Session message generation method and device, storage medium and electronic equipment

Abstract

The invention relates to a session message generation method and device, a storage medium and electronic equipment, and relates to the technical field of natural language process.The method comprises the steps that a to-be-processed problem reported by a user terminal is received, and the intention of the to-be-processed problem is recognized based on a preset intention recognition model to obtain an intention recognition result; matching a historical session scene corresponding to the to-be-processed problem from a preset session database according to the intention recognition result and the to-be-processed problem; distributing a session arrangement mode for the to-be-processed problem according to the historical session scene; and generating a session message corresponding to the to-be-processed problem based on the expert arrangement mode and / or the intelligent arrangement mode, and feeding back the session message to the user terminal, so that a holder of the user terminal performs broadband fault repair according to the session message. According to the invention, the accuracy and efficiency of broadband fault diagnosis are improved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the technical field of natural language processing, and in particular, to a method for generating a conversation message, an apparatus for generating a conversation message, a computer-readable storage medium, and an electronic device. Background Art

[0002] In the existing methods of generating conversation messages, since the large model technology is not yet fully mature, there are problems such as model hallucination, which limits its reliability and effectiveness in practical applications. Therefore, the existing technology has not yet fully utilized the potential of large language models to achieve efficient and accurate home broadband fault diagnosis.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0004] The purpose of the present disclosure is to provide a method for generating a session message, a device for generating a session message, a computer-readable storage medium and an electronic device, thereby at least to a certain extent overcoming the problems of low fault diagnosis efficiency and low fault diagnosis accuracy caused by the limitations and defects of related technologies.

[0005] According to one aspect of the present disclosure, a method for generating a session message is provided, comprising:

[0006] Receive the pending problem reported by the user terminal, and identify the intent of the pending problem based on a preset intent recognition model to obtain an intent recognition result;

[0007] According to the intention recognition result and the problem to be processed, a historical conversation scenario corresponding to the problem to be processed is matched from a preset conversation database;

[0008] Allocating a conversation arrangement mode for the problem to be processed according to the historical conversation scenario; wherein the conversation arrangement mode includes an expert arrangement mode and / or an intelligent arrangement mode;

[0009] A conversation message corresponding to the problem to be processed is generated based on the expert orchestration mode and / or the intelligent orchestration mode, and the conversation message is fed back to the user terminal, so that the holder of the user terminal performs broadband fault repair according to the conversation message.

[0010] In an exemplary embodiment of the present disclosure, the intention of the problem to be processed is identified based on a preset intention identification model to obtain an intention identification result, including:

[0011] Verify the problem to be processed to determine whether the problem to be processed is a normal problem;

[0012] If the problem to be processed is a normal problem, the intention of the problem to be processed is identified based on a preset intention recognition model to obtain an intention recognition result.

[0013] In an exemplary embodiment of the present disclosure, checking the problem to be processed to determine whether the problem to be processed is a normal problem includes:

[0014] Performing word segmentation processing on the problem to be processed to obtain a first word segmentation result, and matching a target word corresponding to the first word segmentation result in a preset rule dictionary;

[0015] If the target word corresponding to the problem to be processed exists in the preset rule dictionary, then the problem to be processed is determined to be an abnormal problem;

[0016] If the target word corresponding to the problem to be processed does not exist in the preset rule dictionary, clustering the problem to be processed based on the first word segmentation result to obtain a problem clustering result;

[0017] The problem category of the problem to be processed is determined according to the problem clustering result, and whether the problem to be processed is a normal problem is determined according to the problem category.

[0018] In an exemplary embodiment of the present disclosure, determining whether the problem to be processed is a normal problem according to the problem category includes:

[0019] If the problem category of the problem to be processed is a normal category, then the problem to be processed is determined to be a normal problem; if the problem category of the problem to be processed is a malicious category, then the problem to be processed is determined to be an abnormal problem;

[0020] If the problem category of the problem to be processed is a suspected normal category and / or a suspected malicious category, the problem category to which the problem to be processed belongs is predicted based on a preset problem classification model to obtain a category prediction result, and determine whether the problem to be processed is a normal problem based on the category prediction result.

[0021] In an exemplary embodiment of the present disclosure, clustering is performed on the problem to be processed based on the first word segmentation result to obtain a problem clustering result, including:

[0022] Performing word embedding processing on the first word segmentation result based on a preset word embedding model to obtain a first text semantic feature of the problem to be processed, and calculating a cosine similarity value between the first text semantic features;

[0023] The similarity scores between the problems to be processed are determined based on the cosine similarity values, and the problems to be processed are clustered according to a preset problem clustering model and the similarity scores to obtain problem clustering results.

[0024] In an exemplary embodiment of the present disclosure, matching a historical conversation scenario corresponding to the problem to be processed from a preset conversation database according to the intention recognition result and the problem to be processed includes:

[0025] Sending the intention recognition result to the user terminal, and receiving intention discrimination feedback information corresponding to the intention recognition result fed back by the user terminal;

[0026] Determine whether the intent recognition result is accurate based on the intent discrimination feedback information, and when it is determined that the intent recognition result is accurate, match the historical conversation scene corresponding to the problem to be processed from a preset conversation database based on the intent recognition result and the problem to be processed.

[0027] In an exemplary embodiment of the present disclosure, assigning a conversation arrangement mode to the problem to be processed according to the historical conversation scenario includes:

[0028] If the historical conversation scenario exists in the preset conversation database, an expert arrangement mode is assigned to the problem to be processed; wherein the expert arrangement mode is a model generated according to a judgment rule;

[0029] If the historical conversation scenario does not exist in the preset conversation database, matching the historical topic corresponding to the problem to be processed from the preset conversation database according to the intention recognition result and the problem to be processed;

[0030] If the historical topic exists in the preset conversation database, an expert orchestration mode is assigned to the problem to be processed; if the historical topic does not exist, an intelligent orchestration mode is assigned to the problem to be processed; wherein the intelligent orchestration mode is an intelligent entity formed by one or more large models.

[0031] In an exemplary embodiment of the present disclosure, generating a conversation message corresponding to the problem to be processed based on the expert orchestration mode and / or the intelligent orchestration mode includes:

[0032] Calling an expert arrangement information library and an API information library, and executing an API information query task included in the expert arrangement information library, querying actual broadband API status information corresponding to the user terminal from the API information library;

[0033] Matching the actual broadband API status information with the preset condition rules in the expert arrangement information library to obtain a status matching result;

[0034] If the status matching result is that the actual broadband API status information is consistent with the preset condition rule, generating a session message corresponding to the problem to be processed according to the status matching result;

[0035] If the status matching result is that the actual broadband API status information is inconsistent with the preset condition rule, a conversation message corresponding to the problem to be processed is generated based on the intelligent orchestration mode.

[0036] In an exemplary embodiment of the present disclosure, generating a conversation message corresponding to the problem to be processed based on the intelligent orchestration mode includes:

[0037] Performing word embedding processing on the actual broadband API status information based on a preset word embedding model to obtain actual bandwidth features, and matching historical bandwidth features corresponding to the actual bandwidth features from a preset knowledge base;

[0038] The actual bandwidth characteristics, historical bandwidth characteristics, and the fault diagnosis results and fault repair results corresponding to the historical bandwidth characteristics are input into a preset content generation model to obtain a session message corresponding to the problem to be processed.

