defects:

[0003] 1. Rule-Reliant: Traditional systems rely on pre-set rules to evaluate responses and are unable to understand complex statements. For example, a candidate might describe their experience solving a complex problem in detail, but the system might misjudge the candidate because the description doesn't match the template, resulting in an accuracy rate of only 30%-40%.

[0004] 2. Keyword matching: The system only mechanically matches keywords, ignoring semantic differences. For example, in the answer to "How to manage team conflict," the system cannot distinguish between "prevent" and "resolve," resulting in a matching success rate of approximately 50%.

[0005] 3. Data limitations: Simple machine learning models struggle to cover complex scenarios due to insufficient training data. For example, when answering the question "How to deal with cross-cultural communication barriers," the system might only have a 60% accuracy rate due to a lack of relevant data.

[0006] 4. Multi-terminal adaptation: The interfaces and interactions of different terminals (PC, mobile, etc.) are inconsistent. Candidates need to adapt again when switching terminals, which affects the experience.

[0007] 5. Delayed knowledge base updates: The knowledge base relies on manual updates, which is inefficient and error-prone. For example, the new question "How to manage a remote team" could not be added to the system in a timely manner, resulting in a lag in evaluation standards.

[0008] The above defects indicate that the existing technology needs to be further optimized to improve the accuracy of evaluation and user experience. Summary of the Invention

[0009] The purpose of the embodiments of the present application is to provide a customer service platform, service method, equipment and medium based on AI digital humans, so as to solve the above-mentioned problems existing in the prior art, and to realize intelligent customer service with efficient multi-terminal interaction by combining the image of AI digital humans.

[0010] In a first aspect, the present invention provides a customer service platform based on AI digital humans, the platform comprising:

[0011] The front-end display layer is configured to generate a chat interface for interaction between the user and the AI digital human in response to a user clicking on an interface generation control; wherein the chat interface includes an input box; in response to the user entering query content through the input box, generate a query request containing the query content; and send the query request to the digital human customer service interaction layer;

[0012] The digital human customer service interaction layer is configured to respond to the query request, perform sentiment analysis on the query content, and obtain the sentiment category corresponding to the query content; send the query request and the sentiment category to the business processing layer; and match the target virtual image and target expression feature corresponding to the sentiment category from a configured comparison table of different sentiment categories, different virtual images, and different expression features; generate a target virtual image of the AI digital human based on the target virtual image and the target expression feature; and, upon receiving the response content corresponding to the query request fed back by the business processing layer, output the response content using the voice of the target virtual image;

[0013] The business processing layer is used to match the response content corresponding to the query content from the configured customer service knowledge base after receiving the query request.

[0014] In an optional embodiment, the chat interface further includes: a first display area, a second display area, a third display area, a fourth display area, and a fifth display area;

[0015] The first display area is used to display the target virtual image;

[0016] The second display area is used to display a pre-configured first prompt information in a configured display mode; the first prompt information is used to prompt the user that an AI digital human is being assigned to the user;

[0017] The third display area is used to display the pre-configured second prompt information in accordance with the configured display mode; the first prompt information is used to prompt the user to enter the query content through the input box;

[0018] The fourth display area is used to display the target allocation status according to the configured display mode; wherein the target allocation status is used to represent the allocation status of the AI digital human assigned to the user;

[0019] The fifth display area is used to display the current interaction status between the user and the AI digital human.

[0020] In an optional embodiment, the input box includes: a text input box and a voice input box;

[0021] The digital human customer service interaction layer is also used to:

[0022] Determining an input form of the query content based on an input box into which the user inputs the query content;

[0023] Match the target processing method corresponding to the input form from the comparison table of different input forms and different processing methods configured;

[0024] The target processing method is used to perform sentiment analysis on the query content to obtain corresponding sentiment categories.

[0025] In an optional implementation manner, the business processing layer is specifically configured to:

[0026] Based on the configured first matching mode, matching the first response content corresponding to the query content from the configured customer service knowledge base;

[0027] If the first response content corresponding to the query content is not matched, then based on the configured second matching mode, the second response content corresponding to the query content is matched from the configured customer service knowledge base;

[0028] If the second response content corresponding to the query content is not matched, matching the third response content corresponding to the query content from the AI model database based on the configured third matching mode;

[0029] If the third response content corresponding to the query content is not matched, the configured third prompt information is output.

[0030] In an optional embodiment, the first matching mode is a complete matching mode;

[0031] The second matching mode is a partial matching mode;

[0032] The third matching mode is a semantic-based vector matching mode.

[0033] In an optional embodiment, the platform further includes: a data storage layer; the data storage layer is used to store the customer service knowledge base and historical interaction data of each user;

[0034] The digital human customer service interaction layer is also used to:

[0035] When receiving query requests generated based on the query content input by each user, the current load of each AI digital human and the total load of the platform, as well as the historical interaction data of each user, are obtained;

[0036] A corresponding AI digital human is assigned to each user based on the current load of each AI digital human, the total platform load and the historical interaction data of each user.

