Message processing method and device, equipment, medium and product

Abstract

The invention provides a message processing method and device, equipment, a medium and a product, and relates to the field of big data and artificial intelligence. According to the method, a historical dialogue message set is obtained, a topic boundary and a topic logic link corresponding to the historical dialogue message set are determined, and a topic venation tree is generated based on the topic boundary and the topic logic link. When a to-be-processed message is obtained, a message intention corresponding to the to-be-processed message is determined, the message intention is used for indicating that the to-be-processed message is a commission message or a question message, and if the to-be-processed message is a to-be-processed message, a to-be-processed list is generated based on the topic context tree; and if the to-be-processed message is the question message, generating a reply statement based on the topic venation tree. According to the method, the topic vein tree can be generated according to the historical dialogue, the to-do list or the reply statement is automatically generated according to the intention of the to-be-processed message, and the dialogue processing efficiency and the intelligent level are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of big data and artificial intelligence, and in particular to a message processing method and device, equipment, medium and product. BACKGROUND

[0002] In the context of accelerating enterprise digital transformation, instant messaging groups have rapidly become the core carrier for cross-department collaboration, customer service and production support in enterprises due to their convenience. It breaks down departmental barriers, makes information transmission more efficient, and greatly improves the flexibility of enterprise operations, playing an increasingly important role in daily business operations.

[0003] Currently, there are some technical means for group communication management. There are basic message filtering tools based on keyword matching or simple rules that can shield some junk information; task management plug-ins allow users to manually mark messages as to-do tasks; static knowledge base systems can provide a pre-recorded common question and answer library; there are rule-based rule-type customer service robots that can automatically respond to limited scenarios.

[0004] However, there are high-frequency, fragmented multi-modal messages such as text, pictures, etc. in the group, which may cause information overload and key decision information may be submerged. At the same time, the task responsibility boundary is blurred, affecting the progress of work. Moreover, repeated inquiries consume a large amount of customer service resources, which reduces the decision-making efficiency and collaboration effectiveness of enterprises. SUMMARY

[0005] The present application provides a message processing method, device, equipment, medium and product to solve the problem that high-frequency, fragmented multi-modal messages in the group cause information overload, key decision information is easily submerged, task responsibility boundary is blurred, and repeated inquiries consume a large amount of customer service resources.

[0006] In a first aspect, the present application provides a message processing method, comprising:

[0007] obtaining a set of historical dialogue messages and determining topic boundaries and topic logical links corresponding to the set of historical dialogue messages;

[0008] generating a topic context tree based on the topic boundaries and the topic logical links;

[0009] when a to-be-processed message is obtained, determining a message intent corresponding to the to-be-processed message, the message intent being used to indicate that the to-be-processed message is a to-do message or a question message;

[0010] if the to-be-processed message is a to-do message, generating a to-do list based on the topic context tree;

[0011] If the to-be-processed message is a question message, a reply sentence is generated based on the topic context tree.

[0012] In a second aspect, the present application provides a message processing apparatus, comprising:

[0013] an acquisition module, configured to acquire a historical dialogue message set, and determine topic boundaries and topic logical links corresponding to the historical dialogue message set;

[0014] a generation module, configured to generate a topic context tree based on the topic boundaries and the topic logical links;

[0015] a determination module, configured to determine a message intention corresponding to a to-be-processed message when the to-be-processed message is acquired, the message intention being used to indicate that the to-be-processed message is a to-do message or a question message;

[0016] the generation module is further configured to generate a to-do list based on the topic context tree if the to-be-processed message is the to-do message;

[0017] the generation module is further configured to generate a reply sentence based on the topic context tree if the to-be-processed message is the question message.

[0018] In a third aspect, the present application provides an electronic device, comprising a processor, and a memory connected with the processor in communication;

[0019] the memory stores computer-executed instructions;

[0020] the processor executes the computer-executed instructions stored in the memory, so as to implement the message processing method according to the first aspect and possible implementation manners of the first aspect.

[0021] In a fourth aspect, the present application provides a computer-readable storage medium, which stores computer-executed instructions, and the computer-executed instructions are executed by a processor to implement the message processing method according to the first aspect and possible implementation manners of the first aspect.

[0022] In a fifth aspect, the present application provides a program product, comprising a computer program, and the computer program is executed by a processor to implement the message processing method.

[0023] The message processing method, device, equipment, medium and product provided by the application collect historical dialogue messages to form a set, and then identify the topic boundaries and logical links between topics in the set. Based on the boundary and link information, a topic context tree structure is constructed. When a new message to be processed is received, the message intent of the message is analyzed to determine whether the message is a task type message requiring a to-do item or a query type message asking a question. Differentiated processing is performed on different types of messages: if the message is determined to be a to-do type message, a to-do list is automatically generated based on the topic context tree; and if the message is determined to be a question type message, a targeted reply is generated by deducing the topic context tree. The method constructs a topic structure map of historical dialogue, intelligently generates a to-do guide or a question response according to the message intent, and effectively improves the processing efficiency and intelligent service capability of the message processing system. BRIEF DESCRIPTION OF DRAWINGS

[0024] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0025] Figure 1 Flowchart of a message processing method provided by an embodiment of the application Figure One ;

[0026] Figure 2 Flowchart of a message processing method provided by an embodiment of the application Figure Two ;

[0027] Figure 3 Flowchart of a message processing method provided by an embodiment of the application Figure Three ;

[0028] Figure 4 Flowchart of a message processing method provided by an embodiment of the application Figure Four ;

[0029] Figure 5 Flowchart of a message processing method provided by an embodiment of the application Figure Five ;

[0030] Figure 6 Structure diagram of a message processing device provided by the application;

[0031] Figure 7 Structure diagram of an electronic device provided by the application.

[0032] The specific embodiments of the application have been shown and described in the above drawings and text. These drawings and text are not meant to limit the scope of the inventive concept in any way but are merely meant to illustrate the inventive concept to one of ordinary skill in the art by reference to a particular embodiment. DETAILED DESCRIPTION

[0033] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description of the exemplary embodiments is intended to apply to any embodiment of the application, unless specified otherwise. Accordingly, when the description of the exemplary embodiments contains language that can imply limitations on the scope of the application, such limitations are not intended to apply to any specific embodiment provided those embodiments are otherwise expressly provided herein.