[0039] According to one aspect of the present disclosure, a device for generating a session message is provided, comprising:

[0040] The intention recognition module is used to receive the pending problem reported by the user terminal, and recognize the intention of the pending problem based on a preset intention recognition model to obtain an intention recognition result;

[0041] A conversation scene matching module, used to match a historical conversation scene corresponding to the problem to be processed from a preset conversation database according to the intention recognition result and the problem to be processed;

[0042] A session arrangement mode allocation module, used to allocate a session arrangement mode to the problem to be processed according to the historical session scenario; wherein the session arrangement mode includes an expert arrangement mode and / or an intelligent arrangement mode;

[0043] A session message generation module is used to generate a session message corresponding to the problem to be processed based on the expert orchestration mode and / or the intelligent orchestration mode, and feed the session message back to the user terminal so that the holder of the user terminal can repair the broadband fault according to the session message.

[0044] According to one aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for generating a session message described in any one of the above is implemented.

[0045] According to one aspect of the present disclosure, there is provided an electronic device, including:

[0046] Processor; and

[0047] A memory, configured to store executable instructions of the processor;

[0048] The processor is configured to execute any one of the above-mentioned methods for generating a session message by executing the executable instructions.

[0049] A method for generating a conversation message provided by an embodiment of the present disclosure comprises the following steps: on the one hand, receiving a pending problem reported by a user terminal, and identifying the intent of the pending problem based on a preset intent recognition model to obtain an intent recognition result; then, according to the intent recognition result and the pending problem, matching the historical conversation scenario corresponding to the pending problem from a preset conversation database; and then allocating a conversation arrangement mode to the pending problem according to the historical conversation scenario; wherein the conversation arrangement mode includes an expert arrangement mode and / or an intelligent arrangement mode; finally, generating a conversation message corresponding to the pending problem based on the expert arrangement mode and / or the intelligent arrangement mode, and feeding back the conversation message to the user terminal, so that the holder of the user terminal can repair the broadband fault according to the conversation message, thereby solving the problem in the prior art that the potential of a large language model is not fully utilized, resulting in low efficiency in broadband fault diagnosis; on the other hand, solving the problem that the accuracy of broadband fault diagnosis results is low, thereby improving the accuracy of broadband fault diagnosis results on the basis of improving the diagnostic efficiency of broadband faults.

[0050] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0052] Figure 1 The following schematically shows a flow chart of a method for generating a session message according to an exemplary embodiment of the present disclosure.

[0053] Figure 2 A schematic diagram shows an example structure of a preset question classification model according to an example embodiment of the present disclosure.

[0054] Figure 3A diagram schematically shows an example structure of a preset word embedding model according to an example embodiment of the present disclosure.

[0055] Figure 4 A scenario example diagram of a system for generating a conversation message according to an example embodiment of the present disclosure is schematically shown.

[0056] Figure 5 A diagram schematically shows an example scenario of a generated chat session message according to an example embodiment of the present disclosure.

[0057] Figure 6 A scenario example diagram schematically illustrates a generated fault diagnosis conversation message according to an example embodiment of the present disclosure.

[0058] Figure 7 The following is a block diagram schematically showing an example of a device for generating a session message according to an example embodiment of the present disclosure.

[0059] Figure 8 An electronic device for implementing a method for generating a conversation message according to an exemplary embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0060] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, the known technical solutions are not shown or described in detail to avoid obscuring the various aspects of the present disclosure.

[0061] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0062] In the field of modern home broadband operation and maintenance (or it can also be called the field of home broadband fault diagnosis), it mainly relies on on-site technicians to perform troubleshooting manually or users to perform self-troubleshooting. Specifically, in the process of manual troubleshooting, technicians usually rely on querying the status information of multiple APIs (Application Programming Interfaces) associated with the broadband account to perform troubleshooting; then, by making regular judgments on the status information of these APIs, technicians can analyze and determine the specific cause of the fault. However, with the rapid growth of home broadband services, the number of APIs that need to be queried has increased significantly, making it impractical to comprehensively list all API status information. In addition, the current query method mainly relies on the experience and knowledge of experts, so when the queried API status information does not match the preset rules, the existing technology cannot accurately diagnose the cause of the fault, nor can it provide an effective solution; further, this method also limits the interaction method with the user, such as only being able to interact through buttons, so the user-perceived usage experience is usually poor.

[0063] In order to solve the above technical problems, in some related technical solutions, home broadband faults can be diagnosed based on large language models. Specifically, with the rapid development of large language models (LLMs), agent technology has also received widespread attention; and in the autonomous agent system supported by LLMs, LLMs, as the intelligent core of agents, can be used to be responsible for key functions such as planning, memory, and tool use; in the process of practical application, the core capabilities of agent technology, especially the accuracy of planning execution, mainly rely on the performance of the underlying large model. However, the large model technology is not yet fully mature, and there are problems such as model hallucination, which limits its reliability and effectiveness in practical applications. Therefore, existing technologies have not yet fully utilized the potential of LLMs to achieve efficient and accurate home broadband fault diagnosis.

[0064] In view of the deficiencies of the prior art, the example embodiments of the present disclosure aim to provide a method for diagnosing home broadband faults based on a large language model. The method can effectively filter out non-compliant questions by reviewing multiple contents; at the same time, through the intention recognition model based on the large model, the scene intention of the user's question can be accurately identified, and the interaction ability with the user can be effectively improved; by combining expert arrangement and intelligent arrangement, while ensuring accuracy and stability, the powerful logical reasoning ability of the large model can be relied on to improve the overall flexibility and generalization ability of the system. Based on the above-mentioned contents, it can be known that the example embodiments of the present disclosure can not only solve the problems existing in the prior art, but also provide users with more efficient, economical, reliable and flexible broadband operation and maintenance solutions through intelligent means, which has significant practical value and broad market prospects.

[0065] In this exemplary embodiment, a method for generating a session message is first provided, which can be run on a server, a server cluster, or a cloud server, etc. Of course, those skilled in the art can also run the method disclosed in this disclosure on other platforms as required, which is not specifically limited in this exemplary embodiment. Figure 1 As shown, the method for generating the session message may include the following steps:

[0066] Step S110: receiving the pending problem reported by the user terminal, and identifying the intention of the pending problem based on a preset intention recognition model to obtain an intention recognition result;

[0067] Step S120. Matching a historical conversation scenario corresponding to the problem to be processed from a preset conversation database according to the intention recognition result and the problem to be processed;

[0068] Step S130. Assign a conversation arrangement mode to the problem to be processed according to the historical conversation scenario; wherein the conversation arrangement mode includes an expert arrangement mode and / or an intelligent arrangement mode;

[0069] Step S140: Generate a conversation message corresponding to the problem to be processed based on the expert orchestration mode and / or the intelligent orchestration mode, and feed back the conversation message to the user terminal, so that the holder of the user terminal can repair the broadband fault according to the conversation message.

