Problem processing method and system, intelligent terminal and computer readable storage medium

Abstract

The invention discloses a problem processing method and system, an intelligent terminal and a computer readable storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a to-be-processed problem and historical processing information; performing retrieval according to the to-be-processed problem and the historical processing information to obtain a target text; according to the to-be-processed question, the historical processing information and the target text, generating a target answer and candidate questions corresponding to the to-be-processed question through a preset question and answer processing model; in response to the received interaction signal of the target object, if the interaction signal meets a preset processing termination condition, outputting a target answer; and if the interaction signal does not meet the preset processing termination condition, respectively updating the to-be-processed question and the historical processing information according to the interaction signal, the candidate question and the target answer, and returning to execute the step of retrieving according to the to-be-processed question and the historical processing information to obtain the target text. The method is beneficial for improving the accuracy of the finally obtained target answer.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a problem-solving method, system, intelligent terminal and computer-readable storage medium. Background Art

[0002] With the development of science and technology, especially artificial intelligence technology, natural language interaction can be provided to users based on large language models to answer questions raised by users.

[0003] At present, in the process of processing questions provided to users, the questions raised by users are usually obtained, and corresponding answer information is directly generated through a preset large language model according to the questions, and the answer information is used as the answer corresponding to the user's question. The problem with the related art is that only using a large language model to generate corresponding answer information based on user questions and directly using it as the answer corresponding to the user's questions is not conducive to improving the accuracy of the answer.

[0004] Therefore, relevant technologies still need to be improved and developed. Summary of the invention

[0005] The main purpose of this application is to provide a question processing method, system, intelligent terminal and computer-readable storage medium, aiming to solve the technical problem in the related technology that only a large language model is used to generate corresponding answer information based on user questions, and directly used as the corresponding answer to the user question, which is not conducive to improving the accuracy of the answer.

[0006] In order to achieve the above-mentioned purpose, the first aspect of the present application provides a problem solving method, wherein the problem solving method comprises:

[0007] Obtain pending issues and historical processing information;

[0008] According to the above-mentioned problem to be processed and the above-mentioned historical processing information, a target text is retrieved;

[0009] Generate a target answer and candidate questions corresponding to the above-mentioned question to be processed through a preset question-answering processing model according to the above-mentioned question to be processed, the above-mentioned historical processing information and the above-mentioned target text;

[0010] In response to receiving an interaction signal from a target object, if the interaction signal satisfies a preset processing termination condition, outputting the target answer;

[0011] If the above-mentioned interaction signal does not meet the preset processing termination condition, then the above-mentioned pending question and the above-mentioned historical processing information are updated respectively according to the above-mentioned interaction signal, the above-mentioned candidate question and the above-mentioned target answer, and the step of retrieving and obtaining the target text according to the above-mentioned pending question and the above-mentioned historical processing information is returned to execute.

[0012] Optionally, the step of retrieving the target text based on the problem to be processed and the historical processing information includes:

[0013] Generate keywords based on the pending issues and the historical processing information, wherein the keywords are associated with the content of the pending issues and / or the content of the historical processing information;

[0014] According to the above-mentioned problem to be processed and the above-mentioned keywords, the above-mentioned target text is retrieved and obtained.

[0015] Optionally, the above-mentioned searching and obtaining the above-mentioned target text according to the above-mentioned problem to be processed and the above-mentioned keywords includes:

[0016] According to the above-mentioned problem to be processed and the above-mentioned keywords, a target feature vector is obtained through a preset embedding model;

[0017] According to the target feature vector, multiple document blocks are retrieved from a preset knowledge base;

[0018] The similarities between each of the document blocks and the problem to be processed are determined respectively, and the target text is obtained after the document blocks are sorted according to the similarities.

[0019] Optionally, the above-mentioned generating a target answer and candidate questions corresponding to the above-mentioned question to be processed by a preset question-answering processing model according to the above-mentioned question to be processed, the above-mentioned historical processing information and the above-mentioned target text includes:

[0020] Based on the above-mentioned pending question and the above-mentioned target text, a query answer corresponding to the above-mentioned pending question is generated by a preset first question-answering processing model;

[0021] Based on the above-mentioned unprocessed questions, the above-mentioned historical processing information and the above-mentioned inquiry answers, the target answers and candidate questions corresponding to the above-mentioned unprocessed questions are generated through a preset second question and answer processing model.

[0022] Optionally, the first question and answer processing model is set on the server side, and the second question and answer processing model is set on the edge device side.

[0023] Optionally, the above method further includes:

[0024] Sending the target answer and the candidate question to the target object to trigger the target object to feedback an interaction signal based on the interaction operation;

[0025] Among them, the above-mentioned target objects include the preset question and answer evaluation platform and / or the users who raised the above-mentioned questions to be processed.