[0037] In an optional embodiment, the platform includes multiple digital human customer service interaction layers and multiple business processing layers;

[0038] The platform also includes: a service distribution layer;

[0039] The service allocation layer is used to obtain the business platform where any user is located, and match the target digital human customer service interaction layer and target business processing layer corresponding to the business platform from the comparison table of different business platforms, different digital human customer service interaction layers and different business processing layers configured;

[0040] Control the target digital human customer service interaction layer and the target business processing layer to process the user's query request.

[0041] In a second aspect, the present invention provides a customer service method based on AI digital human, the method comprising:

[0042] Controlling the front-end display layer to generate a chat interface for user interaction with the AI digital human in response to a user clicking on an interface generation control; wherein the chat interface includes an input box; generating a query request containing the query content in response to the user entering the query content through the input box; and sending the query request to the digital human customer service interaction layer;

[0043] Controlling the digital human customer service interaction layer to respond to the query request, perform sentiment analysis on the query content, and obtain the sentiment category corresponding to the query content; sending the query request and the sentiment category to the business processing layer; and matching the target virtual image and target expression feature corresponding to the sentiment category from a configured comparison table of different sentiment categories, different virtual images, and different expression features; and generating a target virtual image of the AI digital human based on the target virtual image and the target expression feature.

[0044] When the control business processing layer receives the query request, it matches the response content corresponding to the query content from the configured customer service knowledge base; and sends the response content to the digital human customer service interaction layer;

[0045] The digital human customer service interaction layer is controlled to use the target virtual image voice to output the response content corresponding to the query request fed back by the business processing layer.

[0046] In a third aspect, the present invention provides an electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0047] Memory for storing computer programs;

[0048] The processor is configured to implement the method described in the aforementioned embodiment when executing the program stored in the memory.

[0049] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the aforementioned embodiment is implemented.

[0050] This application uses a dynamic AI digital human virtual image to interact with users to provide customer service, which is not only more vivid and more human, but also can improve the efficiency of interaction and help improve the user's customer service experience.

[0051] This application generates a chat interface for users to interact with AI digital humans, and combines sentiment analysis technology to select appropriate avatars and expression characteristics. This can provide a more humane service experience that is closer to real communication, helping to enhance user participation and satisfaction. The business processing layer of this application can quickly and accurately search and return answers to query requests from a preset knowledge base, greatly reducing waiting time and improving service efficiency. At the same time, the target avatar outputs the response content in voice form, making information transmission more intuitive and effective.

[0052] This application can make communication warmer and more friendly by performing emotion recognition on the information input by the user and adjusting the performance of the AI digital person accordingly (such as selecting specific expressions or tones), thereby deepening the emotional connection with customers and promoting brand image building.

[0053] Compared to traditional manual customer service teams, this application utilizes an automated intelligent customer service solution that significantly reduces manpower input while maintaining service quality, lowering the company's long-term operating costs. Furthermore, it offers 24 / 7 uninterrupted service, meeting customer consultation needs at different times.

[0054] This application supports the configuration of multiple types of emotion categories, virtual images, and corresponding expression characteristics based on different scenarios. This provides merchants with highly customizable customer service solution options, enabling them to adjust the appearance and behavior patterns of AI digital humans based on their own brand characteristics or the requirements of specific marketing campaigns.

[0055] The user feedback and interaction data collected by this application can help companies better understand changing market demand trends and continuously improve product features and service processes. By deeply mining this valuable information, potential problems or opportunities can be discovered, driving continuous innovation and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 An architectural diagram of a customer service platform based on AI digital humans provided in an embodiment of the present application;

[0058] Figure 2 A flowchart of a customer service method based on AI digital humans provided in an embodiment of the present application;

[0059] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0061] The customer service method based on AI digital human provided in the embodiment of this application is applied in Figure 1 In the customer service platform based on AI digital humans shown in the figure, the customer service platform based on AI digital humans can be installed in a server or in a terminal with strong computing power. The server can be a physical server or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), as well as basic cloud computing services such as big data and artificial intelligence platforms. The terminal can be a user equipment (UE) such as a mobile phone, smart phone, laptop computer, digital broadcast receiver, personal digital assistant (PDA), tablet computer (PAD), handheld device, vehicle-mounted device, wearable device, computing device or other processing device connected to a wireless modem, mobile station (MS), mobile terminal, etc. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, which is not limited in this application.

[0062] like Figure 1 As shown, the customer service platform based on AI digital human adopted in the embodiment of the present application includes:

[0063] The front-end display layer 110 is used to respond to the user's operation of clicking the interface generation control to generate a chat interface for the user to interact with the AI digital human; respond to the user entering query content through the input box to generate a query request containing the query content; and send the query request to the digital human customer service interaction layer;

[0064] The service allocation layer 120 is used to obtain the business platform where any user is located, and match the target digital human customer service interaction layer and target business processing layer corresponding to the business platform from the comparison table of different configured business platforms, different digital human customer service interaction layers and different business processing layers; control the target digital human customer service interaction layer and the target business processing layer to process the user's query request.