[0070] In the above-described method for generating a session message, on the one hand, by receiving the pending problem reported by the user terminal and identifying the intent of the pending problem based on a preset intent recognition model, an intent recognition result is obtained; then, according to the intent recognition result and the pending problem, a historical session scenario corresponding to the pending problem is matched from a preset session database; and then a session orchestration mode is assigned to the pending problem according to the historical session scenario; wherein the session orchestration mode includes an expert orchestration mode and / or an intelligent orchestration mode; finally, a session message corresponding to the pending problem is generated based on the expert orchestration mode and / or the intelligent orchestration mode, and the session message is fed back to the user terminal, so that the holder of the user terminal can repair the broadband fault according to the session message, thereby solving the problem in the prior art that the potential of the large language model is not fully utilized, resulting in low efficiency in broadband fault diagnosis; on the other hand, the problem of low accuracy of broadband fault diagnosis results is also solved, thereby improving the accuracy of broadband fault diagnosis results on the basis of improving the diagnostic efficiency of broadband faults.

[0071] Hereinafter, the method for generating a session message according to an exemplary embodiment of the present disclosure will be explained and illustrated in detail with reference to the accompanying drawings.

[0072] First, the terms involved in the exemplary embodiments of the present disclosure are explained and described. Specifically:

[0073] Large Model: Large Model usually refers to a large-scale neural network model trained by deep learning, such as GPT (Generative Pre-trained Transformer), Bert (Bidirectional Encoder Representations from Transformers), RoBerta (Optimized Bert), and other similar models.

[0074] Intelligent agent: An intelligent agent is a computer program based on a large language model that has the ability to plan and think, remember, and use tool functions, and can autonomously complete given tasks.

[0075] RAG: The RAG (Retrieval-Augmented Generation) module is an enhanced generation model specifically designed to process information in large-scale knowledge bases. In practical applications, RAG combines large content generation models (such as GPT) and information retrieval technology to improve the model's information understanding and generation capabilities in dealing with open-domain problems and dialogues.

[0076] Prompt: refers to a snippet of input text or code used to guide the model to generate a response. In practical applications, in order to improve the accuracy of the output results of large models, carefully constructed text prompts (Prompt) can be designed to guide generative models (such as GPT) to generate specific replies or behaviors.

[0077] Conversation orchestration mode: The conversation orchestration mode recorded in the example embodiments of the present disclosure may include an expert orchestration mode and an intelligent orchestration mode; wherein the expert orchestration mode is an orchestration based on judgment rules, in which provincial business experts sort out problem troubleshooting experience to form a fixed pattern of orchestration execution steps; the intelligent orchestration mode is an intelligent orchestration based on the reasoning capability of a large model, in which the execution steps and calling tools are determined by the large model; in the actual application process, the expert orchestration mode has the characteristics of high accuracy and stable execution, but has disadvantages such as poor generalization ability to solve problems and poor interactive perception experience; on the contrary, the intelligent orchestration mode has the advantages of good generalization ability and good user interactive perception experience, but intelligent orchestration is based on the reasoning capability of a large model, and has natural disadvantages in terms of accuracy and stability; therefore, in order to achieve the purpose of good generalization ability on the basis of ensuring accuracy and stability, the advantages of expert orchestration and intelligent orchestration can be combined and used in turn.

[0078] Secondly, the technical implementation principle of the example embodiment of the present disclosure is explained and illustrated. Specifically, the method for generating a session message recorded in the example embodiment of the present disclosure can solve the technical limitations existing in the existing field of home broadband fault diagnosis; in the process of actual application, by providing an innovative method for generating a session message, the accuracy and efficiency of fault diagnosis can be improved, and the dependence on expert experience and knowledge can be reduced. In addition, the example embodiment of the present disclosure is also intended to improve the user interaction experience, by providing a more intuitive and friendly user interface, so that users can more easily perceive and participate in the fault diagnosis process; further, by utilizing large language models (LLM) and Agent technology, the example embodiment of the present disclosure will improve the intelligence level of broadband fault diagnosis, thereby bringing significant economic and social benefits to enterprises.

[0079] The following will explain and illustrate the preset question classification model involved in the exemplary embodiment of the present disclosure. Figure 2 As shown, the preset question classification model may include a first input layer 210, a first embedding mapping layer 220, a first encoding layer 230, a first hybrid expert model layer 240 and a first output layer 250; wherein, the functions required to be performed by each model layer in the question classification process will be described in detail later and will not be further elaborated here.

[0080] The following will explain and illustrate the preset word embedding model involved in the exemplary embodiment of the present disclosure. Figure 3 As shown, the preset word embedding model recorded here can be a Bert model, and the Bert model can include a second input layer 310, an encoding module 320 composed of multiple Transformers, and a second output layer 330. Among them, the functions that each model layer needs to perform in the problem classification process will be described in detail later, and no further details will be given here.

[0081] The following will explain and illustrate the system for generating a session message involved in the exemplary embodiment of the present disclosure. Figure 4 As shown, the session message generation system may include a user terminal 410 and a server 420; wherein the user terminal may be connected to the server via a wired network or a wireless network; in actual application, the user terminal may be used to report pending issues to the server and receive session messages fed back by the server; the server may be used to implement the session message generation method described in the exemplary embodiment of the present disclosure.

[0082] The following will be combined Figure 2-Figure 4 right Figure 1 The method for generating the session message shown in the example is further explained and illustrated. Specifically:

[0083] In step S110, the pending problem reported by the user terminal is received, and the intention of the pending problem is identified based on a preset intention recognition model to obtain an intention recognition result.

[0084] In this example embodiment, first, an unprocessed problem reported by a user terminal is received; wherein, the unprocessed problem recorded here may be input by the user to the user terminal by voice or text, and then reported by the user terminal; wherein, the unprocessed problem recorded here may be, for example: "Why can't I access the Internet?" or "Is my mobile phone broken?" and so on; secondly, after receiving the unprocessed problem, the intent of the unprocessed problem can be identified based on a preset intent recognition model to obtain an intent recognition result; specifically, the specific implementation process of intent recognition can be achieved in the following way: first, the unprocessed problem is verified to determine whether the unprocessed problem is a normal problem; secondly, if the unprocessed problem is a normal problem, the intent of the unprocessed problem is identified based on a preset intent recognition model to obtain an intent recognition result.

[0085] In an exemplary embodiment, the problem to be processed is checked to determine whether the problem to be processed is a normal problem, which can be achieved in the following manner: perform word segmentation processing on the problem to be processed to obtain a first word segmentation result, and match the target word corresponding to the first word segmentation result in a preset rule dictionary; if the target word corresponding to the problem to be processed exists in the preset rule dictionary, the problem to be processed is determined to be an abnormal problem; if the target word corresponding to the problem to be processed does not exist in the preset rule dictionary, clustering the problem to be processed based on the first word segmentation result to obtain a problem clustering result; determine the problem category of the problem to be processed according to the problem clustering result, and determine whether the problem to be processed is a normal problem according to the problem category.

[0086] In an exemplary embodiment, determining whether a problem to be processed is a normal problem based on the problem category can be achieved in the following manner: if the problem category of the problem to be processed is a normal category, determining that the problem to be processed is a normal problem; if the problem category of the problem to be processed is a malicious category, determining that the problem to be processed is an abnormal problem; if the problem category of the problem to be processed is a suspected normal category and / or a suspected malicious category, predicting the problem category to which the problem to be processed belongs based on a preset problem classification model to obtain a category prediction result, and determining whether the problem to be processed is a normal problem based on the category prediction result.