[0026] Optionally, the updating of the to-be-processed question and the historical processing information respectively according to the interaction signal, the candidate question and the target answer includes:

[0027] Adding the above-mentioned to-be-processed questions and their corresponding target answers to the above-mentioned historical processing information to form updated historical processing information;

[0028] If the above-mentioned interaction signal indicates at least one target candidate question among the above-mentioned candidate questions and / or indicates at least one new question provided by the above-mentioned target object, the above-mentioned target candidate question and / or the above-mentioned new question are used as updated questions to be processed.

[0029] A second aspect of the present application provides a problem handling system, wherein the problem handling system comprises:

[0030] Data acquisition module, used to obtain pending issues and historical processing information;

[0031] A retrieval module, used for retrieving and obtaining a target text according to the above-mentioned problem to be processed and the above-mentioned historical processing information;

[0032] A data processing module, used to generate a target answer and candidate questions corresponding to the above-mentioned question to be processed through a preset question-answering processing model according to the above-mentioned question to be processed, the above-mentioned historical processing information and the above-mentioned target text;

[0033] an output module, configured to output the target answer in response to receiving an interaction signal from a target object and if the interaction signal satisfies a preset processing termination condition;

[0034] A data updating module is used to update the above-mentioned pending questions and the above-mentioned historical processing information respectively according to the above-mentioned interaction signal, the above-mentioned candidate questions and the above-mentioned target answers if the above-mentioned interaction signal does not meet the preset processing termination conditions, and return to execute the above-mentioned step of retrieving and obtaining the target text according to the above-mentioned pending questions and the above-mentioned historical processing information.

[0035] A third aspect of the present application provides an intelligent terminal, which includes a memory, a processor, and a problem handling program stored in the memory and executable on the processor, wherein the problem handling program implements any one of the steps of the problem handling method when executed by the processor.

[0036] A fourth aspect of the present application provides a computer-readable storage medium, on which a problem handling program is stored. When the problem handling program is executed by a processor, the steps of any one of the above-mentioned problem handling methods are implemented.

[0037] As can be seen from the above, in the present application scheme, the pending questions and historical processing information are obtained; the target text is retrieved based on the above pending questions and the above historical processing information; the target answer and candidate questions corresponding to the above pending questions are generated through a preset question-answer processing model based on the above pending questions, the above historical processing information and the above target text; in response to receiving the interaction signal of the target object, if the above interaction signal satisfies the preset processing termination condition, the above target answer is output; if the above interaction signal does not satisfy the preset processing termination condition, the above pending questions and the above historical processing information are updated respectively according to the above interaction signal, the above candidate questions and the above target answer, and the step of retrieving the target text according to the above pending questions and the above historical processing information is returned to execute.

[0038] Compared with the prior art, the present application does not directly generate corresponding answer information based on user questions, but first retrieves the target text by comprehensively searching for the pending questions and historical processing information, and then generates the corresponding target answers and candidate questions in combination with the pending questions, historical processing information and target text. In addition, the interaction signal of the target object is obtained. If the interaction signal does not meet the preset processing termination condition, the pending questions and historical processing information are updated according to the interaction signal, candidate questions and target answers, and the question processing is returned again. That is, the present application can perform multiple rounds of question processing to generate a more accurate target answer until the interaction signal given by the target object meets the processing termination condition. In this way, multiple rounds of question and answer processing are performed in combination with various data such as pending questions, historical processing information and target text, which is conducive to improving the accuracy of the target answer finally obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 It is a flowchart of a problem solving method provided by an embodiment of the present application;

[0041] Figure 2 It is a specific flow chart of a problem solving method provided by an embodiment of the present application;

[0042] Figure 3 It is a schematic diagram of system module interaction provided by an embodiment of the present application;

[0043] Figure 4 is a schematic diagram of a candidate question generation process provided by an embodiment of the present application;

[0044] Figure 5 It is a schematic diagram of component modules of a problem handling system provided in an embodiment of the present application;

[0045] Figure 6 It is a block diagram of the internal structure principle of a smart terminal provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0047] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0048] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0049] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0050] As used in this specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to being classified into," depending on the context. Similarly, the phrase "if it is determined" or "if classified into [described condition or event]" may be interpreted as meaning "upon determination" or "in response to determining" or "upon classification into [described condition or event]" or "in response to being classified into [described condition or event]," depending on the context.

[0051] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0053] At present, users have higher and higher demands and requirements for natural language interaction. In actual application scenarios, natural language interaction can be provided to users based on large language models to answer questions raised by users. Large language models (LLM) can provide users with friendly natural language interaction and personalized services. These models help users solve various problems by understanding and generating natural language, from information query to task automation, significantly improving user experience and service quality.

[0054] In one application scenario, a large language model can be deployed in network edge devices (such as mobile servers, personal computers, and edge computing servers) to reduce latency, improve response speed, and enhance data privacy and security, because data processing can be done more locally, reducing reliance on cloud transmission.