[0065] The Digital Human Customer Service Interaction Layer 130 (i.e., the target Digital Human Customer Service Interaction Layer) is configured to respond to received query requests, perform sentiment analysis on the query content, and obtain the sentiment category corresponding to the query content; send the query request and sentiment category to the business processing layer; and match the target virtual image and target expression characteristics corresponding to the sentiment category from a configured comparison table of different sentiment categories, different virtual images, and different expression characteristics; generate the target virtual image of the AI Digital Human based on the target virtual image and target expression characteristics; and, upon receiving the response content corresponding to the query request fed back by the business processing layer, output the response content using the voice of the target virtual image;

[0066] The business processing layer 140 (i.e., the target business processing layer) is used to match the query content with the corresponding response content from the configured customer service knowledge base after receiving the query request;

[0067] The data storage layer 150 is used to store the customer service knowledge base and historical interaction data of each user.

[0068] In this embodiment of the present application, the front-end display layer 110 is integrated with multiple business platforms, including the web, WeChat mini-programs, and apps. The business platform integrated with the front-end display layer is the user's business platform. The front-end display layer is used to display the user's interactive dialogue with the AI digital human. It receives user input through an input box and displays the AI digital human's response.

[0069] In the embodiment of the present application, the front-end display layer 110 adopts responsive design technology to automatically adjust the interface layout according to the screen size and resolution of the user's terminal; assuming that the terminal screen width is w, the height is h, and the interface element set is E = {e1, e2, ... e x}, x is a positive integer; each interface element ei The layout attributes (such as position, size) are attr(e i ), i∈{1,2,…,x}.

[0070] The layout rule set corresponding to different screen size ranges is predefined as L = {l1,l2,…l m}, each layout rule l j Contains the adjustment strategy for the layout attributes of interface elements; j∈{1,2,…,m}, m is a positive integer;

[0071] For example, on mobile phones (w≤600), a single-column layout is used, and the width of multiple interface elements is set to a certain proportion of the screen width; on PCs (w greater than 600), a multi-column layout is used, and the element width is reasonably distributed according to the screen width.

[0072] The layout adjustment algorithm is:

[0073]

[0074] Among them, range(l j ) represents the layout rule l j Applicable screen size range, adjust(e i ,l j ) indicates according to layout rules.

[0075] In an embodiment of the present application, the chat interface includes: an input box, a first display area, a second display area, a third display area, a fourth display area, and a fifth display area;

[0076] Specifically, the input box includes: a text input box and a voice input box;

[0077] The first display area is used to display the target virtual image; the target virtual image can be static or dynamic; when the response content corresponding to the query content is being generated, the static target virtual image can be displayed; after the response content is generated, the dynamic target virtual image is used to output the response content;

[0078] The second display area is used to display a pre-configured first prompt message in accordance with the configured display mode; the first prompt message is used to inform the user that an AI digital human is being assigned to the user; the first prompt message is a waiting prompt, which can be randomly matched by the platform from a pre-built waiting prompt library, or can be uniformly set by the platform, or generated based on the user's historical preferences; for example, it can be "Hello, a dedicated AI digital human customer service representative is being assigned to you. Please wait a moment~";

[0079] The third display area is used to display the pre-configured second prompt information in accordance with the configured display mode; the first prompt information is used to prompt the user to input the query content through the input box; the second prompt information is an input prompt, which can be randomly matched by the platform from a pre-built waiting prompt library, or can be uniformly set by the platform, or generated according to the user's historical preferences; for example, it can be "You can enter the question content in advance, and it can be sent immediately after being assigned to a digital person"; the second prompt information prompts users to input first to improve interaction efficiency.

[0080] The fourth display area is used to display the target allocation status according to the configured display mode; the target allocation status is used to represent the allocation status of the AI digital human to the user; the target allocation status includes being allocated, waiting for allocation, and allocated;

[0081] The fifth display area is used to display the current interaction status between the user and the AI digital human.

[0082] In one embodiment of the present application, the display areas of the static target virtual image and the dynamic target virtual image may be the same or different; the first display area may include a first sub-area and a second sub-area; the first sub-area is located in the right area of the chat interface; the second sub-area is located in the middle area of the chat interface; when displaying a static target virtual image, the first sub-area is used for display; when displaying a dynamic target virtual image, the second sub-area is used for display.

[0083] In the embodiment of the present application, the display mode can be static text display, dynamic text flashing display or other configuration display mode; static text display can also set the text color, line, font and size and other display characteristics.

[0084] In an embodiment of the present application, when the user is waiting for access, the front-end display layer 110 abandons the traditional exclusive waiting interface and directly enters the chat interface, displays the waiting prompt information (i.e., the first prompt information) and the dynamic allocation status (i.e., the second prompt information), and supports the user to enter the question content in advance.

[0085] In an embodiment of the present application, different digital human customer service interaction layers and business processing layers are set up for different business platforms, so that the same user or different users can be served on different business platforms at the same time without affecting the customer service speed of different business platforms; for example, users can generate corresponding chat interfaces on both WeChat mini-programs and APP business platforms at the same time, and the platform calls different digital human customer service interaction layers and business processing layers to process query requests from different business platforms.

[0086] In an embodiment of the present application, a dynamic routing gateway is used to assign a corresponding target digital human customer service interaction layer and the target business processing layer to the user to control the corresponding target digital human customer service interaction layer and the target business processing layer to process the query request of the corresponding user.