[0034] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portal for user to choose authorization or refusal.

[0035] And the present application involves big data analysis of user information (including but not limited to personal biological characteristics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and uses artificial intelligence technology for automatic decision-making, and makes technical solutions based on automatic decision-making results that have a significant impact on personal rights and interests, provides corresponding operation portal for users to choose to agree or refuse automatic decision-making results; if the user chooses to refuse, the expert decision-making process is entered.

[0036] It should be noted that the message processing method, device, equipment, medium and product provided by the present application can be used in the field of big data and artificial intelligence, and can also be used in any field other than big data and artificial intelligence. The application field of the message processing method, device, equipment, medium and product in the present application is not limited.

[0037] Instant messaging groups are now the core hub of cross-department collaboration, customer service and production support. However, the massive, fragmented multi-modal messages (including text, pictures, voice, files, etc.) in the group are causing increasingly severe information management challenges. High-frequency information interaction leads to key decision content being submerged, task responsibility boundaries being blurred due to scattered discussions, and repetitive consultation questions continuously consuming customer service team resources. This information overload phenomenon not only directly reduces enterprise decision-making efficiency and team collaboration effectiveness, but also affects customer satisfaction due to impaired user experience.

[0038] Current methods for addressing group communication management needs mainly include: basic message filtering tools that shield spam information through preset keyword libraries or simple rule engines, task management plug-ins that support users manually marking to-do items and assigning responsible persons, and static knowledge base systems that rely on pre-entered common question and answer libraries to provide standardized problem solving. These tools alleviate information overload pressure to some extent, for example, filtering tools can reduce low-value content interference, task plug-ins can assist in tracking execution progress, and knowledge bases can improve information retrieval efficiency through structured storage.

[0039] However, the semantic understanding mechanism based on keyword matching is difficult to capture the context association of messages, often misjudging high-value discussions as noise or missing potential risk points; task management is highly dependent on human intervention, not only failing to automatically extract to-do items, responsible subjects and deadlines from conversations, but also lacking intelligent reminder mechanisms, leading to task overruns; static knowledge bases are limited by the pre-entered update mode, unable to absorb new problems that arise in the group, and difficult to continuously optimize solution strategies through machine learning, resulting in high repeat consultation rates.

[0040] To address the above problems, the present application proposes a message processing method, which integrates fragmented historical messages into a structured topic context tree, and then combines text processing and analysis techniques to judge the message intent of instant messages, and makes targeted replies to instant messages or automatically generates to-do items to remind relevant users to handle them in a timely manner. In this way, intelligent analysis and filtering of information and automatic generation of to-do items are achieved, effectively improving cross-department collaboration efficiency and accelerating decision-making response speed.

[0041] The present application can be applied to enterprise cross-department collaboration office scenarios. In daily enterprise operations, a large number of fragmented exchange messages will occur between different departments, and this method can help efficient collaboration. It can also be applied to large project team communication scenarios, where members exchange frequently and information is complex. This method can intelligently analyze and filter information, automatically generate to-do items, and ensure that key project tasks are handled in a timely manner. The execution subject of the present application can be a message processing system integrated in a chat software.

[0042] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0043] Figure 1 Flowchart of a message processing method provided by an embodiment of the present application Figure One . As shown in Figure 1 , the message processing method provided by the embodiment includes:

[0044] S101: Obtain a historical conversation message set and determine topic boundaries and topic logical links corresponding to the historical conversation message set.

[0045] It can be understood that obtaining the historical conversation message set can extract all conversation messages in the past period of time or under a specific conversation scenario from a data source such as a database storing historical conversations, a chat record file, or a message queue, to form a complete message set.

[0046] The topic boundary refers to the starting and ending positions of each topic. In the historical conversation, a topic can be composed of multiple related messages, and there is a clear distinction between different topics. The topic boundary can be determined by analyzing keywords, semantic transitions, topic switching cues, and the like in the message content.

[0047] The topic logical link describes the logical relationship between topics, such as cause and effect, parallel relationship, progressive relationship, and the like. Through semantic analysis and context understanding of the historical conversation messages, these logical relationships can be sorted out.

[0048] Optionally, the specific determination process of the topic boundary and the topic logical link can be:

[0049] Determine the close contact degree between each historical message in the historical conversation message set.

[0050] It can be understood that determining the close contact degree of each historical message can be analyzed based on semantic similarity, using a word vector model in natural language processing technology to convert each message into a vector representation, and then measuring the semantic association degree between messages by calculating the cosine similarity between vectors. The higher the similarity, the more closely related the two messages are in semantics.

[0051] The keyword overlap in the messages can also be considered to determine the close contact degree. Specifically, the number of keywords commonly appearing in the two messages is counted, and the more the keyword overlap, the higher the close contact degree can be. In addition, the context of the messages can also be combined to analyze the reference and reply relationship between the messages. If one message is a direct reply or reference to another message, the close contact degree between them will be significantly enhanced.

[0052] Based on the time interval of the historical messages, the topic boundary is determined.

[0053] It can be understood that when a long time interval appears in the conversation, it mostly means that the user in the chat may have temporarily interrupted the discussion of the current topic and turned to handle other matters or think of a new topic, and then continued to communicate, which may start a new topic.

[0054] When determining the topic boundary according to the time interval, the threshold of the time interval is crucial, which needs to be set according to the specific application scenario and the characteristics of the conversation. A suitable value can be selected as the threshold by statistical analysis of a large amount of historical conversation data to observe the time interval distribution when the topic is switched. For example, if the time interval when most of the topics are switched is concentrated above 30 minutes, 30 minutes can be used as a preliminary threshold. When the time interval between adjacent messages exceeds this threshold, it is considered that a topic boundary may appear. At the same time, in order to more accurately determine, other factors such as the mutation of the message content can also be combined. If the time interval is long and the message content is obviously unrelated to the previous topic, the credibility of determining the topic boundary is higher.

[0055] According to the close contact degree, the topic logical link is determined.