[0087] In an exemplary embodiment, clustering the problems to be processed based on the first word segmentation result to obtain problem clustering results can be achieved in the following manner: performing word embedding processing on the first word segmentation result based on a preset word embedding model to obtain the first text semantic features of the problems to be processed, and calculating the cosine similarity values ​​between the first text semantic features; determining the similarity scores between the problems to be processed based on the cosine similarity values, and clustering the problems to be processed according to a preset problem clustering model and the similarity scores to obtain problem clustering results.

[0088] The intent recognition process will be further explained and illustrated below. Specifically, in the actual application process, after receiving the problem to be processed, it is first necessary to conduct a content review on the problem to be processed; specifically, in the process of content review, first of all, it is necessary to pre-process the problem to be processed; that is, it is necessary to remove irrelevant characters, stop words, punctuation marks, etc.; then, the pre-processed problem to be processed is segmented to obtain a first segmentation result; wherein, the segmentation process recorded here can be implemented based on corresponding segmentation tools (such as word2vec), or based on corresponding segmentation models, and this example does not impose special restrictions on this; the first segmentation result recorded here may include discrete words or phrase units; secondly, a search is performed in the preset rule dictionary. If the first segmentation result involves relevant content in the rule dictionary (that is, it matches the target word), then it is determined. The problem to be processed is an abnormal problem and the relevant words of the user terminal are returned for a second question; if the first word segmentation result does not involve relevant content in the rule dictionary (that is, the target word is not matched), the user question (that is, the first word segmentation result) can be semantically clustered to analyze the category of the problem to be processed; further, if the question category is a normal category, the subsequent intent recognition step is executed; if it is a malicious category, the problem to be processed is determined to be an abnormal problem and the relevant words of the user terminal are returned for a second question; if it is a suspected malicious category or a suspected normal category, it will be reviewed and verified through a large model (that is, a question classification model), and the user question will be finally verified using the powerful logical reasoning ability of the large model; further, if it is a malicious category, the first word segmentation result corresponding to the problem to be processed can also be stored in the database for updating the rule dictionary.

[0089] In an example embodiment, the rule dictionary recorded above can be constructed in the following manner: collect data related to sensitive topic types commonly found on the Internet to construct the rule dictionary. At the same time, since the number of this part of the dictionary may be limited, the big model can be used to expand the rule dictionary; in the process of expanding the rule dictionary, the synonyms or near synonyms of the rule dictionary can be obtained by constructing relevant prompts (Prompt), so as to combine the two parts of the dictionary to complete the construction of the rule dictionary. Among them, the construction of the rule dictionary can be achieved in the following ways: ① Data collection: collect common sensitive topic words from channels such as the Internet and social media to establish a preliminary rule dictionary; ② Big model expansion: use the reasoning ability of the big model to generate synonyms and near synonyms by constructing prompts to expand the coverage of the existing dictionary; at the same time, the big model recorded here can be a big model based on Bert; ③ Dictionary integration: integrate the collected sensitive words with the synonyms and near synonyms generated by the big model to form a comprehensive rule dictionary. Based on this method, not only the richness and accuracy of the rule dictionary are improved, but also it can adapt to the ever-changing language environment and social needs, thereby achieving the purpose of improving the accuracy of the verification results of the problems to be processed.

[0090] In an example embodiment, in the process of clustering the problems to be processed, first, a pre-trained language model (i.e., a preset word embedding model) is used to map the user text (i.e., the first word segmentation result) to a high-dimensional vector space as a means of feature extraction, where each vector represents the semantic feature of the text (the first text semantic feature); then, the cosine similarity between the first text semantic vectors is calculated to evaluate the similarity between them, and the metric is based on the cosine value of the vector angle, thereby obtaining a similarity matching score. Finally, the DBSCAN (Density-Based SpatialClustering of Applications with Noise) clustering algorithm is used to classify user questions, which can identify clusters of any shape and divide user questions into three categories: normal, suspected, and malicious.

[0091] In an example embodiment, if the obtained category is a suspected normal category or a suspected malicious category, further audit verification can be performed based on the large model. In the process of auditing and verifying the large model, you can first build a prompt for content auditing, and use the reasoning ability of the large model to realize the audit and verification of user questions; at the same time, when building the prompt, you can add a prefix to guide the behavior of the model, for example: "You are a content audit expert, mainly responsible for auditing and identifying user questions to avoid sensitive information in user questions." In addition, you can also add restrictive instructions after the prompt to standardize the output of the model, for example: "Your answer results are limited to two categories: 'normal' and 'malicious', and no other additional descriptions are allowed." And so on. In the actual application process, you can set the prompt information according to your actual needs, and this example does not impose special restrictions on this. By adding restrictive conditions for relevant prompts, the verification ability of the large model in content auditing can be effectively improved; further, if the verification result is "malicious", the user's question will be used to retrain the semantic matching model to enhance the model's ability to identify sensitive information.

[0092] In an example embodiment, in the process of predicting the problem category to which the problem to be processed belongs based on a preset problem classification model to obtain a category prediction result, it can be achieved in the following way: first, context information is generated according to the problem to be processed and the prompt information; secondly, the context information is embedded and mapped based on the first embedding mapping layer to obtain a context marker sequence; then, the context marker sequence is encoded based on the first encoding layer to obtain a context overall representation, and the context overall representation is predicted based on the hybrid expert model layer to obtain a category prediction result.

[0093] In an exemplary embodiment, if the problem to be processed is a normal problem, the intention of the problem to be processed is identified based on a preset intention recognition model to obtain an intention recognition result. Specifically, in the process of intention recognition, the question raised by the user can be deeply analyzed to identify the specific intention behind it; after obtaining the intention recognition result, the information such as the problem to be processed, the intention recognition result and the answer of the intelligent agent (that is, the conversation message) can be stored in the database as long-term memory to support multi-round conversation management and tracing of historical information. The specific implementation process can be achieved in the following way: first, receive the user question (that is, the problem to be processed), and use natural language processing technology, such as a large language model, to determine the user's intention; secondly, store the historical record of the user input, the historical result of the intention recognition and the answer history of the intelligent agent, based on this method, context awareness and conversation coherence can be achieved; at the same time, by performing intention recognition and storing the corresponding conversation message, the continuity of the user conversation can be effectively maintained, and historical information can be provided to assist the conversation when necessary.

[0094] In an example embodiment, during the training of the intent recognition model, first, a data set needs to be prepared; wherein, the data set recorded here should include user questions and their corresponding intent labels; at the same time, these data can be collected from multiple sources, including but not limited to business experts and business developers; after the data is collected, it needs to be cleaned and preprocessed to ensure data quality; after the preprocessing is completed, the training data can be annotated, including user questions and their corresponding intent labels; in the actual application process, the data can be organized in JSON format, including two fields: user questions and intent labels; in the process of model training, a training data set is obtained by constructing a user question intent QA pair, and then the Lora algorithm is used to fine-tune the large model based on the training data set to obtain the intent recognition model. In the process of fine-tuning the large model, the low-rank adaptation model can be trained based on the training data set to obtain the corresponding low-rank matrix parameters, and then the large model is fine-tuned based on the low-rank matrix parameters to obtain the intent recognition model; when the user question passes the content review, the intent recognition model will analyze the user's intent.