[0087] In the embodiment of the present application, the digital human customer service interaction layer 130 is used to:

[0088] When receiving query requests generated based on the query content input by each user, the current load of each AI digital human and the total platform load, as well as the historical interaction data of each user, are obtained; based on the current load of each AI digital human, the total platform load and the historical interaction data of each user, a corresponding AI digital human is assigned to each user; based on the input box where the user inputs the query content, the input form of the query content is determined; from the comparison table of different configured input forms and different processing methods, the target processing method corresponding to the input form is matched; and the target processing method is used to perform sentiment analysis on the query content to obtain the corresponding sentiment category.

[0089] In the embodiment of the present application, the method for calculating the current load of each AI digital human includes:

[0090] For any AI digital human, obtain the value of the load assessment indicator corresponding to the AI digital human and the weight of each configured load assessment indicator; among which, the load assessment indicators include: the number of users, the estimated remaining service time of each user, and the CPU usage rate; perform a weighted sum of the values of each load assessment indicator to obtain the current load of the AI digital human.

[0091] In an embodiment of the present application, the estimated remaining service time of each user is predicted using a time prediction model based on the current processing status of the user's query request and the user's historical interaction data.

[0092] In the embodiment of the present application, the total platform load is obtained by adding up the current loads of all AI digital humans.

[0093] In an embodiment of the present application, the user's historical interaction data includes: the frequency of interaction between the user and each AI digital human, the interaction time of each interaction, and feedback data for each interactive user (including satisfaction score, etc.).

[0094] In the embodiment of the present application, a dynamic load balancing algorithm is used to allocate AI digital humans to users; specifically, the following steps are included:

[0095] For any user, the user's historical interaction data is analyzed to obtain the user's personal preferences for AI digital humans; each AI digital human is sorted in ascending order according to its current load; the first n AI digital humans after sorting are used as alternative AI digital humans; n is a positive integer, and the value of n can be customized according to actual needs; based on the user's personal preferences for AI digital humans, the user's intimacy score for each alternative AI digital human is calculated; the current load of each alternative AI digital human, the total platform load and the user's intimacy score for each alternative available AI digital human are input into the pre-built AI digital human matching model to obtain the target AI digital human corresponding to the user.

[0096] In an embodiment of the present application, the AI digital human matching model adopts a machine learning-based model or a reinforcement learning model; a machine learning-based model is a variant of a multi-armed bandit algorithm (e.g., the UCB algorithm); the AI digital human matching model takes the current load of each candidate AI digital human, the total platform load, and the user's intimacy score for each available alternative AI digital human as input. The lower the current load, the higher the reward weight given; the higher the user's intimacy score for the digital human, the higher the reward weight given; if it is a reinforcement learning model, the objective function is to minimize user waiting time and maximize user satisfaction.

[0097] In an embodiment of the present application, a user's personal preferences for AI digital humans are determined based on the frequency of user interactions with each AI digital human, the interaction time of each interaction, and feedback data for each user interaction (including satisfaction scores, etc.), using natural language processing technology or a preference analysis model based on machine learning.

[0098] In the embodiment of the present application, the user's intimacy score for each candidate AI digital human is calculated based on the user's personal preference for the AI digital human, including:

[0099] The result is obtained by weighted summing the frequency of user interaction with each AI digital human, the interaction time of each interaction, and the feedback data of each user interaction according to the configured frequency weight, time weight, and feedback weight;

[0100] Alternatively, it can be obtained based on the user's historical interaction data and the intimacy score of other similar users for each AI digital person; similar users are users with similar personal preferences for AI digital people as the user; the intimacy score of other similar users for each AI digital person can be obtained based on the feedback data of the corresponding users after the interaction.

[0101] This application uses the above features to prioritize requests to AI digital humans with lower load and higher user intimacy, reducing user wait time. At the same time, it monitors the load changes of the digital human customer service in real time and dynamically adjusts the load weight to ensure the rationality of the allocation.

[0102] In an embodiment of the present application, the customer service platform is equipped with multiple AI digital humans, and each AI digital human corresponds to a different business scenario; for example, in the customer service scenario of the trading platform, different AI digital humans can be set up for pre-sales scenarios, sales scenarios, and after-sales scenarios, respectively, so as to serve users in different business scenarios.

[0103] In other embodiments of the present application, further refinement can be performed under business scenarios, so that users in the same business scenario correspond to different AI digital humans due to different specific business parameters or historical interaction data, so as to improve the user experience.

[0104] In an embodiment of the present application, a text input box and a voice input box are provided for users to input query content in the form of text input or voice input; at the same time, different processing methods are set for query content in different input forms.

[0105] In the embodiment of the present application, when the query content is input in the form of speech, a deep learning-based speech recognition model (such as a Transformer-based ASR model) is used to process the query content; specifically, the following steps are performed:

[0106] A deep learning-based speech recognition model is used to extract features from the query content voice signal. The extracted features include MFCC and Mel-Spectrogram. The speech features are converted into text sequences. The text sequences are recognized using a pre-trained emotion recognition model to obtain the corresponding emotion categories.

[0107] In an embodiment of the present application, when the query content is input in the form of text, natural language processing technology is used to extract features of the query content to obtain a text sequence; and a pre-trained emotion recognition model is used to recognize the text sequence to obtain the corresponding emotion category.