[0056] It can be understood that the close contact degree reflects the degree of association between historical messages, and based on this degree of association, the logical link of the topic can be sorted out. When determining the topic logical link, the starting message of the conversation can be started, and the subsequent messages related to it can be found in order of close contact degree from high to low. For example, if the close contact degree between message A and message B is high, and message B has a high close contact degree with message C, then A-B-C can be preliminarily determined to constitute part of a topic logical link.

[0057] In order to more comprehensively and accurately determine the topic logical link, a graph theory method can be used. Each historical message is regarded as a node in the graph, and the close degree of connection between messages is regarded as the weight of the edge between nodes. By constructing such a graph structure, a graph algorithm (such as a depth-first search, a breadth-first search, or a shortest path algorithm, etc.) is used to find a logical path from the starting node to other nodes, thereby determining the logical link of the entire topic. At the same time, when constructing the logical link, the order of the conversation also needs to be considered to ensure that the messages on the link are sequentially expanded in chronological order to conform to the normal conversation logic.

[0058] S102: generating a topic context tree based on the topic boundary and the topic logical link.

[0059] It can be understood that the topic context tree is a model that visually displays the hierarchical relationship and logical connection between the topic structures in the historical conversation in a tree structure. According to the determined topic boundary, the entire conversation is divided into multiple independent topic modules, and each topic module corresponds to a branch or node in the tree. Then, for each topic module, according to the determined topic logical link, the messages are organized in a logical order. For example, the starting message is taken as the root node of the tree or the root node of the sub-tree, and then the subsequent messages are sequentially added to the corresponding parent node as child nodes according to the logical link, forming a hierarchical tree structure.

[0060] In the process of generating the topic context tree, some special cases also need to be considered. For example, there can be multiple branch logical links within a topic module, that is, multiple parallel sub-topics appear when the conversation discusses a topic, and at this time the branch relationship needs to be reasonably reflected in the tree structure, and multiple child nodes are added to represent different sub-topics.

[0061] S103: determining the message intent corresponding to the to-be-processed message when the to-be-processed message is obtained.

[0062] Among them, the message intent is used to indicate that the to-be-processed message is a to-do message or a question message.

[0063] It can be understood that when a new message enters the message processing system, this new message is the to-be-processed message, which can be a new text message, a voice message, etc. input by the user, for example, in a chat software, the message sent by the user through the input box is the to-be-processed message.

[0064] The message type is used to indicate the purpose of the user sending the message. Through natural language processing techniques such as text classification algorithms, semantic analysis models, etc., the message to be processed is analyzed and judged to determine whether it is a to-do message or a question message. For example, if the message content is "Help me book a conference room at 3 pm tomorrow", it is obviously a to-do message, i.e. informing the corresponding user to complete a task; if the message content is "Is the conference room available at 3 pm tomorrow", it is a question message, i.e. the user wants to obtain certain information.

[0065] Optionally, in the case where there are multiple to-be-processed messages, the order of the to-be-processed messages can be determined according to the message type, specifically including:

[0066] Determining the message type of the to-be-processed message.

[0067] It can be understood that the message type can include text messages, file messages, picture messages, etc. In actual application scenarios, text messages can also be divided into ordinary text messages, @ messages, specific instruction messages, etc.

[0068] In determining the message type, a preliminary judgment can be made based on the format and content of the message. For example, if the format of the message conforms to the common file suffix (such as.docx,.pdf,.xlsx, etc.), it can be classified as a file message; if the message starts with a specific interrogative word (such as "what", "how", "where", etc.), or contains sentences with characteristics such as asking for information or seeking help, it can be judged as a question message; and when the message contains explicit to-do item descriptions such as "Please submit the report before this Friday" or "Arrange a meeting", it can be identified as a to-do message.

[0069] Natural language processing techniques and machine learning algorithms can also be used. By building a message classification model, a large amount of labeled message data is used to train the model, allowing the model to learn the characteristics and patterns of different message types. In actual application, the to-be-processed message is input into the trained model, and the model can output the type to which the message belongs. This method can handle more complex and diverse messages, improving the accuracy and diversity of message classification.

[0070] According to the message type and the preset weight, the priority of the to-be-processed message is determined.

[0071] The preset weight is pre-set according to the importance and urgency of the message type. For example, an urgent @ message may involve an urgent reminder of an important matter, so it has high importance and urgency, and therefore can be given a higher weight; a file message may contain important data or task requirements, and the weight is second; the weight of an ordinary text message is relatively low. This is a simple and effective way to set the basic weight based on the message type, which can quickly sort the messages.

[0072] It can be understood that in addition to relying on the basic weight of the message type to reflect the message priority, a multi-feature fusion strategy can also be used to dynamically calculate the message priority. In addition to the message type, a keyword matching method can also be combined. By pre-setting some keywords such as "fault", "deadline", "approval", etc., when the message contains these keywords, it indicates that the message may involve an urgent problem or important task, and the priority of the message needs to be improved. For example, a normal text message contains the word "fault", which may mean that the system has a problem that needs to be handled immediately, at which time the priority of the message can be adjusted according to the weight of the keyword.

[0073] Context correlation analysis is also one of the important methods to determine the message priority. The session coherence modeling based on LSTM (Long Short-Term Memory Network) can analyze the context relationship between messages and understand the position and role of the message in the entire conversation or business process. If a message is highly related to the important task or emergency situation currently being handled, its priority should also be correspondingly improved. For example, in a conversation discussing a project, the progress of the project has been discussed all the time, and suddenly a message about a problem in a critical part of the project appears. Through context correlation analysis, it can be determined that this message is closely related to the current important task, thereby improving its priority.

[0074] Based on the above multiple features, the priority of each message to be processed is dynamically calculated, and a ranked list is output, such as urgent > high > medium > low. This can ensure that core problems and urgent problems are prioritized to reach the processing node, improving processing efficiency and response speed.

[0075] According to the priority, the messages to be processed are pushed to the processing node, and the message intent of the message to be processed is determined in turn.

[0076] It can be understood that after the priority of the message to be processed is determined and the ranked list is generated, the messages can be pushed to the processing node in order of priority from high to low. The processing node can be a server or a specific processing program, etc., responsible for further analysis and processing of the message to be processed.

[0077] S104: If the message to be processed is a to-do message, generate a to-do list based on the topic context tree.