[0055] In another application scenario, in order to further improve the reliability and accuracy of edge services, the retrieval-augmented generation (RAG) technology can be further integrated into the edge large language model service. RAG is a technology that combines retrieval and generation models to improve the performance of large language models (LLMs). The core idea of ​​RAG is to enhance the generation capability of the model by introducing external knowledge bases or documents, so that LLM can solve the hallucination problem during generation to a certain extent when answering questions through more accurate and context-related information provided by the retriever.

[0056] Existing retrieval enhancement solutions usually adopt a "retrieve-and-generate" setting, that is, to retrieve relevant information based on the input content at one time, and then generate a complete answer. This method does not work well in long text generation tasks because the information requirements are complex and not always obvious from the input. For example, when a user asks a complex query, the simple similarity retrieval in RAG may have difficulty in correctly understanding the intent and context of the query, resulting in the retrieved documents not being highly relevant to the query.

[0057] In another application scenario, the retrieval ability of RAG in the face of complex queries can be improved through query rewriting, iterative query, recursive query, etc. For example, a query rewriter can be trained to optimize the original query so that it can more accurately capture the user's intent, thereby improving the quality of the retrieval results. However, this method faces problems with training data requirements and generalization capabilities. The training data needs to be high quality and diverse, and the collection and annotation costs are high and time-consuming. When the data is insufficient, the model performance may not be ideal. In terms of generalization ability, the model may overfit a specific data set and have difficulty performing well in new fields or different query types, requiring better regularization and data augmentation strategies. Dynamic updates and adaptive capabilities are also challenges, and the model needs to continue to learn to adapt to changing user needs and language patterns.

[0058] In another application scenario, the knowledge base can be repeatedly searched based on the initial query and the text generated so far to retrieve more relevant and precise knowledge. However, when the query information is not clear or ambiguous, the initial search results may contain a lot of irrelevant information. In this case, the subsequent iteration process may rely on this irrelevant information, causing the model to gradually deviate from the actual needs of the user. In addition, as the number of iterations increases, the accumulation of noise and irrelevant information may also cause the generated content to gradually deviate from the user's original intention or the direction they originally wanted to explore.

[0059] For example, if the user's initial query is "ecological environment of marine life", but does not further specify which type of marine life or which region's ecological environment they are interested in, the model may retrieve a large amount of general marine ecological information in subsequent iterations, which may not meet the user's true intention. This deviation may gradually amplify in each round of iteration, and ultimately result in the generated content being far from the user's actual needs.

[0060] In another application scenario, the retrieval system can be used to find documents related to the fuzzy question first, and then all possible explanations of the question (i.e., clear questions of different dimensions) are recursively generated through a small number of example prompts, and a tree structure is constructed to organize these explanations. In the generation stage, the generated tree structure is pruned to remove those clear questions that are inconsistent or irrelevant to the original question. This pruning process ensures that the retained nodes are highly relevant to the user's original fuzzy question, thereby improving the accuracy and relevance of the answer. Ultimately, all valid explanations are combined to generate a comprehensive long-form answer that covers all possible dimensions of the question and provides a detailed and multi-angle answer. However, the construction and processing of this tree structure may require a lot of computing resources, because the generation and pruning of each level involves complex calculations and a lot of information processing.

[0061] In order to solve at least one of the above-mentioned multiple technical problems, in the solution of the present application, a pending question and historical processing information are obtained; a target text is retrieved based on the pending question and the historical processing information; a target answer and candidate questions corresponding to the pending question are generated through a preset question-answer processing model based on the pending question, the historical processing information and the target text; in response to receiving an interaction signal from the target object, if the interaction signal satisfies a preset processing termination condition, the target answer is output; if the interaction signal does not satisfy the preset processing termination condition, the pending question and the historical processing information are updated respectively based on the interaction signal, the candidate question and the target answer, and the step of retrieving the target text based on the pending question and the historical processing information is returned to execute.

[0062] Compared with the prior art, the present application does not directly generate corresponding answer information based on user questions, but first retrieves the target text by comprehensively searching for the pending questions and historical processing information, and then generates the corresponding target answers and candidate questions in combination with the pending questions, historical processing information and target text. In addition, the interaction signal of the target object is obtained. If the interaction signal does not meet the preset processing termination condition, the pending questions and historical processing information are updated according to the interaction signal, candidate questions and target answers, and the question processing is returned again. That is, the present application can perform multiple rounds of question processing to generate a more accurate target answer until the interaction signal given by the target object meets the processing termination condition. In this way, multiple rounds of question and answer processing are performed in combination with various data such as pending questions, historical processing information and target text, which is conducive to improving the accuracy of the target answer finally obtained.

[0063] Specifically, the above-mentioned problem handling method provided in this application can deploy the models involved in some steps on the edge device side (i.e., the user side), so that the edge side model can actively interact and explore with the user, and improve the accuracy of generated answers based on user feedback.