[0108] In an embodiment of the present application, in order to improve recognition accuracy, a speech recognition model based on deep learning is trained on a large amount of speech data to learn the mapping relationship between speech and text.

[0109] In an embodiment of the present application, the emotion recognition model is an SVM (support vector machine) based on machine learning or an LSTM (long short-term memory network) model based on deep learning; emotion categories include positive, neutral and negative.

[0110] In the embodiments of the present application, the virtual images of AI digital humans include: anthropomorphic images, animal images, and symbolic images, etc.; expression features include facial expressions, tone, timbre, speaking speed, and intonation; facial expressions include: happiness, anger, and sadness, etc.; for example, when the user is in a negative mood, an animal image is generated, and a happy facial expression and cheerful tone are used for output.

[0111] In the embodiment of the present application, the business processing layer 140 is specifically used to:

[0112] Based on the configured first matching mode, the first response content corresponding to the query content is matched from the configured customer service knowledge base; if the first response content corresponding to the query content is not matched, based on the configured second matching mode, the second response content corresponding to the query content is matched from the configured customer service knowledge base; if the second response content corresponding to the query content is not matched, based on the configured third matching mode, the third response content corresponding to the query content is matched from the AI model database; if the third response content corresponding to the query content is not matched, the configured third prompt information is output.

[0113] In the embodiment of the present application, the first matching mode is a complete matching mode; the second matching mode is a partial matching mode; and the third matching mode is a semantic-based vector matching mode.

[0114] In an embodiment of the present application, the first matching mode is an exact match, and a hash table data structure is used to store standard questions; based on each standard question in the customer service knowledge base, a set of standard questions is generated, a hash function is constructed, and each standard question is mapped to a different position in the hash table; for the query content Q input by the user, the hash value is preprocessed, and the hash table is searched for a matching standard question based on the calculated hash value; if there is, the corresponding response content is directly returned; if not, it is returned as not present to prompt entering the second matching mode.

[0115] In an embodiment of the present application, the customer service knowledge base also includes: a first vector library and a second vector library; wherein the first vector library is used to store the knowledge vectors of each knowledge item contained in the customer service knowledge base; and the second vector library is used to store the knowledge encoding vectors of each knowledge item encoded using a large model.

[0116] In an embodiment of the present application, based on the configured second matching mode, matching the second response content corresponding to the query content from the configured customer service knowledge base includes:

[0117] Use a pre-trained word vector model (such as Word2Vec, GloVe, etc.) to convert the query content into a query word vector; calculate the cosine similarity between the query word vector and the knowledge vector of each knowledge item in the customer service knowledge base; if the calculated cosine similarity is greater than the configured first similarity threshold, adjust the response content corresponding to the knowledge item corresponding to the corresponding knowledge vector to obtain a second response content.

[0118] In an embodiment of the present application, based on the configured third matching mode, matching the third response content corresponding to the query content from the AI model database includes:

[0119] Encoding the query content using the large model to obtain a first encoding vector; receiving a plurality of first knowledge encoding vectors and corresponding knowledge base questions retrieved by the vector retrieval library;

[0120] Calculate the similarity between each knowledge base question and the query content. If the calculated similarity is greater than the configured person similarity threshold, obtain the corresponding answer content of the knowledge base question;

[0121] The query content is input into the large model to obtain the probability distribution p(b) output by the large model based on its pre-trained language generation ability and learning from a large amount of text data, where b represents a word in the vocabulary; the answer text is sampled from the probability distribution p(b); to control the quality and diversity of the generated text, some strategies can be adopted, such as setting a temperature parameter (Temperature) to adjust the smoothness of the probability distribution. A higher temperature value will make the generated text more diverse, but may reduce the accuracy; the generated answer text is verified and adjusted in combination with relevant knowledge in the customer service knowledge base; and the third answer content is generated based on the adjusted text and the answer content corresponding to the corresponding knowledge base question obtained.

[0122] In an embodiment of the present application, the generated answer text is verified and adjusted in combination with the relevant knowledge in the customer service knowledge base, including: checking whether the generated answer text contains key information in the knowledge base. If not, it is regenerated or supplemented to ensure that the generated answer is both creative and accurate and reliable.

[0123] This application adopts the Faiss vector retrieval library and uses clustering and quantization technology to build an index to improve retrieval efficiency; the similarity between each knowledge encoding vector and the first encoding vector can be cosine similarity or other distance measurement.

[0124] In the embodiment of the present application, the business processing layer 140 is further used to: manage the customer service knowledge base; the management of the customer service knowledge base includes adding, renaming, and deleting operations on the knowledge base, document knowledge and question and answer knowledge management, and user data management;

[0125] Specifically include:

[0126] Add a new knowledge base: Select the default knowledge base on the relevant interface, click the "Add" button, and a text input box will pop up. Enter the knowledge base name (up to 50 characters) and press Enter to take effect. The system will call the verification interface to ensure that there is no knowledge base with the same name in the same directory.

[0127] Rename a knowledge base: Click the three dots to the right of a knowledge base and select "Rename." The knowledge base name becomes editable. Enter a new name (up to 50 characters) and press Enter to take effect. The system will then check if a knowledge base with the same name already exists.