[0078] It can be understood that the to-do list is to arrange and list the specific tasks in the to-do message according to certain rules, facilitating subsequent execution and management. It can help users clearly understand the task content and order to be completed.

[0079] Once the pending message is identified as a task, its relationship with various topics in the topic hierarchy tree is analyzed. Based on the logical relationships and hierarchical structure between topics, the tasks are appropriately inserted into their corresponding positions in the topic hierarchy tree, and key information is extracted to generate a to-do list. For example, in a project management dialogue, the pending message is "Complete the market analysis section of the project report." Combining this with the topics related to the various chapters of the project report in the topic hierarchy tree, this task is added to the relevant topics under the market analysis chapter, generating a to-do list containing information such as task name, responsible person, and deadline.

[0080] S105: If the message to be processed is a question message, then generate a reply statement based on the topic context tree.

[0081] Understandably, a response should accurately and clearly answer the questions raised in the message, and should be consistent and coherent with relevant information in the topic structure. It should satisfy the user's need for information while maintaining the logic and flow of the conversation.

[0082] When the message to be processed is a question, the first step is to locate the relevant topic node in the topic context tree. Then, based on the specific information and logical relationships under the topic node, and combined with natural language generation technology, an appropriate response is generated. For example, if a user asks, "What is the after-sales service like for this product?", the topic node related to the product's after-sales service is found in the topic context tree, the service content is extracted, and organized into natural language to reply to the user, such as, "Our product's after-sales service includes a one-year free warranty, lifetime maintenance, and other services."

[0083] This embodiment provides a message processing method that acquires a set of historical dialogue messages and determines the topic boundaries and logical links corresponding to the set. Based on these boundaries and links, a topic context tree is generated. When a message to be processed is received, the message intent is determined. The intent indicates whether the message is a task or a question. If the message is a task, a task list is generated based on the topic context tree; if it is a question, a response is generated. This method can generate a topic context tree based on historical dialogues and automatically generate task lists or responses based on the intent of the message to be processed, improving dialogue processing efficiency and intelligence.

[0084] Figure 2 A flowchart illustrating a message processing method provided in this application embodiment. Figure Two .like Figure 2 As shown, in Figure 1 Based on the embodiments, the process of processing multimodal messages is described in detail, including:

[0085] S201: Determine a first message and a second message in the to-be-processed messages.

[0086] The first message is in a non-text format, and the second message is in a text format.

[0087] It can be understood that the to-be-processed messages can come from various channels, and in order to prevent important messages from being missed, the messages can be classified and processed. The message format can be determined by checking the file extension of the message (such as.jpg,.png,.gif for picture format, such as.wav,.mp3,.amr for voice format, such as.doc,.docx,.xls,.xlsx,.pdf for file format).

[0088] In addition to the file extension, the file header information can also be read to determine whether it is in a picture format; the audio signal processing technology can be used to analyze the audio characteristics of the message to determine whether the to-be-processed message is a voice message; by reading the content structure of the file, it can be checked whether it contains document-specific markers and format information, such as paragraph markers and table structures in Word documents, to determine whether it is a document message.

[0089] For the second message in the text format, simple text encoding detection can be used to confirm whether it is a valid text message.

[0090] S202: Perform character conversion and extraction processing on the first message to obtain a corresponding third message.

[0091] The third message is in a text format.

[0092] It can be understood that the first message can include messages in picture format, voice format, and text format. Different methods can be used to convert messages in different formats into a third message in a text format.

[0093] For messages in a picture format, OCR technology (Optical Character Recognition) can be used to recognize the text and tables in the picture. Image segmentation algorithms can also be used. The image segmentation algorithm can separate the table from the picture, and then process the separated table separately to improve the accuracy of table recognition.

[0094] For messages in voice format, ASR technology (Automatic Speech Recognition) can be used to convert voice to text format. To improve the accuracy of voice-to-text conversion, acoustic models and language models can be combined for optimization. Acoustic models are used to convert voice signals into phoneme sequences, while language models are used to perform syntax and semantic analysis on phoneme sequences, thereby improving the coherence and accuracy of the text. In addition, context correction algorithms in natural language processing can be used to further optimize text quality.

[0095] For messages in file format, document parsing tools can be used to read, write, convert, and other operations on documents. At the same time, rule engines and text summarization techniques (such as generative models) can be combined to extract document titles, core conclusions, and data tables, and generate content summaries, achieving structured output of document content.

[0096] S203: Filtering the second message and the third message to obtain the to-be-processed text corresponding to the to-be-processed message.

[0097] It can be understood that after converting the format of all to-be-processed messages into text form, the text classification model (such as the bidirectional encoder representation model based on Transformer) can be used to classify the semantics of the message in combination with the rule engine (regular matching emoticons, stop word library), automatically filter emoticons, invalid greetings, and low-value short texts.

[0098] In order to retain high-value text containing decision keywords, a decision keyword library can be constructed, and common decision keywords such as "demand", "fault", "emergency", "approval", "decision", etc. are included. When classifying and filtering text messages, check if the text contains these decision keywords, and retain the text containing the decision keywords. At the same time, in order to improve the accuracy of keyword matching, fuzzy matching, stem extraction and other technologies can be used to ensure that the text related to decision can be accurately identified.

[0099] The message processing method provided by the embodiment distinguishes a first message (covering messages in formats of pictures, voice, documents, etc.) from a second message of text in the to-be-processed message, then, for the first message, uses OCR, ASR, document analysis, etc. in combination with various auxiliary algorithms to realize text conversion and extraction, and obtains a third message of text, then, for the second and third messages, uses a text and image classification model, a rule engine, etc. to carry out filtering processing, and retains high-value text containing decision keywords. Through this series of processing, efficient classification and accurate processing of multi-modal messages are realized, invalid noise information is effectively filtered, the extraction accuracy and integrity of key information are greatly improved, data support is provided for subsequent generation of to-do tasks, reply to questions, etc. based on key information, and the efficiency and value of information processing are enhanced.

[0100] Figure 3 The flowchart of the message processing method provided by the embodiment of the application Figure Three . As Figure 3 shown, on the basis of the embodiment, the process of generating a to-do list is described in detail, including: Figure 1

[0101] S301: Determine a first topic branch related to the to-do message according to the topic context tree.