[0064] Exemplary Methods

[0065] like Figure 1 As shown, the embodiment of the present application provides a method for solving the problem. Specifically, the method includes the following steps:

[0066] Step S100, obtaining issues to be processed and historical processing information.

[0067] The pending questions are questions that need to be processed to generate corresponding answers. The pending questions can be raised by the target object (such as a user) or obtained in other ways (such as sent by other devices to the device for processing the questions), which is not specifically limited here.

[0068] The above-mentioned historical processing information is the information generated during the processing corresponding to the above-mentioned pending problems. Specifically, based on the problem processing method provided in the embodiment of the present application, multiple rounds of processing can be performed on the pending problems (including the initial pending problems and the pending problems updated during multiple rounds of iterative processing), for example, multiple rounds of dialogues are processed to generate the final answer. In this process, for the pending problems in any round of processing, the historical processing information includes the answers generated in the previous rounds, and may also include the problems corresponding to the previous rounds or other information involved in the previous rounds of processing, which is not specifically limited here. It should be noted that in the initial round, the historical processing information corresponding to the initial pending problems may be empty, or be a preset prompt information, or include the corresponding target object (ie, user)'s question preferences, areas of interest and other information, which is not specifically limited here.

[0069] It should be noted that the steps corresponding to the problem handling method provided in the embodiment of the present application can be completely deployed on the cloud server, and all steps are executed by the cloud server; they can also be completely deployed on the edge device side, and all steps are executed by the edge device; they can also be partially deployed on the cloud server and partially deployed on the edge device side, using the computing power of the cloud server and the rapid response capability of the edge device to provide users with a better user experience.

[0070] In the embodiment of the present application, the steps of the above problem handling method are partially deployed on the cloud server and partially deployed on the edge device end for specific description, but it is not a specific limitation. Specifically, the user interacts with the edge device, and the edge device end obtains the problem to be processed and the historical processing information.

[0071] Step S200, obtaining the target text according to the above-mentioned problem to be processed and the above-mentioned historical processing information.

[0072] The content of the target text is associated with the content of the pending question and / or the content of the historical processing information. In an embodiment of the present application, a query question is generated based on the pending question and / or the historical processing information to retrieve the target text associated with the content, so as to generate an answer in combination with the target text, thereby improving the accuracy of the answer generation.

[0073] Specifically, the above-mentioned step of retrieving and obtaining the target text according to the above-mentioned problem to be processed and the above-mentioned historical processing information includes:

[0074] Generate keywords based on the pending issues and the historical processing information, wherein the keywords are associated with the content of the pending issues and / or the content of the historical processing information;

[0075] According to the above-mentioned problem to be processed and the above-mentioned keywords, the above-mentioned target text is retrieved and obtained.

[0076] Specifically, in an embodiment of the present application, an LLM model is deployed on an edge device, and the above-mentioned problems to be processed and the above-mentioned historical processing information are summarized and extracted based on the LLM model to obtain corresponding keywords. It should be noted that keywords can also be extracted by other means, which are not specifically limited here. These keywords are used to limit the scope of the retrieved target text, thereby improving the relevance and accuracy of the retrieval.

[0077] Furthermore, the above-mentioned target text is retrieved based on the above-mentioned problem to be processed and the above-mentioned keywords, including:

[0078] According to the above-mentioned problem to be processed and the above-mentioned keywords, a target feature vector is obtained through a preset embedding model;

[0079] According to the target feature vector, multiple document blocks are retrieved from a preset knowledge base;

[0080] The similarities between each of the document blocks and the problem to be processed are determined respectively, and the target text is obtained after the document blocks are sorted according to the similarities.

[0081] In the embodiment of the present application, after the edge device retrieves and obtains the keywords, it uploads the pending issues, keywords and historical processing information to the cloud server, retrieves the target text through the cloud server, and utilizes the cloud server's more powerful data processing capabilities to improve the retrieval effect and search results.

[0082] Specifically, the cloud server (or remote server) uses a retriever to perform similarity retrieval in a preset knowledge base under the constraints of similarity and keywords. The retriever calculates the corresponding target feature vector based on the preset embedding model according to the keywords extracted by LLM and the questions to be processed, and then searches for document blocks based on the target feature vector.

[0083] After finding multiple document blocks, a rearranger based on a cross encoder can be used to determine the similarity between each of the above document blocks and the above-mentioned problem to be processed, so as to sort the document blocks based on the similarity between the document blocks and the problem to be processed, and put document blocks with higher similarity in a closer position, and select document blocks for content generation based on the similarity, thereby improving the efficiency of subsequent processing and ensuring that document blocks with higher similarity are processed first, thereby improving the speed and accuracy of generated answers.