[0095] In step S120, historical conversation scenarios corresponding to the question to be processed are matched from a preset conversation database according to the intention recognition result and the question to be processed.

[0096] Specifically, the specific matching process of the historical conversation scene can be implemented in the following manner: sending the intention recognition result to the user terminal, and receiving the intention discrimination feedback information corresponding to the intention recognition result fed back by the user terminal; determining whether the intention recognition result is accurate according to the intention discrimination feedback information, and when determining that the intention recognition result is accurate, matching the historical conversation scene corresponding to the problem to be processed from the preset conversation database according to the intention recognition result and the problem to be processed. That is, in the actual application process, after the specific intention (that is, the intention recognition result) is inferred, the intention recognition result will be fed back to the user, and the user will confirm or judge whether the recognized intention is correct; then, according to the user's feedback, the user question and the wrong intention with the intention recognition error will be saved to the intention recognition database, and the intention recognition QA pair will be automatically generated to construct a training data set, and the intention recognition large model will be automatically iteratively updated to continuously improve the accuracy and precision of the intention recognition unit; then, after the accuracy of the intention is confirmed, the TOP3 closest conversations in the historical conversation can be retrieved from the database according to the problem to be processed and the intention recognition result, and the historical conversation scene of the problem to be processed is determined based on the closest conversation.

[0097] In step S130, a conversation orchestration mode is allocated to the problem to be processed according to the historical conversation scenario; wherein the conversation orchestration mode includes an expert orchestration mode and / or an intelligent orchestration mode.

[0098] Specifically, the specific allocation process of the conversation arrangement mode can be implemented in the following manner: if the historical conversation scenario exists in the preset conversation database, the expert arrangement mode is allocated to the problem to be processed; wherein the expert arrangement mode is a model generated according to the judgment rule; if the historical conversation scenario does not exist in the preset conversation database, the historical topic corresponding to the problem to be processed is matched from the preset conversation database according to the intention recognition result and the problem to be processed; if the historical topic exists in the preset conversation database, the expert arrangement mode is allocated to the problem to be processed, and if the historical topic does not exist, the intelligent arrangement mode is allocated to the problem to be processed; wherein the intelligent arrangement mode is an intelligent agent formed by one or more large models. That is, in actual business scenarios, users may ask some questions that are not related to the business in the process of calling the intelligent agent; for such questions, the intention recognition unit will identify them as "chatting"; in addition, in the process of multiple interactions between the intelligent agent and the user, the user will feedback some answers that are related to the interaction of the historical business scenario but are intentionally identified as chatting, and the intention recognition module will still identify them as "chatting". For this type of question, you cannot directly enter the chat mode, which will greatly reduce the user experience perception, so it is necessary to add an orchestration judgment step; specifically, the purpose of the orchestration judgment step is to match the user question intention recognition result with the known scenario library. If the user intention matches the scenario library successfully (that is, there is a historical conversation scenario), it is considered that the user is consulting related business and enters the expert orchestration module; if there is no match, it enters the intelligent orchestration mode; in this mode, first determine whether there is a historical topic, if so, assign the expert orchestration mode; otherwise, assign the intelligent orchestration mode.

[0099] In step S140, a session message corresponding to the problem to be processed is generated based on the expert orchestration mode and / or the intelligent orchestration mode, and the session message is fed back to the user terminal, so that the holder of the user terminal can perform broadband fault repair according to the session message.

[0100] In this example embodiment, first, a session message is generated; specifically, it can be achieved in the following manner: calling an expert scheduling information library and an API information library, and executing an API information query task included in the expert scheduling information library, querying the actual broadband API status information corresponding to the user terminal from the API information library; matching the actual broadband API status information and the preset condition rules in the expert scheduling information library to obtain a status matching result; if the status matching result is that the actual broadband API status information is consistent with the preset condition rules, then generating a session message corresponding to the problem to be processed according to the status matching result; if the status matching result is that the actual broadband API status information is inconsistent with the preset condition rules, then generating a session message corresponding to the problem to be processed based on the intelligent scheduling mode.

[0101] In an exemplary embodiment, generating a conversation message corresponding to the problem to be processed based on the intelligent orchestration mode can be achieved in the following manner: performing word embedding processing on the actual broadband API status information based on a preset word embedding model to obtain actual bandwidth characteristics, and matching historical bandwidth characteristics corresponding to the actual bandwidth characteristics from a preset knowledge base; inputting the actual bandwidth characteristics, historical bandwidth characteristics, and fault diagnosis results and fault repair results corresponding to the historical bandwidth characteristics into a preset content generation model to obtain a conversation message corresponding to the problem to be processed.

[0102] The specific process of generating the session message will be further explained and illustrated below.

[0103] First, the expert arrangement information library and API information library recorded in the exemplary embodiment of the present disclosure are explained and illustrated. (1) The expert arrangement information library recorded here is mainly used to store arrangement information; wherein, the arrangement information recorded here is mainly composed of various branches, each branch includes branch name, branch code, query API, user interaction, conditional rules, execution logic, whether to beautify the speech and the next branch, etc.; at the same time, the execution logic recorded here may include but is not limited to suggested text information, output type (button, picture and link, etc.), execution function item name, function item content, executor and other information; in the actual process of generating conversation messages, after calling the expert information arrangement library, it can be judged whether the current branch meets the requirements according to the conditional rules; if the conditions are met, the API and user interaction information can be queried according to the execution items in the execution logic, and the query status API and the interaction with the user can be completed. The next operation can be directly and accurately pointed to according to the next branch. For example, the query API (that is, the API information query task) can be executed based on the execution logic to obtain the actual broadband API status information, and of course, the purpose of user interaction can also be achieved.

[0104] (2) The API information library recorded here is mainly used to store API related information when calling the capability platform, which may include but is not limited to API basic information, contract information, header, input parameter table, request example and output parameter table; at the same time, after calling the API information library, the API basic information can be obtained, and the relevant basic information can be parsed to obtain the input parameters, output parameters, call header and other parameters when calling the API, thereby completing the API call and output parameter parsing.

[0105] Secondly, further explanation and description are given on how to generate session messages based on the expert orchestration information library and the API information library. Specifically, first, determine whether it is necessary to call the capability platform based on whether there is a value in the query API field in the expert orchestration information library; if there is a value in the query API field, call the API information library to find the input parameters, output parameters, Header and other parameters of the API, and obtain the actual broadband API status information by calling the capability platform, and match the actual broadband API status information with the conditional rules in the branch name corresponding to the next branch. If the rules match, enter the relevant branch and continue to execute the next capability platform call or interact with the user; if there is no match, it means that the actual broadband API status information queried conflicts with the conditional rules written by the user, and the expert orchestration cannot meet the requirements. At this time, it is necessary to enter the intelligent orchestration mode and use large model reasoning to complete task planning and execution operations.