[0128] Deleting a knowledge base: The system first verifies whether all documents associated with the current knowledge base are in the "Effective Status". If so, they cannot be deleted. If there are no valid documents, a background logical deletion is performed.

[0129] Document list display: Displays data in reverse order of update time, supports filtering by knowledge base and keyword search. The list is paginated, with 10 rows of data per page.

[0130] New Documents: The "Document Name" and "Document Content" fields on the pop-up page are required. A rich text editor is used, supporting text input but not non-text content. The document name is limited to 50 characters, and the document content is limited to 20,000 characters. The system will display a warning if the limit is exceeded.

[0131] Document status management: The "Save Draft" button puts the document into the "Unlearned" state; the "Save and Synchronize Model" button puts the document into the "Learning" state and synchronizes it to the Dify framework.

[0132] Other functions: support editing and deleting documents, viewing document details and change logs, handling learning failures, etc.

[0133] In the embodiment of the present application, the management of the knowledge base also includes:

[0134] Addition and renaming verification algorithm: When adding or renaming a knowledge base, the system checks the uniqueness of the name; if the same name exists, the user is prompted to re-enter it; if the name is unique, the corresponding addition or renaming operation is performed;

[0135] When deleting a knowledge base, first check the status of its associated documents; based on the status of the knowledge base's associated documents, determine whether the corresponding knowledge base meets the configured deletion conditions. If so, perform a logical deletion operation, such as marking the knowledge base as deleted in the database; if there are documents in a valid state, the user will be prompted that the deletion cannot be made.

[0136] In the embodiment of the present application, document knowledge management also includes:

[0137] The document list is displayed in reverse order of update time, and the sorting algorithm can use efficient sorting algorithms such as quick sort or merge sort;

[0138] Supports filtering by knowledge base and keyword searching. When adding or editing documents, word count limits and formatting checks are applied to document names and content. Formatting checks can use techniques such as regular expressions to check whether the document content conforms to the specified format. When saving a draft, the document status is set to "unlearned." When saving and synchronizing a model, the document status is set to "learning," and the corresponding interface is called to synchronize the document content to the Dify framework for processing.

[0139] In an embodiment of the present application, the customer service knowledge base of the present application can not only store corresponding knowledge items, but also store corresponding customer service knowledge in the form of documents.

[0140] In the embodiment of the present application, question-answering knowledge management includes:

[0141] Question and answer list display: sorted in reverse order of update time, supporting status filtering and keyword search. The list is displayed in pages, with 20 data items per page;

[0142] New Q&A: Fill in the standard question (required, up to 140 characters), similar questions (optional, up to 140 characters, maximum 30), status, effective date (optional), and answer (required, up to 20,000 characters). The system detects questions with the same name and has a mechanism to verify required fields.

[0143] Other functions: support editing Q&A, viewing Q&A details, batch importing Q&A (supports XLSX and XLS formats, with a complete validation mechanism), and deleting Q&A, etc.

[0144] In the embodiment of the present application, question-answering knowledge management also includes:

[0145] The Q&A list is sorted in reverse order of update time, with status filtering and keyword search; it can be displayed in pages;

[0146] Verify the content of standard questions, similar questions, answers, etc.; at the same time, check whether the required fields are filled in, such as standard questions and answers cannot be empty; it is also necessary to detect questions with the same name, traverse the existing question and answer sets, and check whether there are the same standard questions or similar questions; if there are questions with the same name, prompt the user to modify the question; when importing questions and answers, first check whether the file format is XLSX or XLS; then, check the file content format according to the pre-defined template structure; finally, perform length verification on the imported standard questions, similar questions and answers, using the same length verification algorithm when adding new questions and answers.

[0147] In the embodiment of the present application, the management of user data includes functions such as viewing the user list, adding new users, editing users, and resetting passwords;

[0148] The user list page displays the user data in the current system, including user ID, user account, user name, user role, status, and creation time. The list is sorted in descending order by creation time and supports paging and keyword search.

[0149] When adding a new user, enter the user name (up to 10 Chinese characters), user account (up to 50 characters), user password (up to 50 characters), select the user role and status. The user ID is generated by the system and automatically increments from 1.

[0150] When editing a user, the original user information will be displayed. Click the "OK" button after the modification to take effect. After resetting the password, the user must use the new password (the default is "XY888888") to log in.

[0151] In the personal center, users can modify their passwords. After modification, the system will automatically log out of the current login state and require the user to log in again.

[0152] In this embodiment of the present application, the customer service knowledge base uses an inverted index-based retrieval structure. Let the knowledge base be K, which contains multiple knowledge items, each of which consists of a question and an answer (i.e., the response content). For the query content entered by the user, preprocessing is first performed, including operations such as word segmentation, part-of-speech tagging, and stop word filtering, to obtain the processed query content.

[0153] During retrieval, the document frequency (DF) of each word in the processed query content in the knowledge base is calculated;

[0154] For each knowledge item, the classic TF-IDF algorithm is used to calculate its relevance score with the processed query content;

[0155] The knowledge items are sorted by relevance score and the knowledge item with the highest score is selected as the matching result. If the score exceeds the set threshold, the answer is returned; otherwise, the process enters the large model processing stage.

[0156] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application can be combined with each other if there is no conflict.