[0084] In an application scenario, the target text retrieval process can also be supervised based on other methods. For example, the document blocks retrieved in each round can be further screened by a similarity restriction method, and only a few document blocks with high similarity can be retained. For example, a preset number of document blocks with high similarity can be retained, or document blocks with similarity higher than a preset similarity threshold can be retained, which is not specifically limited here.

[0085] Step S300, based on the above-mentioned unprocessed question, the above-mentioned historical processing information and the above-mentioned target text, a target answer and candidate questions corresponding to the above-mentioned unprocessed question are generated through a preset question-answering processing model.

[0086] The preset question-answering processing model is a pre-set large language model that has been trained. It should be noted that in one application scenario, a large language model can be trained to implement the function of step S300. In another application scenario, multiple large language models can also be trained to jointly implement the function of step S300.

[0087] In the embodiment of the present application, the preset question-answering processing model includes a first question-answering processing model and a second question-answering processing model. The above-mentioned target answer and candidate questions corresponding to the above-mentioned question to be processed are generated by the preset question-answering processing model according to the above-mentioned question to be processed, the above-mentioned historical processing information and the above-mentioned target text, including:

[0088] Based on the above-mentioned pending question and the above-mentioned target text, a query answer corresponding to the above-mentioned pending question is generated by a preset first question-answering processing model;

[0089] Based on the above-mentioned unprocessed questions, the above-mentioned historical processing information and the above-mentioned inquiry answers, the target answers and candidate questions corresponding to the above-mentioned unprocessed questions are generated through a preset second question and answer processing model.

[0090] Among them, the above-mentioned first question and answer processing model is set on the server side, and the above-mentioned second question and answer processing model is set on the edge device side.

[0091] Specifically, the first question-answering processing model of the remote device (i.e., the server side) generates a query answer based on the question to be processed and the target text of this round of retrieval. The second question-answering processing model on the edge device side generates a target answer and candidate questions for the question to be processed based on the historical processing information and the query answer. Among them, the above-mentioned candidate questions are questions whose content is related to the question to be processed and / or the target answer, and are used to provide users with further question directions for the question to be processed.

[0092] Step S400, in response to receiving an interaction signal from a target object, if the interaction signal satisfies a preset processing termination condition, outputting the target answer.

[0093] It should be noted that, before step S400, the above method further includes:

[0094] Sending the target answer and the candidate question to the target object to trigger the target object to feedback an interaction signal based on the interaction operation;

[0095] Among them, the above-mentioned target objects include the preset question and answer evaluation platform and / or the users who raised the above-mentioned questions to be processed.

[0096] Specifically, the target object can be a specific user or a pre-set question-and-answer evaluation platform for evaluating the target answer. For example, a trained answer evaluation model can be set on the question-and-answer evaluation platform to supervise or guide the entire process. Specifically, a supervisor or LLM can be used to determine whether the existing target answer needs to be further retrieved in certain directions, thereby making the entire process more automated and reducing possible misguidance by users.

[0097] Furthermore, the above-mentioned target objects may also include specific users and question-and-answer evaluation platforms at the same time. When the target objects include both users and question-and-answer evaluation platforms, the priorities of the two may be set in advance, and the interaction signals actually used may be determined based on the priorities of the two. For example, when the user priority is set higher, subsequent processing is performed based on the interaction signal provided by the user. The interaction signals of the two may also be comprehensively considered to determine the subsequent operations to be performed. For example, the problem handling process is terminated only when both give a termination signal. If the two indicate different problems, the problems indicated by both are considered at the same time, but this is not specifically limited here.

[0098] The above-mentioned interactive operation is used to indicate the target object's intention for the next operation. Specifically, it is used to indicate that the target object believes that the current answer can answer the pending question, or to indicate the direction of further questioning.

[0099] The above-mentioned preset processing termination condition can be set and adjusted according to actual needs. For example, the preset processing termination condition is receiving a termination signal. At this time, if the above-mentioned interaction signal is a termination signal, the question termination condition is met and there is no need to generate the next round of answers. Otherwise, the next round of answer generation is entered. The above-mentioned preset processing termination condition can also be set to the number of iterations reaching a preset round number threshold, etc., which is not specifically limited here.

[0100] In an embodiment of the present application, if the above-mentioned interaction signal meets the preset processing termination condition, the above-mentioned target answer is directly output to the user through the edge device to complete the problem processing process.

[0101] Step S500, if the above-mentioned interaction signal does not meet the preset processing termination condition, then according to the above-mentioned interaction signal, the above-mentioned candidate question and the above-mentioned target answer, the above-mentioned pending question and the above-mentioned historical processing information are updated respectively, and the step of retrieving and obtaining the target text according to the above-mentioned pending question and the above-mentioned historical processing information is returned to execute.

[0102] Specifically, if the above interaction signal is not a termination signal, that is, the target object (user) believes that further questions are needed, then the next round of question processing process is entered.