[0106] In an example embodiment, when the pending question is judged as an unknown scenario (that is, there is no historical conversation scenario) after intent recognition, the pending question will enter the intelligent orchestration mode; in the intelligent orchestration mode, the semantic relevance will first be judged in combination with the user's current question and the TOP3 in the historical conversation. If the current question is related to the historical conversation, it will be judged as a historical topic, and the user question will be topic-changed, and the intelligent orchestration mode will be converted to the expert orchestration mode, thereby realizing flexible conversion between intelligent orchestration and expert orchestration; further, if the user question is not a historical topic, it will be judged as chat, and the chat prompt will be directly constructed through the beautification and interaction module, and the large model will be called to generate chat words to obtain the conversation message; among which, the scene example diagram of the obtained chat conversation message can be referred to Figure 5 shown.

[0107] In an example embodiment, in the expert orchestration mode, when the actual broadband API status information queried does not match the preset condition rules, the intelligent orchestration mode will be entered; in the intelligent orchestration mode, it is first necessary to build a knowledge base for different scenarios based on RAG, and the knowledge base is stored in the format of QA; in this knowledge base, the status and diagnosis results of each API are taken as Q, and the corresponding solutions or steps are taken as A; and it is also necessary to convert the knowledge base into a vector library through the Embedding model and store it in the server as long-term memory; wherein, the Embedding model recorded here can be a Bert model; at the same time, a scenario-API mapping management library is also constructed. When the user's specific intent (that is, the intent recognition result) is obtained, the API name can be obtained based on the mapping relationship between the intent recognition result and the API; further, when the obtained After finding the specific API name, the capability platform needs to be called to query the actual bandwidth API status information, and based on the queried status information, the embedding model is used to search from the knowledge base, and the reranking model is used to re-rank, so as to obtain the retrieval recall results ranked as TOP3 (that is, the historical bandwidth characteristics); then, the actual API status information is queried and combined with the results retrieved from the knowledge base to build a new task planning prompt, and the large model is used to complete the detection task planning and execution; further, after calling the relevant API, the LLM reasoning capability is used to perform fault diagnosis and location, so as to determine the specific diagnosis result; further, the fault repair prompt can be built through the knowledge information retrieved from the knowledge base, and the large model can complete the specific fault repair function based on the fault diagnosis and location results, so as to generate the following: Figure 6 The fault diagnosis and repair session message shown in the figure. After completing the fault repair function, by building a status repair confirmation prompt, you can call the relevant API again, analyze the API status by retrieving the knowledge base, and complete the final status repair.

[0108] It is also necessary to further explain that in the actual application process, the expert orchestration mode and the intelligent orchestration mode can be converted to each other based on the method: first, the historical conversation scene corresponding to the problem to be processed is matched; if the historical conversation scene exists, the conversation message is generated based on the expert orchestration mode; if the historical conversation scene does not exist, the conversation message is generated based on the intelligent orchestration mode;

[0109] Secondly, in the expert orchestration mode, it is necessary to obtain the actual broadband API status information and determine whether the actual broadband API status information matches the preset condition rules; if it matches, a session message is directly generated; if it does not match, the intelligent orchestration mode is entered;

[0110] Furthermore, if the historical conversation scenario does not exist, it is necessary to match the historical topic corresponding to the problem to be processed in the intelligent orchestration mode; if the historical topic exists, it is switched to the expert orchestration mode and the matching step of the actual broadband API status information and the preset condition rules is performed; if the historical topic does not exist, it is determined to be a chat, and it is necessary to call the content generation model to generate a chat session message;

[0111] Furthermore, if the actual broadband API status information does not match the preset condition rules, it is necessary to obtain the API name based on the mapping relationship between the intent recognition result and the API; further, after obtaining the specific API name, it is necessary to call the capability platform to query the actual bandwidth API status information, and retrieve from the knowledge base based on the queried status information by using the embedding model, and rerank using the Reranking model to obtain the TOP3 retrieval recall results (that is, historical bandwidth features); then, the queried actual API status information is combined with the results retrieved from the knowledge base to construct a new task planning prompt, and the large model is used to complete the detection task planning and execution; further, after calling the relevant API, the LLM reasoning capability is used to perform fault diagnosis and location, so as to determine the specific diagnosis result and generate the corresponding fault diagnosis session message.

[0112] Finally, in the interaction with users, some words have rigorous attributes, and the original meaning of the words may be changed when the big model is used to generate related words. Therefore, it is necessary to add a "whether to beautify the words" label in the orchestration library; in the process of beautification and interaction, first determine whether to beautify the words. If the words do not need to be beautify, the configured related words will be given to the content review module for compliance review; after the review is passed, it can be fed back to the user terminal; if the words need to be beautify, then by setting prompts for different roles, the big model can generate various types of words such as professional and lively to improve the interactive experience between the intelligent agent and the user, and feed back the beautified conversation message to the user terminal.

[0113] So far, the method for generating a session message described in the exemplary embodiments of the present disclosure has been fully implemented. Based on the above-mentioned contents, it can be known that the method for generating conversation messages recorded in the example embodiment of the present disclosure has at least the following advantages: on the one hand, by setting up triple audits of rule dictionaries, semantic clustering and large model audit verification, illegal information can be effectively filtered out, thereby improving the overall security compliance of the intelligent body; in addition, user questions that have not passed the large model audit verification will be used to retrain the semantic matching model. By automatically and dynamically updating the training semantic clustering, the precision and accuracy of the semantic clustering model can be continuously improved, and ultimately the accuracy of the content audit module can be improved; on the other hand, by constructing user question-intent label QA pairs and fine-tuning the large model using the lora algorithm, an intent recognition large model can be obtained; and, according to the user feedback mechanism, an intent recognition QA pair training data set can be automatically generated, and the intent recognition large model can be automatically iterated and updated, thereby continuously improving the accuracy and precision of the intent recognition unit; on the other hand, the example embodiment of the present disclosure also provides a fusion solution of expert orchestration and intelligent orchestration. Expert orchestration uses an orchestration library and an API information library to ensure that business experts effectively and timely maintain professional information. Intelligent orchestration uses retrieval enhancement generation and the deep reasoning capabilities of the large model to ensure the accuracy of the generated solution and reduce model hallucinations. This fusion solution has the advantages of good generalization ability and good user interaction perception experience while ensuring accuracy and stability. It can automatically identify and solve broadband failures, thereby significantly reducing dependence on manual operation and maintenance.

[0114] The following are embodiments of the device disclosed herein, which can be used to execute the method embodiments disclosed herein. For details not disclosed in the device embodiments disclosed herein, please refer to the method embodiments disclosed herein.