[0157] Figure 2 This is a flowchart of a customer service method based on AI digital human provided in the embodiment of this application. Figure 2 As shown, the method may include:

[0158] Step S210: Control the front-end display layer to respond to the user clicking on the interface generation control operation, generate a chat interface for the user and the AI digital human to interact; respond to the user entering query content through the input box, generate a query request containing the query content; and send the query request to the digital human customer service interaction layer.

[0159] Step S220: Control the digital human customer service interaction layer to respond to the received query request, perform sentiment analysis on the query content, and obtain the sentiment category corresponding to the query content; and send the query request and the sentiment category to the business processing layer.

[0160] Step S230: Match the target virtual image and target expression feature corresponding to the emotional category from the configured comparison table of different emotional categories, different virtual images, and different expression features; generate the target virtual image of the AI digital human based on the target virtual image and target expression feature.

[0161] Step S240: After receiving the query request, the control business processing layer matches the response content corresponding to the query content from the configured customer service knowledge base; and sends the response content to the digital human customer service interaction layer.

[0162] Step S250: Control the digital human customer service interaction layer to use the target virtual image voice to output the response content corresponding to the query request fed back by the business processing layer.

[0163] In an embodiment of the present application, the platform administrator can automatically update the customer service knowledge base, or update the customer service knowledge base on a regular basis; it can also be set that when any platform administrator uploads a corresponding document to a designated database, the customer service knowledge base automatically stores the document in the customer service knowledge base, extracts the data therein, generates knowledge entries and / or question and answer knowledge, and updates the question and answer list and the customer service knowledge base.

[0164] In this embodiment of the present application, after a user enters a query, the platform immediately starts a timer and displays a transitional text message "Helping you query..." within the timer duration (e.g., 1 second). Simultaneously, the query entered by the user is sent to the Digital Human customer service interaction layer and the business processing layer for processing. Once the processing is complete and an answer is received, the timer is stopped and the response is displayed. If no answer is received before the timer expires, the transitional text message continues to display until the response is returned.

[0165] In an embodiment of the present application, the method further includes generating a system state transition diagram, which is used to illustrate the platform's state changes under different operations and event triggers. For example, in a digital human customer service session, the initial state is "Waiting for Access." After the stream is pulled, the state changes to "Interactive." After the user sends a query, the state enters "Problem Handling." Depending on the processing results, the system may enter the "Answer Output" or "Exception Handling" states. If the user does not operate for an extended period, the system will enter the "Prompt User Input" and "Close Session" states, respectively.

[0166] In an embodiment of the present application, the method further includes generating a data flow diagram, which is used to illustrate the flow paths and interactions of various data types within the platform. The query content entered by the user is transmitted via the front-end to the Digital Human customer service interaction layer, and then to the business processing layer. The business processing layer first searches the knowledge base for the answer, and if no answer is found, it calls the large model. The answer data is returned to the user along the reverse path. The data flow for knowledge base management and user management operations is also clearly displayed in the diagram.

[0167] In an embodiment of the present application, the method further includes generating an interface call sequence diagram, which is used to demonstrate the order and timing of interface calls between the various layers of the platform. After a user submits a query, the front-end invokes the send query interface. The data is then distributed through the dynamic routing gateway of the business processing layer to the four-level decision engine, which in turn invokes the knowledge base query interface and the large model interface, ultimately returning the answer to the front-end for display.

[0168] The present application also provides an electronic device, such as Figure 3 As shown, it includes a processor 310 , a communication interface 320 , a memory 330 and a communication bus 340 , wherein the processor 310 , the communication interface 320 , and the memory 330 communicate with each other via the communication bus 340 .

[0169] Memory 330, for storing computer programs;

[0170] The processor 310 is configured to execute the program stored in the memory 330 by performing the following steps:

[0171] Control the front-end display layer to respond to a user clicking on an interface generation control, generating a chat interface for interaction between the user and the AI digital human; wherein the chat interface includes an input box; in response to the user entering query content through the input box, generate a query request containing the query content; and send the query request to the digital human customer service interaction layer;

[0172] Controlling the digital human customer service interaction layer to respond to received query requests, perform sentiment analysis on the query content, and obtain the corresponding sentiment category of the query content; sending the query request and sentiment category to the business processing layer; and matching the target sentiment category with the target sentiment category from a configured comparison table of different sentiment categories, different avatars, and different expression characteristics; and generating the target avatar of the AI digital human based on the target avatar and target expression characteristics.

[0173] When the control business processing layer receives a query request, it matches the query content with the corresponding response content from the configured customer service knowledge base and sends the response content to the digital human customer service interaction layer;

[0174] The digital human customer service interaction layer is controlled to use the target virtual image voice to output the response content corresponding to the query request received from the business processing layer.

[0175] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0176] The communication interface is used for communication between the above electronic device and other devices.

[0177] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0178] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0179] The implementation methods and beneficial effects of the various components of the electronic device in the above embodiments to solve the problems can be found in Figure 2 The various steps in the embodiment shown are implemented, therefore, the specific working process and beneficial effects of the electronic device provided by the embodiment of the present application are not repeated here.

[0180] In another embodiment provided in the present application, a computer-readable storage medium is also provided, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the customer service method based on AI digital human described in any of the above embodiments.