[0102] It can be understood that when the to-do message appears, the message processing system can analyze the core theme and key information involved in the to-do message, then start from the root node of the topic context tree, and gradually investigate along each branch, find the branch that is most closely related to the to-do message and has the most direct logical connection by comparing the theme features of the message content and each branch, and this branch is the first topic branch. It provides a clear topic range and context background for subsequent processing of the to-do message, and ensures that the message processing system can accurately grasp the position and meaning of the to-do item in the entire dialogue context.

[0103] S302: Perform marking processing on the to-do entity in the first topic branch, and output a structured task list.

[0104] It can be understood that after the first topic branch is determined, a named entity recognition model (for example, a bidirectional long short-term memory network-conditional random field model) can be used to perform marking processing on the to-do entity in the first topic branch. The bidirectional long short-term memory network-conditional random field model has strong sequence labeling capability and can identify various entities in the text.

[0105] ​For the personnel mentioned in the topic branch message, the model can identify the mentioned personnel, who are often the responsible persons of the to-do list, and accurately label them to clarify the task ownership. For example, the expression "complete before X" can be identified by the model as a time node and deadline requirement, which is crucial for reasonable task scheduling and monitoring of task completion. Task items are the core content of the to-do list, such as "fix program defects", which can be accurately extracted and labeled by the model to help users know what specific work to do.

[0106] At the same time, in order to enhance the comprehensiveness and accuracy of entity coverage, the message processing system can also combine the field word library, such as the post name table, to include professional vocabulary in the specific field into the recognition range to avoid missing important entities. After a series of marking processing, the message processing system can organize the labeled information and output the structured task list in a specific format. The list clearly lists the responsible person, task content and deadline, etc. key information, so as to eliminate the blind area of responsibility and make each task have a clear responsible person and time requirement.

[0107] S303: According to the historical messages and to-do messages in the first topic branch, determine the to-do event elements, and generate a to-do summary according to the preset rule template.

[0108] It can be understood that, in order to fully and accurately grasp the to-do event, the message processing system not only pays attention to the to-do message itself, but also analyzes the historical messages in the first topic branch. These historical messages may contain information about the background of the to-do event, the pre-discussion situation and the relevant decision basis, which is crucial for a complete understanding of the to-do event.

[0109] The message processing system can use a text classification model, such as the TextCNN model (Text Convolutional Neural Network), to identify the conversation intent of the historical messages. By analyzing the context, words and expression of the conversation, the system can determine whether the conversation belongs to decision discussion, demand collection or problem feedback, etc. Different types of conversation intent will have different understanding and processing methods for the to-do event.

[0110] After determining the dialogue intent, the message processing system inputs the dialogue text of the pending topic branch into a generative model. This model organizes and integrates the dialogue content according to the structure of "context → task → action → result", generating coherent paragraphs. At the same time, to ensure that the summary is logical and executable, the system can use rule templates such as "Question: {Situation}; Goal: {Task}" to force the structure of the generated content, ensuring that the pending summary not only contains the key elements of the pending event, such as the context of the event and the goal to be achieved, but also has a clear logical structure, making it easy for relevant personnel to quickly understand and execute.

[0111] S304: Based on the structured task list and the pending summary, generate a pending list.

[0112] It can be understood that after having a structured task list and a pending summary, the message processing system begins to generate the final pending list. To ensure that the to-do items are operable and measurable, the system can refine and perfect each to-do item according to the SMART principle. The SMART principle refers to the principles of Specificity, Measurability, Attainability, Relevance, and Time-bound.

[0113] The message processing system can clearly mark the responsible person, so that everyone is clear about their task responsibilities; set a reasonable time limit and specify that the task must be completed within the specified time; and establish clear delivery standards, such as "passing QA (Quality Assurance)", so that the completion of the task has a clear basis for evaluation.

[0114] During the generation of the pending list, the message processing system will also mark possible points of contention. These points of contention may be caused by unclear information, differences in understanding, etc., and are marked for manual review to ensure the accuracy and completeness of the pending list.

[0115] In addition, the message processing system can also combine the domain template library, for example, for operation and maintenance tasks, the template library contains specific fields such as "system module" and "impact range", and the system can automatically fill in these information according to the actual needs of the task. At the same time, automatically assign priority labels (P0-P3) to each task, and reasonably sort them according to the importance and urgency of the task.

[0116] Finally, the to-do list generated by the message processing system is standardized and can be directly imported into a project management tool, such as Jira (Atlassian JIRA), to achieve full-cycle management of tasks. From task allocation, execution, to monitoring and completion, all can be effectively tracked and managed in the project management tool, improving work efficiency and the quality of project management.

[0117] The message processing method provided by the embodiment accurately locates a first topic branch related to the to-do message according to the constructed topic context tree, provides a clear context for subsequent processing, accurately marks the to-do entity in the first topic branch by using a named entity recognition model in combination with a domain word library, outputs a structured task list with clear responsibility person, task content, and deadline, and eliminates the responsibility blind area; then, a dialogue intent is recognized by using a text classification model, and a to-do summary with clear logic and strong executability is generated by using a generative model; finally, a standardized to-do list is generated based on the structured task list and the to-do summary. The method comprehensively integrates dialogue information, accurately extracts task elements, realizes full-cycle management of tasks, and effectively improves the accuracy of task allocation and the efficiency of execution.

[0118] Figure 4 Flowchart of a message processing method provided by the embodiment of the application Figure Four . As shown in Figure 4 , on the basis of the embodiment, the process of generating a reply sentence is described in detail, including: Figure 1

[0119] S401: Determine the confidence degree of the question message and all question indexes in the knowledge base based on the knowledge base.

[0120] Among them, the knowledge base stores a plurality of question indexes and answers corresponding to the question indexes.

[0121] It can be understood that the confidence degree is an index for measuring the matching degree of the question message and the question index in the knowledge base, and can be calculated by natural language processing technology (such as word vector matching, semantic analysis or deep learning model). The goal of this process is to quickly locate whether there is a question in the knowledge base that is highly related to the current question, to provide a basis for subsequent automatic reply.

[0122] S402: Determine whether there is a case where the confidence degree is greater than a first preset confidence degree, if yes, execute step S403, if not, execute step S404.