[0115] The exemplary embodiment of the present disclosure also provides a device for generating a session message. Specifically, refer to Figure 7 As shown, the conversation message generation device may include an intention recognition module 710, a conversation scenario matching module 720, a conversation arrangement mode allocation module 730, and a conversation message generation module 740. Among them:

[0116] The intention recognition module 710 may be used to receive the pending problem reported by the user terminal, and recognize the intention of the pending problem based on a preset intention recognition model to obtain an intention recognition result;

[0117] The conversation scene matching module 720 may be used to match the historical conversation scene corresponding to the problem to be processed from a preset conversation database according to the intention recognition result and the problem to be processed;

[0118] The conversation arrangement mode allocation module 730 may be used to allocate a conversation arrangement mode to the problem to be processed according to the historical conversation scenario; wherein the conversation arrangement mode includes an expert arrangement mode and / or an intelligent arrangement mode;

[0119] The session message generation module 740 can be used to generate a session message corresponding to the problem to be processed based on the expert orchestration mode and / or the intelligent orchestration mode, and feed the session message back to the user terminal so that the holder of the user terminal can repair the broadband fault according to the session message.

[0120] In an exemplary embodiment of the present disclosure, the intent of the problem to be processed is identified based on a preset intention recognition model to obtain an intention recognition result, including: verifying the problem to be processed to determine whether the problem to be processed is a normal problem; if the problem to be processed is a normal problem, identifying the intent of the problem to be processed based on the preset intention recognition model to obtain an intention recognition result.

[0121] In an exemplary embodiment of the present disclosure, the problem to be processed is verified to determine whether the problem to be processed is a normal problem, including: performing word segmentation processing on the problem to be processed to obtain a first word segmentation result, and matching a target word corresponding to the first word segmentation result in a preset rule dictionary; if the target word corresponding to the problem to be processed exists in the preset rule dictionary, determining that the problem to be processed is an abnormal problem; if the target word corresponding to the problem to be processed does not exist in the preset rule dictionary, clustering the problem to be processed based on the first word segmentation result to obtain a problem clustering result; determining the problem category of the problem to be processed according to the problem clustering result, and determining whether the problem to be processed is a normal problem according to the problem category.

[0122] In an exemplary embodiment of the present disclosure, determining whether the problem to be processed is a normal problem is based on the problem category, including: if the problem category of the problem to be processed is a normal category, determining that the problem to be processed is a normal problem; if the problem category of the problem to be processed is a malicious category, determining that the problem to be processed is an abnormal problem; if the problem category of the problem to be processed is a suspected normal category and / or a suspected malicious category, predicting the problem category to which the problem to be processed belongs based on a preset problem classification model to obtain a category prediction result, and determining whether the problem to be processed is a normal problem based on the category prediction result.

[0123] In an exemplary embodiment of the present disclosure, clustering processing is performed on the questions to be processed based on the first word segmentation result to obtain question clustering results, including: performing word embedding processing on the first word segmentation result based on a preset word embedding model to obtain the first text semantic features of the questions to be processed, and calculating the cosine similarity values ​​between the first text semantic features; determining the similarity scores between the questions to be processed based on the cosine similarity values, and clustering the questions to be processed according to a preset question clustering model and the similarity scores to obtain question clustering results.

[0124] In an exemplary embodiment of the present disclosure, historical conversation scenarios corresponding to the problem to be processed are matched from a preset conversation database based on the intention recognition result and the problem to be processed, including: sending the intention recognition result to the user terminal, and receiving intention discrimination feedback information corresponding to the intention recognition result fed back by the user terminal; determining whether the intention recognition result is accurate based on the intention discrimination feedback information, and when it is determined that the intention recognition result is accurate, matching the historical conversation scenarios corresponding to the problem to be processed from a preset conversation database based on the intention recognition result and the problem to be processed.

[0125] In an exemplary embodiment of the present disclosure, a conversation orchestration mode is assigned to the problem to be processed according to the historical conversation scenario, including: if the historical conversation scenario exists in the preset conversation database, an expert orchestration mode is assigned to the problem to be processed; wherein the expert orchestration mode is a model generated according to a judgment rule; if the historical conversation scenario does not exist in the preset conversation database, a historical topic corresponding to the problem to be processed is matched from the preset conversation database according to the intention recognition result and the problem to be processed; if the historical topic exists in the preset conversation database, an expert orchestration mode is assigned to the problem to be processed, and if the historical topic does not exist, an intelligent orchestration mode is assigned to the problem to be processed; wherein the intelligent orchestration mode is an intelligent entity formed by one or more large models.

[0126] In an exemplary embodiment of the present disclosure, a session message corresponding to the problem to be processed is generated based on the expert orchestration mode and / or the intelligent orchestration mode, including: calling an expert orchestration information library and an API information library, and executing an API information query task included in the expert orchestration information library, querying the actual broadband API status information corresponding to the user terminal from the API information library; matching the actual broadband API status information and the preset condition rules in the expert orchestration information library to obtain a status matching result; if the status matching result is that the actual broadband API status information is consistent with the preset condition rules, then generating a session message corresponding to the problem to be processed according to the status matching result; if the status matching result is that the actual broadband API status information is inconsistent with the preset condition rules, then generating a session message corresponding to the problem to be processed based on the intelligent orchestration mode.

[0127] In an exemplary embodiment of the present disclosure, a conversation message corresponding to the problem to be processed is generated based on the intelligent orchestration mode, including: performing word embedding processing on the actual broadband API status information based on a preset word embedding model to obtain actual bandwidth features, and matching historical bandwidth features corresponding to the actual bandwidth features from a preset knowledge base; inputting the actual bandwidth features, historical bandwidth features, and fault diagnosis results and fault repair results corresponding to the historical bandwidth features into a preset content generation model to obtain a conversation message corresponding to the problem to be processed.

[0128] The specific details of each module in the above-mentioned session message generation device have been described in detail in the corresponding session message generation method, so they are not repeated here.

[0129] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0130] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0131] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided. A person skilled in the art will appreciate that various aspects of the present disclosure may be implemented as a system, method, or program product. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects, which may be collectively referred to herein as a circuit, module, or system.

[0132] Refer to the following Figure 8 The electronic device 800 according to this embodiment of the present disclosure is described. Figure 8 The electronic device 800 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0133] like Figure 8 As shown, the electronic device 800 is in the form of a general computing device. The components of the electronic device 800 may include, but are not limited to: the at least one processing unit 810, the at least one storage unit 820, a bus 830 connecting different system components (including the storage unit 820 and the processing unit 810), and a display unit 840.

[0134] The storage unit stores program codes, which can be executed by the processing unit 810, so that the processing unit 810 performs the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 810 can perform the following steps: Figure 1 The step S110 shown in the figure is as follows: receiving the pending problem reported by the user terminal, and identifying the intent of the pending problem based on a preset intent recognition model to obtain an intent recognition result; step S120: matching the historical conversation scenario corresponding to the pending problem from a preset conversation database according to the intent recognition result and the pending problem; step S130: allocating a conversation orchestration mode to the pending problem according to the historical conversation scenario; wherein the conversation orchestration mode includes an expert orchestration mode and / or an intelligent orchestration mode; step S140: generating a conversation message corresponding to the pending problem based on the expert orchestration mode and / or the intelligent orchestration mode, and feeding back the conversation message to the user terminal, so that the holder of the user terminal performs broadband fault repair according to the conversation message.

[0135] The storage unit 820 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 8201 and / or a cache memory unit 8202 , and may further include a read-only memory unit (ROM) 8203 .