[0181] In another embodiment provided in the present application, a computer program product containing instructions is also provided. When the computer program product is run on a computer, it enables the computer to execute the customer service method based on AI digital human described in any of the above embodiments.

[0182] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the embodiments of the present application can be implemented in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the embodiments of the present application can be implemented in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0183] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0184] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0185] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0186] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0187] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims and their equivalents, the embodiments of the present application are also intended to include these modifications and variations.

Claims

1. A customer service platform based on AI digital human, characterized by: The platform includes: The front-end display layer is configured to generate a chat interface for interaction between the user and the AI digital human in response to a user clicking on an interface generation control; wherein the chat interface includes an input box; in response to the user entering query content through the input box, generate a query request containing the query content; and send the query request to the digital human customer service interaction layer; The digital human customer service interaction layer is configured to respond to the query request, perform sentiment analysis on the query content, and obtain the sentiment category corresponding to the query content; send the query request and the sentiment category to the business processing layer; and match the target virtual image and target expression feature corresponding to the sentiment category from a configured comparison table of different sentiment categories, different virtual images, and different expression features; generate a target virtual image of the AI digital human based on the target virtual image and the target expression feature; and, upon receiving the response content corresponding to the query request fed back by the business processing layer, output the response content using the voice of the target virtual image; The business processing layer is used to match the response content corresponding to the query content from the configured customer service knowledge base after receiving the query request.

2. The platform according to claim 1, wherein The chat interface further includes: a first display area, a second display area, a third display area, a fourth display area, and a fifth display area; The first display area is used to display the target virtual image; The second display area is used to display a pre-configured first prompt information in a configured display mode; the first prompt information is used to prompt the user that an AI digital human is being assigned to the user; The third display area is used to display the pre-configured second prompt information in accordance with the configured display mode; the first prompt information is used to prompt the user to enter the query content through the input box; The fourth display area is used to display the target allocation status according to the configured display mode; wherein the target allocation status is used to represent the allocation status of the AI digital human assigned to the user; The fifth display area is used to display the current interaction status between the user and the AI digital human.

3. The platform according to claim 1, wherein The input box includes: a text input box and a voice input box; The digital human customer service interaction layer is also used to: Determining an input form of the query content based on an input box into which the user inputs the query content; Match the target processing method corresponding to the input form from the comparison table of different input forms and different processing methods configured; The target processing method is used to perform sentiment analysis on the query content to obtain corresponding sentiment categories.

4. The platform according to claim 1, wherein The business processing layer is specifically used to: Based on the configured first matching mode, matching the first response content corresponding to the query content from the configured customer service knowledge base; If the first response content corresponding to the query content is not matched, then based on the configured second matching mode, the second response content corresponding to the query content is matched from the configured customer service knowledge base; If the second response content corresponding to the query content is not matched, matching the third response content corresponding to the query content from the AI model database based on the configured third matching mode; If the third response content corresponding to the query content is not matched, the configured third prompt information is output.

5. The platform according to claim 4, wherein: The first matching mode is a complete matching mode; The second matching mode is a partial matching mode; The third matching mode is a semantic-based vector matching mode.

6. The platform according to claim 1, wherein The platform also includes: a data storage layer; the data storage layer is used to store the customer service knowledge base and historical interaction data of each user; The digital human customer service interaction layer is also used to: When receiving query requests generated based on the query content input by each user, the current load of each AI digital human and the total load of the platform, as well as the historical interaction data of each user, are obtained; A corresponding AI digital human is assigned to each user based on the current load of each AI digital human, the total platform load and the historical interaction data of each user.

7. The platform according to claim 1, wherein The platform includes multiple digital human customer service interaction layers and multiple business processing layers; The platform also includes: a service distribution layer; The service allocation layer is used to obtain the business platform where any user is located, and match the target digital human customer service interaction layer and target business processing layer corresponding to the business platform from the comparison table of different business platforms, different digital human customer service interaction layers and different business processing layers configured; Control the target digital human customer service interaction layer and the target business processing layer to process the user's query request.

8. A customer service method based on AI digital human, characterized in that: The method comprises: Controlling the front-end display layer to generate a chat interface for user interaction with the AI digital human in response to a user clicking on an interface generation control; wherein the chat interface includes an input box; generating a query request containing the query content in response to the user entering the query content through the input box; and sending the query request to the digital human customer service interaction layer; Controlling the digital human customer service interaction layer to respond to the query request, perform sentiment analysis on the query content, and obtain the sentiment category corresponding to the query content; sending the query request and the sentiment category to the business processing layer; and matching the target virtual image and target expression feature corresponding to the sentiment category from a configured comparison table of different sentiment categories, different virtual images, and different expression features; and generating a target virtual image of the AI digital human based on the target virtual image and the target expression feature. When the control business processing layer receives the query request, it matches the response content corresponding to the query content from the configured customer service knowledge base; and sends the response content to the digital human customer service interaction layer; The digital human customer service interaction layer is controlled to use the target virtual image voice to output the response content corresponding to the query request fed back by the business processing layer.

9. An electronic device, characterized in that: The electronic device includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement the method described in claim 8 when executing the program stored in the memory.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to claim 8 is implemented.

Images