[0123] ​It can be understood that after obtaining the confidence of the question message and all question indexes in the knowledge base, the message processing system can determine whether there is at least one confidence exceeding a preset threshold (a first preset confidence). This threshold is set by the system designer according to actual needs when designing the system, and is used to distinguish between "high matching" and "low matching" questions. If there is a question index with a confidence exceeding the threshold, it means that there is a standard question in the knowledge base that is highly matched with the current question, and the system can directly obtain a reply sentence based on the question index; otherwise, it means that the question may not be covered by the knowledge base and needs to be further processed.

[0124] S403: Determine a reply sentence according to the question index corresponding to the confidence.

[0125] It can be understood that when the system detects that there is a question index with a confidence exceeding the threshold, the standard answer corresponding to the question index can be directly extracted from the knowledge base as the reply sentence. This step reflects the fast response capability of the system and is suitable for the reply of common questions or standardized scenarios. For example, in a customer service group, a user asks "how to return and exchange?", if there is a same question index in the knowledge base, the system can immediately return the preset return and exchange process description without further analysis.

[0126] S404: Determine a matching result according to the confidence.

[0127] The matching result is the question index and answer corresponding to the maximum confidence.

[0128] It can be understood that if the confidence of all question indexes does not exceed the threshold, the system can select the question index with the highest confidence and its answer as the "matching result". Although this matching may not be completely accurate, it provides the closest reference information in the absence of a more optimal option. For example, a user asks "how long does it take to return?", and the knowledge base only has "return and exchange process" related questions, the system can extract the answer to the process, so that subsequent reply sentences consistent with the question message can be generated based on the answer.

[0129] S405: Determine a historical message related to the question message according to the topic context tree.

[0130] Understandably, the topic context tree is a categorized storage structure for historical dialogue messages. It stores historical messages according to topics or dialogue contexts, facilitating the tracing of context. The system can analyze the topic to which the current question belongs and extract historical messages related to that topic from the topic context tree. These historical messages may include previous questions from the user, system replies, or discussions from other group members, providing contextual support for generating more accurate responses. For example, in a technical discussion group, if a user currently asks "How to solve code errors?", the system can retrieve previously mentioned code snippets or error logs to understand the specific problem background.

[0131] S406: Generate a response statement based on the question message, historical messages, and matching results.

[0132] Understandably, after integrating the question message, historical messages, and matching results, the system can invoke a generative model (such as a Transformer-based large language model) to generate the final response. The generative model can combine contextual information to understand the deeper meaning of the question and generate a natural and fluent response.

[0133] After generating a response, the system can push it to the user's response window for the user to assess its reliability. If the user deems the response accurate and highly credible, they can directly send the response, replying to the question. If the user deems the response less credible, they can manually adjust the response content and finally send the modified response. The question and response are then stored in the knowledge base, enabling dynamic updates and optimization of the knowledge base. This process not only improves the quality of responses but also continuously refines the system's knowledge base through user feedback, forming a closed-loop optimization mechanism.

[0134] This embodiment provides a message processing method that first calculates the confidence level between the question message and each question index based on a knowledge base to determine if a high-confidence match exists. If one exists, the corresponding standard answer is directly invoked; otherwise, the best approximate match result is selected. Simultaneously, relevant historical messages are traced using a topic context tree to supplement the current question with contextual information. Finally, a generative model integrates the question content, historical records, and matching results to generate a natural and fluent response. This method also supports a user feedback mechanism, allowing for the correction of response content and dynamic updates to the knowledge base based on user feedback. This method significantly reduces the need for manual intervention and effectively improves the efficiency and response quality of group message management.

[0135] Figure 5 A flowchart illustrating a message processing method provided in this application embodiment. Figure Five .like Figure 5 As shown, in Figure 1 Based on the examples, the process of generating the knowledge base is described in detail, including:

[0136] S501: Obtain the question information in the topic context tree and generate a semantic vector of the question information.

[0137] It can be understood that the question sentence is the basis for building the question and answer knowledge base, and the topic context tree stores historical information, so all question and answer sentences can be found in the topic context tree, and the knowledge base is built according to the question and answer sentences. In terms of clearly defining the specific semantics of each question, using an encoder model (such as the Sentence-BERT model) is an efficient and accurate method. The Sentence-BERT model is based on advanced natural language processing technology and can deeply understand and analyze the input text, converting each user question into a semantic vector. This semantic vector is a digital representation of the question sentence in the semantic space, which captures the key semantic features in the question, allowing the semantic relationship between different questions to be measured through mathematical operations between vectors, providing strong data support for subsequent similarity calculation and grouping operations.

[0138] S502: Determine the similarity between the question information based on the semantic vector.

[0139] It can be understood that the semantic vector can be regarded as a point in a multi-dimensional space in mathematics, and the similarity between two vectors can be measured by calculating the distance or angle between them. A commonly used calculation method is cosine similarity, which evaluates the similarity of two vectors by calculating the cosine value of the angle between them. The closer the cosine value is to 1, the more similar the two vectors are, i.e., the more similar the corresponding question information is in semantics; on the contrary, the closer the cosine value is to 0, the greater the semantic difference between the two question information. In this way, the similarity between each question information can be quantified, providing a clear basis for subsequent grouping operations.

[0140] S503: Group the question information according to the similarity, determine the problem index of each group of question information, and determine multiple vector spaces.

[0141] Among them, a vector space corresponds to the question information of a group.

[0142] It can be understood that according to the calculated similarity, the question information can be divided into multiple groups, i.e., the repeated questions are integrated, and by setting a suitable similarity threshold, the question information with a similarity higher than the threshold is classified into the same group. In this way, similar sentences can be effectively gathered together to form a group with common semantic features.

[0143] For each group, determine its problem index, which is the problem corresponding to this vector space, which represents the core semantics of the group's question information, facilitating subsequent knowledge base query and management. At the same time, the question information of each group corresponds to an independent vector space, which is composed of the semantic vectors of all question information in the group. In the vector space, each question information semantic vector has its specific position, and the relative position relationship between them reflects the semantic similarity and difference between the question information, providing an intuitive geometric representation for further analysis and processing of question information.

[0144] S504: For one of the vector spaces, determine the answer information of the vector space according to the topic context tree.