[0136] The storage unit 820 may also include a program / utility 8204 having a set (at least one) of program modules 8205, such program modules 8205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0137] Bus 830 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0138] The electronic device 800 may also communicate with one or more external devices 900 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 800, and / or communicate with any device that enables the electronic device 800 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 850. Furthermore, the electronic device 800 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 860. As shown, the network adapter 860 communicates with other modules of the electronic device 800 via a bus 830. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0139] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0140] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary implementations of the present disclosure described in the above "Exemplary Method" section of the present specification.

[0141] A program product for implementing the above method according to an embodiment of the present disclosure is described, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, apparatus, or device.

[0142] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0143] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0144] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0145] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0146] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0147] Other embodiments of the present disclosure will be readily apparent to those skilled in the art after considering the specification and practicing the inventions invented herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not invented by the present disclosure. The specification and examples are to be considered merely exemplary, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A method for generating a session message, characterized in that: include: Receive the pending problem reported by the user terminal, and identify the intent of the pending problem based on a preset intent recognition model to obtain an intent recognition result; According to the intention recognition result and the problem to be processed, a historical conversation scenario corresponding to the problem to be processed is matched from a preset conversation database; Allocating a conversation arrangement mode for the problem to be processed according to the historical conversation scenario; wherein the conversation arrangement mode includes an expert arrangement mode and / or an intelligent arrangement mode; A conversation message corresponding to the problem to be processed is generated based on the expert orchestration mode and / or the intelligent orchestration mode, and the conversation message is fed back to the user terminal, so that the holder of the user terminal performs broadband fault repair according to the conversation message.

2. The method for generating a session message according to claim 1, characterized in that: Based on the preset intention recognition model, the intention of the problem to be processed is recognized to obtain the intention recognition results, including: Verify the problem to be processed to determine whether the problem to be processed is a normal problem; If the problem to be processed is a normal problem, the intention of the problem to be processed is identified based on a preset intention recognition model to obtain an intention recognition result.

3. The method for generating a session message according to claim 2, characterized in that: Verifying the problem to be processed to determine whether the problem to be processed is a normal problem includes: Performing word segmentation processing on the problem to be processed to obtain a first word segmentation result, and matching a target word corresponding to the first word segmentation result in a preset rule dictionary; If the target word corresponding to the problem to be processed exists in the preset rule dictionary, then the problem to be processed is determined to be an abnormal problem; If the target word corresponding to the problem to be processed does not exist in the preset rule dictionary, clustering the problem to be processed based on the first word segmentation result to obtain a problem clustering result; The problem category of the problem to be processed is determined according to the problem clustering result, and whether the problem to be processed is a normal problem is determined according to the problem category.

4. The method for generating a conversation message according to claim 3, characterized in that: Determining whether the problem to be processed is a normal problem according to the problem category includes: If the problem category of the problem to be processed is a normal category, then the problem to be processed is determined to be a normal problem; if the problem category of the problem to be processed is a malicious category, then the problem to be processed is determined to be an abnormal problem; If the problem category of the problem to be processed is a suspected normal category and / or a suspected malicious category, the problem category to which the problem to be processed belongs is predicted based on a preset problem classification model to obtain a category prediction result, and determine whether the problem to be processed is a normal problem based on the category prediction result.

5. The method for generating a conversation message according to claim 3, characterized in that: Based on the first word segmentation result, clustering processing is performed on the problem to be processed to obtain a problem clustering result, including: Performing word embedding processing on the first word segmentation result based on a preset word embedding model to obtain a first text semantic feature of the problem to be processed, and calculating a cosine similarity value between the first text semantic features; The similarity scores between the problems to be processed are determined based on the cosine similarity values, and the problems to be processed are clustered according to a preset problem clustering model and the similarity scores to obtain problem clustering results.

6. The method for generating a conversation message according to claim 1, characterized in that: According to the intention recognition result and the problem to be processed, a historical conversation scenario corresponding to the problem to be processed is matched from a preset conversation database, including: Sending the intention recognition result to the user terminal, and receiving intention discrimination feedback information corresponding to the intention recognition result fed back by the user terminal; Determine whether the intent recognition result is accurate based on the intent discrimination feedback information, and when it is determined that the intent recognition result is accurate, match the historical conversation scene corresponding to the problem to be processed from a preset conversation database based on the intent recognition result and the problem to be processed.

7. The method for generating a conversation message according to claim 1, characterized in that: Allocating a conversation arrangement mode to the problem to be processed according to the historical conversation scenario includes: If the historical conversation scenario exists in the preset conversation database, an expert arrangement mode is assigned to the problem to be processed; wherein the expert arrangement mode is a mode generated according to a judgment rule; If the historical conversation scenario does not exist in the preset conversation database, matching the historical topic corresponding to the problem to be processed from the preset conversation database according to the intention recognition result and the problem to be processed; If the historical topic exists in the preset conversation database, an expert orchestration mode is assigned to the problem to be processed; if the historical topic does not exist, an intelligent orchestration mode is assigned to the problem to be processed; wherein the intelligent orchestration mode is an intelligent entity formed by one or more large models.

8. The method for generating a conversation message according to claim 1, characterized in that: Generating a conversation message corresponding to the problem to be processed based on the expert orchestration mode and / or the intelligent orchestration mode includes: Calling an expert arrangement information library and an API information library, and executing an API information query task included in the expert arrangement information library, querying actual broadband API status information corresponding to the user terminal from the API information library; Matching the actual broadband API status information with the preset condition rules in the expert arrangement information library to obtain a status matching result; If the status matching result is that the actual broadband API status information is consistent with the preset condition rule, generating a session message corresponding to the problem to be processed according to the status matching result; If the status matching result is that the actual broadband API status information is inconsistent with the preset condition rule, a conversation message corresponding to the problem to be processed is generated based on the intelligent orchestration mode.

9. The method for generating a conversation message according to claim 8, characterized in that: Generating a conversation message corresponding to the problem to be processed based on the intelligent orchestration mode includes: Performing word embedding processing on the actual broadband API status information based on a preset word embedding model to obtain actual bandwidth features, and matching historical bandwidth features corresponding to the actual bandwidth features from a preset knowledge base; The actual bandwidth characteristics, historical bandwidth characteristics, and the fault diagnosis results and fault repair results corresponding to the historical bandwidth characteristics are input into a preset content generation model to obtain a session message corresponding to the problem to be processed.

10. A device for generating a conversation message, characterized in that: include: The intention recognition module is used to receive the pending problem reported by the user terminal, and recognize the intention of the pending problem based on a preset intention recognition model to obtain an intention recognition result; A conversation scene matching module, used to match a historical conversation scene corresponding to the problem to be processed from a preset conversation database according to the intention recognition result and the problem to be processed; A session arrangement mode allocation module, used to allocate a session arrangement mode to the problem to be processed according to the historical session scenario; wherein the session arrangement mode includes an expert arrangement mode and / or an intelligent arrangement mode; A session message generation module is used to generate a session message corresponding to the problem to be processed based on the expert orchestration mode and / or the intelligent orchestration mode, and feed the session message back to the user terminal so that the holder of the user terminal can repair the broadband fault according to the session message.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating a session message according to any one of claims 1 to 9 is implemented.

12. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the method for generating a session message according to any one of claims 1 to 9 by executing the executable instructions.

Images