[0103] In the embodiment of the present application, the updating of the to-be-processed question and the historical processing information respectively according to the interaction signal, the candidate question and the target answer includes:

[0104] Adding the above-mentioned to-be-processed questions and their corresponding target answers to the above-mentioned historical processing information to form updated historical processing information;

[0105] If the above-mentioned interaction signal indicates at least one target candidate question among the above-mentioned candidate questions and / or indicates at least one new question provided by the above-mentioned target object, the above-mentioned target candidate question and / or the above-mentioned new question are used as updated questions to be processed.

[0106] Furthermore, the target candidate question and / or the new question may be taken together with the currently existing questions to be processed as updated questions to be processed.

[0107] Specifically, users can select target candidate questions from candidate questions through interactive signals, and / or input new questions, that is, raise new questions to determine the exploration direction and / or continue exploration on the basis of given candidate questions, thereby avoiding retrieval in unnecessary directions, saving resources and time, and improving the accuracy of target answers generated during question processing to better meet user needs.

[0108] In the embodiment of the present application, the method for solving the above problem is further described in detail based on a specific application scenario. Figure 2is a specific flow chart of a problem solving method provided in an embodiment of the present application. Figure 3 is a schematic diagram of system module interaction provided by an embodiment of the present application, wherein: Figure 3 The system shown is used to execute the above problem solving method. Figure 3 The interaction among the user device module, the edge device module and the remote server module is shown in the figure. It should be noted that during the problem handling process, the user device (user terminal) used by the user can be an edge device or an additional user terminal device, which is not specifically limited here. After the user raises a question (i.e., the problem to be processed) through the user terminal device, a recursive search is performed on the basis of the problem based on the problem handling method provided in this embodiment. This solution not only involves deploying the database and the retriever in the cloud to improve data retrieval capabilities, flexibility and data management efficiency, but also includes deploying a large language model (LLM) on edge devices, such as mobile servers, personal computers and edge computing servers, to improve processing efficiency and data security. Each round of retrieval includes Figure 2 Steps shown.

[0109] Specifically, after a user asks a question, the LLM on the edge device will first analyze the question and summarize keywords based on the question and conversation history prompts. These keywords are used to limit the scope of document blocks retrieved by the retriever, thereby improving the relevance and accuracy of the retrieval.

[0110] The remote server uses the retriever to perform similar searches in the knowledge base under the constraints of similarity and keywords. The retriever will search for related text blocks based on the keywords extracted by the LLM and the feature vector of the entire question obtained by the embedding model. This process utilizes the computing power of the remote server to ensure that the retrieval process is fast and efficient. The retrieved text blocks are then handed over to the rearranger for processing. It should be noted that the embedding model can calculate feature vectors for keywords and questions to be processed separately, or it can take keywords and questions to be processed as a whole to obtain a feature vector, which is not specifically limited here.

[0111] After receiving the retrieved text blocks, the cloud-based rearranger uses a cross encoder to calculate their relevance score (i.e., relevance) to the original question and re-ranks them accordingly.

[0112] After the above round of retrieval and rearrangement, recursive retrieval is performed according to the following steps. Specifically, the LLM on the remote device generates query answers based on the questions and the documents retrieved in this round, and the edge-side LLM generates the target answers and corresponding candidate questions in this round based on the query history (i.e., historical processing information) and the query answers. Figure 4 is a schematic diagram of a candidate question generation process provided by an embodiment of the present application. Specifically, Figure 4As shown, when the edge device side LLM generates the target answer for this round, if a sentence with a low-confidence token is generated (target answer sentence), this sentence is used to generate candidate questions for the low-confidence token, and the sentence is modified so that it does not contain relevant information about these tokens. Figure 4 Token#1 in represents the first identifier with low confidence. Specifically, if the confidence is lower than a preset confidence threshold, it is considered to be a sentence with a low confidence identifier.

[0113] After receiving the target answer and candidate questions of this round, if the user decides to ask a new question for retrieval or continue to explore the given candidate questions, the corresponding pending questions and historical processing information will be updated, and the next round of question processing will be returned. In this way, by continuously optimizing keywords and search results, the real intention of the user is gradually approached, which can reduce the machine's search in unnecessary directions, thereby saving resources and time. If the user gives a termination signal, the conversation ends, the historical processing information is cleared, and the final target answer is output.

[0114] As can be seen from the above, in the problem processing method provided by the embodiment of the present application, the corresponding answer information is not directly generated based on the user's question, but the target text is first obtained by combining the pending problem and the historical processing information to retrieve, and then the corresponding target answer and candidate question are generated in combination with the pending problem, the historical processing information and the target text. In addition, the interactive signal of the target object is also obtained. When the interactive signal does not meet the preset processing termination condition, the pending problem and the historical processing information are updated according to the interactive signal, the candidate question and the target answer, and the problem processing is returned to be re-performed, that is, the present application scheme can perform multiple rounds of problem processing to generate a more accurate target answer until the interactive signal given by the target object meets the processing termination condition. In this way, multiple rounds of question and answer processing are performed in combination with various data such as the pending problem, the historical processing information and the target text, which is conducive to improving the accuracy of the target answer finally obtained.