[0145] It can be understood that the topic context tree stores rich historical dialogue information, which contains multiple rounds of dialogue related to the target problem. The target problem is the problem corresponding to the current vector space, and by tracing the dialogue records related to the problem in the topic context tree, complete context information can be obtained. In order to accurately locate the answer key sentence from these multiple rounds of dialogue, a rule engine (such as regular matching keywords "solution" "step" etc.) can be combined. These keywords are often closely related to the answer, and by matching these keywords with regular expressions, sentences that may contain answers can be quickly filtered out.

[0146] However, the filtered sentences may have problems such as incomplete expression and redundancy, so the filtered sentences can be further processed by a generative model (based on the Transformer architecture). The generative model can understand the semantics and syntax structure of the sentence, and generate a concise and coherent standard reply template according to the context information, such as "Please check the network connection and restart the application". Such a reply template is more in line with the needs of practical applications, and is convenient for users to understand and operate.

[0147] S505: Generate a knowledge base according to the problem index and the answer information.

[0148] It can be understood that after determining the problem index and the corresponding answer information, the last step is to save them to generate a knowledge base. The problem index serves as an entry for query, allowing quick positioning to the problem of interest to the user; the answer information provides specific solutions or answers for the user. Storing the problem and answer in a structured manner in the knowledge base, such as in the form of a database, facilitates query, update and maintenance. The generation of the knowledge base not only realizes the effective integration and utilization of historical dialogue information, but also provides rich knowledge support for the subsequent intelligent question answering system, enabling the system to quickly and accurately answer user questions, improving user experience and service quality.

[0149] The message processing method provided by the embodiment extracts a question sentence in a topic context tree, generates a semantic vector of the question sentence, quantifies the similarity between question information, groups similar questions, and determines a question index and a corresponding vector space. For each vector space, the topic context tree is traced back to relevant multi-round dialogues, a rule engine is combined to locate an answer key sentence, and a generative model is used to generate a standard reply template. Finally, the question index and the answer information are saved to generate a knowledge base. The structured integration of historical dialogue information is realized, the question semantics can be accurately captured, similar questions can be efficiently grouped, accurate and concise answers can be quickly located and generated, the response speed and accuracy of the question message are improved, and high-quality and efficient intelligent question and answer services are provided for users.

[0150] Figure 6 A structural schematic diagram of a message processing device provided by the application is shown in the figure. Figure 6 The message processing device 600 provided by the application comprises:

[0151] The acquisition module 601 is configured to acquire a historical dialogue message set and determine a topic boundary and a topic logical link corresponding to the historical dialogue message set.

[0152] The generation module 602 is configured to generate a topic context tree based on the topic boundary and the topic logical link.

[0153] The determination module 603 is configured to determine a message intent corresponding to a to-be-processed message when the to-be-processed message is acquired, the message intent being used to indicate that the to-be-processed message is a to-do message or a question message.

[0154] The generation module 602 is configured to generate a to-do list based on the topic context tree if the to-be-processed message is a to-do message.

[0155] The generation module 602 is configured to generate a reply sentence based on the topic context tree if the to-be-processed message is a question message.

[0156] Optionally, the determination module 603 is specifically configured to determine a close contact degree between each historical message in the historical dialogue message set, determine a topic boundary based on a time interval of the historical message, and determine a topic logical link according to the close contact degree.

[0157] Optionally, the device further comprises a processing module 604.

[0158] The determination module 603 is further configured to determine a first message and a second message in the to-be-processed message, the first message being in a non-text format, and the second message being a message in a text format.

[0159] The processing module 604 is configured to perform character conversion and extraction processing on the first message to obtain a corresponding third message, the third message being a text format message; and perform filtering processing on the second message and the third message to obtain a to-be-processed text corresponding to the to-be-processed message.

[0160] Optionally, the generation module 602 is specifically configured to determine a first topic branch related to the to-be-dealt-with message according to the topic context tree; perform marking processing on a to-be-dealt-with entity in the first topic branch to output a structured task list; determine a to-be-dealt-with event element according to a historical message in the first topic branch and the to-be-dealt-with message, and generate a to-be-dealt-with summary according to a preset rule template; and generate a to-be-dealt-with list based on the structured task list and the to-be-dealt-with summary.

[0161] Optionally, the generation module 602 is specifically configured to determine a confidence degree of the question message and all question indexes in a knowledge base based on the knowledge base, the knowledge base storing a plurality of question indexes and answers corresponding to the question indexes; determine whether the confidence degree is greater than a first preset confidence degree; if the confidence degree is greater than the first preset confidence degree, determine a reply statement according to a question index corresponding to the confidence degree; if the confidence degree is not greater than the first preset confidence degree, determine a matching result according to the confidence degree, the matching result being a question index and an answer corresponding to a maximum confidence degree; determine a historical message related to the question message according to the topic context tree; and generate a reply statement according to the question message, the historical message, and the matching result.

[0162] Optionally, the knowledge base is constructed in the following manner:

[0163] Obtain question information in the topic context tree, and generate a semantic vector of the question information;

[0164] Determine a similarity degree between the question information based on the semantic vector;

[0165] Group the question information according to the similarity degree, determine a question index of each group of question information, and determine a plurality of vector spaces, one vector space corresponding to one group of question information;

[0166] For one of the plurality of vector spaces, determine answer information of the vector space according to the topic context tree;

[0167] Generate the knowledge base according to the question index and the answer information.

[0168] Optionally, the determining module 603 is further configured to determine a message type of the to-be-processed message when there are multiple to-be-processed messages; determine a priority of the to-be-processed message according to the message type and a preset weight, the preset weight being preset according to an importance and an urgency of the message type; and push the to-be-processed message to a processing node according to the priority, and determine a message intent of the to-be-processed message in sequence.

[0169] The message processing apparatus provided by the embodiments of the present application has similar implementation principles and technical effects to the implementation manners of the various parts of the message processing method described above, and thus detailed descriptions are omitted here.

[0170] Figure 7 A structural schematic diagram of an electronic device is provided in the present application. As shown in Figure 7 The present application provides an electronic device, which comprises a receiver 701, a transmitter 702, a processor 703 and a memory 704.