[0115] Professional field information is obtained from the retrieved content to mine complex queries. At the same time, the relevant documents retrieved are limited through interaction with keywords and users to reduce the impact of irrelevant documents on the generation during the retrieval process. Specifically, in each round of edge-side LLM generation, LLM will generate at least one potentially relevant candidate question for each low-confidence content, and it is up to the user to decide whether they want to further explore these questions. This solution determines the direction of question exploration through interaction with the user, thereby reducing the proportion of irrelevant information in the final retrieval results, making LLM more accurate when generating answers.

[0116] In addition, large language models and related retrieval systems can be deployed on edge devices, significantly improving the response speed and processing efficiency of user queries. Edge devices undertake most of the large language model calculation tasks, reducing the burden on cloud servers. In addition, through recursive retrieval technology, this solution can continuously optimize query results, making the target answers generated in each round closer to user needs, thereby providing more accurate and personalized services.

[0117] Furthermore, this solution has a faster response time, occupies fewer resources, and limits the direction of exploration through user interaction, which can reduce the impact of unnecessary document blocks on answers. In addition, the spread of erroneous information is reduced through user interaction and keyword restrictions, thereby reducing resource consumption on irrelevant information. In addition, this solution does not need to implement recursive retrieval based on an additional structured data set, but is implemented by users and models based on more general text data, which can reduce the difficulty of implementation.

[0118] Exemplary Devices

[0119] like Figure 5 As shown in , corresponding to the above problem handling method, the embodiment of the present application also provides a problem handling system, and the above problem handling system includes:

[0120] Data acquisition module 510, used to acquire pending issues and historical processing information;

[0121] Retrieval module 520, used for retrieving and obtaining target text according to the above-mentioned problem to be processed and the above-mentioned historical processing information;

[0122] The data processing module 530 is used to generate a target answer and candidate questions corresponding to the above-mentioned question to be processed through a preset question-answering processing model according to the above-mentioned question to be processed, the above-mentioned historical processing information and the above-mentioned target text;

[0123] The output module 540 is used to respond to receiving the interaction signal of the target object and output the target answer if the interaction signal meets the preset processing termination condition;

[0124] The data updating module 550 is used to update the above-mentioned pending questions and the above-mentioned historical processing information respectively according to the above-mentioned interaction signal, the above-mentioned candidate questions and the above-mentioned target answers if the above-mentioned interaction signal does not meet the preset processing termination conditions, and return to execute the above-mentioned step of retrieving the target text according to the above-mentioned pending questions and the above-mentioned historical processing information.

[0125] In this way, the corresponding answer information is not directly generated based on the user's question, but the target text is first retrieved by combining the pending questions and historical processing information, and then the corresponding target answers and candidate questions are generated in combination with the pending questions, historical processing information and target text. In addition, the interaction signal of the target object is also obtained. If the interaction signal does not meet the preset processing termination condition, the pending questions and historical processing information are updated according to the interaction signal, candidate questions and target answers, and the question processing is returned again. That is, the present application scheme can perform multiple rounds of question processing to generate a more accurate target answer until the interaction signal given by the target object meets the processing termination condition. In this way, multiple rounds of question and answer processing are performed in combination with various data such as pending questions, historical processing information and target text, which is conducive to improving the accuracy of the target answer finally obtained.

[0126] It should be noted that the specific structure and implementation of the above-mentioned problem handling system and its various modules or units can refer to the corresponding description in the above-mentioned method embodiment, and will not be repeated here.

[0127] It should be noted that the division method of each module of the above-mentioned problem handling system is not unique and is not specifically limited here.

[0128] Based on the above embodiments, the present application also provides a smart terminal, whose principle block diagram can be as follows: Figure 6 As shown. The above-mentioned intelligent terminal includes a processor, a memory, a network interface and a display screen connected through a system bus. Among them, the processor of the intelligent terminal is used to provide computing and control capabilities. The memory of the intelligent terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a problem handling program. The internal memory provides an environment for the operation of the operating system and the problem handling program in the non-volatile storage medium. The network interface of the intelligent terminal is used to communicate with an external terminal through a network connection. When the problem handling program is executed by the processor, the steps of any one of the above-mentioned problem handling methods are implemented. The display screen of the intelligent terminal can be a liquid crystal display screen or an electronic ink display screen.

[0129] Those skilled in the art will understand that Figure 6 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the smart terminal to which the solution of the present application is applied. The specific smart terminal may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0130] In one embodiment, a smart terminal is provided, which includes a memory, a processor, and a problem handling program stored in the memory and executable on the processor. When the problem handling program is executed by the processor, the steps of any one of the problem handling methods provided in the embodiments of the present application are implemented.