[0171] The receiver 701 is configured to receive instructions and data.

[0172] The transmitter 702 is configured to transmit instructions and data.

[0173] The memory 704 is configured to store computer execution instructions.

[0174] The processor 703 is configured to execute the computer execution instructions stored in the memory 704 to implement the various steps performed by the message processing method in the above embodiments. For details, refer to the related descriptions in the message processing method embodiments described above.

[0175] Optionally, the memory 704 can be independent or integrated with the processor 703.

[0176] When the memory 704 is independently arranged, the electronic device further comprises a bus configured to connect the memory 704 and the processor 703.

[0177] The implementation principles and technical effects of the electronic device provided by the present embodiment can refer to the descriptions of the various embodiments described above, and thus detailed descriptions are omitted here.

[0178] The embodiments of the present application further provide a computer readable storage medium, which stores computer execution instructions. When a processor executes the computer execution instructions, the method described in any of the above embodiments is implemented.

[0179] The embodiments of the present application further provide a computer program product, which comprises a computer program. When the computer program is executed by a processor, the method described in any of the above embodiments is implemented.

[0180] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all described as a combination of a series of actions, but those skilled in the art should know that the present application is not limited to the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0181] Further, it should be noted that although each step in the flowchart is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or sub-steps or stages of other steps.

[0182] It should be understood that the above-described device embodiments are only illustrative, and the device of the present application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and actual implementation can have another division method. For example, multiple units, modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed.

[0183] In addition, unless otherwise specified, each functional unit / module in each embodiment of the present application can be integrated in one unit / module, or each unit / module can exist physically, or two or more units / modules can be integrated together. The integrated unit / module can be realized in the form of hardware or in the form of a software program module.

[0184] If the integrated units / modules are implemented in the form of hardware, the hardware can be a digital circuit, an analog circuit, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor can be any appropriate hardware processor, such as a CPU, a GPU, an FPGA, a DSP, an ASIC, etc. Unless otherwise specified, the storage unit can be any appropriate magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc.

[0185] If the integrated units / modules are implemented in the form of software program modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the essential part or all or part of the technical solutions that make contributions to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0186] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0187] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0188] It is to be understood that the application is not limited to the precise construction herein disclosed and shown in the drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the application. The scope of the application is limited only by the claims that follow.

Claims

1. A message processing method, characterized in that, include: Obtain a set of historical dialogue messages and determine the topic boundaries and topic logic links corresponding to the set of historical dialogue messages; Based on the topic boundaries and the topic logical links, a topic context tree is generated; When a message to be processed is obtained, the message intent corresponding to the message to be processed is determined. The message intent is used to indicate that the message to be processed is a to-do message or a question message. If the message to be processed is a pending message, then a pending list is generated based on the topic context tree; If the message to be processed is a question, then a response statement is generated based on the topic context tree.

2. The method according to claim 1, characterized in that, Determining the topic boundaries and topic logic links corresponding to the historical dialogue message set includes: Determine the degree of coherence between each historical message in the historical dialogue message set; Determine topic boundaries based on the time intervals of the historical messages; Based on the degree of closeness of connection, determine the logical link of the topic.

3. The method according to claim 1, characterized in that, The message to be processed is a multimodal message. Before determining the message intent corresponding to the message to be processed, the method further includes: The first message and the second message in the messages to be processed are determined, wherein the first message is in non-text format and the second message is in text format. The first message is processed by text conversion and extraction to obtain the corresponding third message, which is a text-formatted message. The second message and the third message are filtered out to obtain the text to be processed corresponding to the message to be processed.

4. The method according to claim 1, characterized in that, When the message to be processed is a pending message, the step of generating a pending list based on the topic context tree includes: Based on the topic network tree, determine the first topic branch related to the pending message; The pending entities in the first topic branch are marked and processed, and a structured task list is output. Based on the historical messages in the first topic branch and the pending messages, determine the elements of the pending events and generate a pending summary according to the preset rule template; A to-do list is generated based on the structured task list and the to-do summary.

5. The method according to claim 1, characterized in that, When the message to be processed is a question message, generating a response statement based on the topic context tree includes: Based on the knowledge base, the confidence level between the question message and all question indexes in the knowledge base is determined. The knowledge base stores multiple question indexes and the answers corresponding to the question indexes. Determine whether there is a situation where the confidence level is greater than the first preset confidence level; If there is a situation where the confidence level is greater than the first preset confidence level, then the response statement is determined according to the question index corresponding to the confidence level; If there is no case where the confidence level is greater than the first preset confidence level, then the matching result is determined based on the confidence level, and the matching result is the question index and answer corresponding to the maximum confidence level; Based on the topic context tree, determine the historical messages related to the question message; A response statement is generated based on the question message, the historical messages, and the matching results.

6. The method according to claim 5, characterized in that, The knowledge base was constructed in the following manner: Obtain the question information from the topic context tree and generate the semantic vector of the question information; Based on the semantic vectors, the similarity between the question information is determined; Based on the similarity, the question information is grouped, the question index of each group of question information is determined, and multiple vector spaces are determined, with one vector space corresponding to one group of question information; For one of the multiple vector spaces, the answer information of the vector space is determined according to the topic context tree; The knowledge base is generated based on the question index and the answer information.

7. The method according to claim 1, characterized in that, The method further includes: If there are multiple messages to be processed, determine the message type of each message to be processed. The priority of the messages to be processed is determined based on the message type and the preset weight, wherein the preset weight is pre-set according to the importance and urgency of the message type; The messages to be processed are pushed to the processing nodes according to the priority, and the message intent of the messages to be processed is determined in turn.

8. A message processing device, characterized in that, include: The acquisition module is used to acquire a set of historical dialogue messages and determine the topic boundaries and topic logic links corresponding to the set of historical dialogue messages. The generation module is used to generate a topic context tree based on the topic boundaries and the topic logical links; The determination module is used to determine the message intent corresponding to the message to be processed when a message to be processed is obtained. The message intent is used to indicate whether the message to be processed is a to-do message or a question message. The generation module is used to generate a to-do list based on the topic context tree if the message to be processed is a message to be done. The generation module is used to generate a response statement based on the topic context tree if the message to be processed is a question message.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 7.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 7.

Images