[0131] An embodiment of the present application further provides a computer-readable storage medium, on which a problem handling program is stored. When the problem handling program is executed by a processor, the steps of any one of the problem handling methods provided in the embodiment of the present application are implemented.

[0132] It should be understood that the serial numbers of the steps in the above embodiments do not imply a sequence of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0133] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the above-mentioned device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned device can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0134] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0135] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0136] In the embodiments provided in the present application, it should be understood that the disclosed system / terminal device and method can be implemented in other ways. For example, the system / terminal device embodiments described above are only schematic, for example, the division of the above modules or units is only a logical function division, and in actual implementation, other division methods can be used, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0137] If the above-mentioned integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The above-mentioned computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the above-mentioned computer program includes computer program code, and the above-mentioned computer program code can be in source code form, object code form, executable file or some intermediate form. The above-mentioned computer-readable medium may include: any entity or device capable of carrying the above-mentioned computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the above-mentioned computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0138] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A problem solving method, characterized in that: The method comprises: Obtain pending issues and historical processing information; Retrieve and obtain target text according to the problem to be processed and the historical processing information; Generate a target answer and candidate questions corresponding to the question to be processed through a preset question-answering processing model according to the question to be processed, the historical processing information and the target text; In response to receiving an interaction signal from a target object, if the interaction signal satisfies a preset processing termination condition, outputting the target answer; If the interaction signal does not satisfy the preset processing termination condition, the question to be processed and the historical processing information are updated respectively according to the interaction signal, the candidate question and the target answer, and the step of retrieving the target text according to the question to be processed and the historical processing information is returned to execute.

2. The problem solving method according to claim 1, characterized in that: The step of retrieving and obtaining a target text according to the problem to be processed and the historical processing information includes: Generating keywords according to the problem to be processed and the historical processing information, wherein the keywords are associated with the content of the problem to be processed and / or the content of the historical processing information; The target text is retrieved based on the problem to be processed and the keywords.

3. The problem solving method according to claim 2, characterized in that: The step of retrieving the target text according to the problem to be processed and the keyword includes: According to the problem to be processed and the keyword, a target feature vector is obtained through a preset embedding model; According to the target feature vector, a plurality of document blocks are retrieved from a preset knowledge base; The similarities between each of the document blocks and the problem to be processed are determined respectively, and the document blocks are sorted according to the similarities to obtain the target text.

4. The problem solving method according to claim 1, characterized in that: The step of generating a target answer and candidate questions corresponding to the question to be processed by a preset question-answer processing model according to the question to be processed, the historical processing information and the target text includes: Based on the question to be processed and the target text, generating a question answer corresponding to the question to be processed by a preset first question-answer processing model; Based on the question to be processed, the historical processing information and the inquiry answer, a target answer and candidate questions corresponding to the question to be processed are generated through a preset second question and answer processing model.

5. The problem solving method according to claim 4, characterized in that: The first question and answer processing model is set on the server side, and the second question and answer processing model is set on the edge device side.

6. The problem solving method according to claim 1, characterized in that: The method further comprises: Sending the target answer and the candidate question to the target object to trigger the target object to feedback an interaction signal based on the interaction operation; The target objects include a preset question-and-answer evaluation platform and / or a user who raises the question to be processed.

7. The problem solving method according to any one of claims 1 to 6, characterized in that: The updating of the to-be-processed question and the historical processing information respectively according to the interaction signal, the candidate question and the target answer comprises: Adding the to-be-processed question and its corresponding target answer to the historical processing information to form updated historical processing information; If the interaction signal indicates at least one target candidate question among the candidate questions and / or indicates at least one new question provided by the target object, the target candidate question and / or the new question are used as updated questions to be processed.

8. A problem handling system, characterized in that: The system comprises: Data acquisition module, used to obtain pending issues and historical processing information; A retrieval module, used for retrieving and obtaining a target text according to the problem to be processed and the historical processing information; A data processing module, configured to generate a target answer and candidate questions corresponding to the question to be processed through a preset question-answering processing model according to the question to be processed, the historical processing information and the target text; an output module, configured to output the target answer in response to receiving an interaction signal from a target object, if the interaction signal satisfies a preset processing termination condition; A data updating module is used to update the pending question and the historical processing information respectively according to the interaction signal, the candidate question and the target answer if the interaction signal does not meet the preset processing termination condition, and return to execute the step of retrieving and obtaining the target text according to the pending question and the historical processing information.

9. An intelligent terminal, characterized in that: The intelligent terminal includes a memory, a processor, and a problem handling program stored in the memory and executable on the processor. When the problem handling program is executed by the processor, the steps of the problem handling method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a problem handling program, and when the problem handling program is executed by the processor, the steps of the problem handling method according to any one of claims 1 to 7 are implemented.

Images