Self-verification intelligent reply method and device, equipment and storage medium

Abstract

The invention provides a self-verification intelligent reply method and device, equipment and a storage medium, and relates to the field of artificial intelligence. The method comprises the following steps: receiving a to-be-processed problem sent by a client; analyzing the problem content of the to-be-processed problem to obtain a task target and a keyword; generating a problem constraint through the task target, the keyword and a preset information base; querying in a preset resource database according to the problem constraint to obtain callable resources; generating an execution path based on the problem constraint and the callable resources; performing resource calling according to the execution path to generate a question answer; verifying the question answer according to the task target and the question constraint; and sending the verified question answer to the client. Resource calling is carried out by extracting task targets and keywords and generating question constraints, so that the answer generation efficiency and accuracy are improved; and self-verification is carried out after the answer is generated, so that the provided answer is reliable information after verification, and the user experience and the answer credibility are effectively improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a self-verifying intelligent reply method, device, equipment and storage medium. Background Art

[0002] Existing intelligent reply systems can answer almost any question and respond to various tasks and needs, but it is difficult to guarantee the accuracy of the answers. Especially when faced with complex questions or tasks, various large models and intelligent agent systems are difficult to give practical and usable results, which brings inconvenience to users in practical applications.

[0003] Real-life problems usually contain many implicit goals and constraints beyond the problem statement. The same problem requires different processing and answers in different scenarios. Existing smart replies usually generate answers immediately based on the questions but ignore many constraints, resulting in many errors in the analysis execution and answer verification stages.

[0004] Existing smart replies can give answers based on questions, but they are not verified to be right or wrong. When faced with actual complex tasks, the high error rate leads to poor performance. Some smart reply models are based on pattern matching based on training data in principle, which is only suitable for questions that can directly match existing knowledge in the training data. For questions that lack corresponding knowledge or require complex analysis, they will directly generate wrong or even illusory answers. Summary of the invention

[0005] The present application provides a self-verifying intelligent reply method, device, equipment and storage medium to solve the problem that the existing intelligent reply has a high error rate and leads to poor usage effect.

[0006] In a first aspect, the present application provides a self-verification intelligent reply method, comprising:

[0007] Receive pending issues sent by the client;

[0008] Analyze the problem content of the problem to be processed to obtain task objectives and keywords;

[0009] Generate problem constraints through the task objectives, keywords and preset information base;

[0010] According to the problem constraints, query the preset resource database to obtain available resources;

[0011] generating an execution path based on the problem constraints and the callable resources;

[0012] Calling resources according to the execution path to generate an answer to the question;

[0013] Verifying the answer to the question according to the task objective and the problem constraints;

[0014] Send the verified answer to the question to the client.

[0015] Optionally, in the above method, analyzing the content of the problem to be processed to obtain task objectives and keywords includes:

[0016] Input the question content into a preset natural language processing model for parsing to obtain the task objective;

[0017] Performing semantic analysis on the question content to obtain multiple words;

[0018] Calculating similarity between the vocabulary and the task target to obtain similarity data;

[0019] A plurality of keywords whose similarity data is greater than a preset similarity threshold are selected from the vocabulary.

[0020] Optionally, the method described above, generating problem constraints through the task objectives, keywords and a preset information base, includes:

[0021] According to the task objective and the keywords, relevant information is obtained from a preset information library by matching, wherein the relevant information includes environmental information and task guidance information;

[0022] The task objectives, keywords, environmental information and task guidance information are input into a preset multi-dimensional learning model to generate problem constraints.

[0023] Optionally, the method as described above, before querying the preset resource database to obtain the callable resources according to the problem constraint, further includes:

[0024] Get historical search logs;

[0025] Matching the task goal and the problem constraint with the historical search log to obtain historical similar problems and available resources and search success rate data corresponding to the historical similar problems;

[0026] Calculating based on the available resources, the success rate data and a preset weighted value to obtain confidence data;

[0027] Comparing the confidence data with a preset confidence threshold;

[0028] When the confidence data is less than the preset confidence threshold, a problem warning is directly generated and problem feedback is performed;

[0029] When the confidence data is greater than or equal to the preset confidence threshold, a query is performed in a preset resource database according to the problem constraints to obtain available resources.

[0030] Optionally, the method as described above, generating an execution path based on the problem constraints and the callable resources, comprises:

[0031] Inputting the problem constraints and the callable resources into a preset analysis model to obtain multiple executable paths;

[0032] Performing a weighted analysis on the resource utilization efficiency, response time and execution cost of each executable path to obtain path comprehensive data;

[0033] A path with the highest value is selected from a plurality of executable paths according to the path comprehensive data as an execution path.

[0034] Optionally, in the above method, generating the answer to the question by calling resources according to the execution path includes:

[0035] Calling resources from the preset resource database according to the sequence of steps in the execution path;

[0036] Obtaining the data returned by the resource and integrating it to obtain resource information;

[0037] Performing data cleaning and information extraction on the resource information to obtain target information;

[0038] The target information is sorted based on a preset answer format to obtain an answer to the question.

[0039] Optionally, the method as described above, verifying the answer to the question according to the task objective and the question constraint, includes:

[0040] Checking the answer to the question based on the task goal and the problem constraints to determine whether the answer to the question satisfies the problem constraints and the task goal;

[0041] When the answer to the question satisfies the question constraints and the task objectives, the answer to the question passes verification;

[0042] When the answer to the question does not satisfy the question constraints and / or the task objectives, the question to be processed is input into a preset analysis model to improve the question content, and the step of analyzing the question content of the question to be processed to obtain the task objectives and keywords is returned.

[0043] In a second aspect, the present application provides a self-verifying intelligent reply device, comprising:

[0044] The acquisition module is used to receive pending issues sent by the client;

[0045] An analysis module, used to analyze the problem content of the problem to be processed to obtain task objectives and keywords;

[0046] A constraint module, used to generate problem constraints through the task objectives, keywords and a preset information library;

[0047] A query module, used to query a preset resource database to obtain available resources according to the problem constraints;

[0048] A generation module, used for generating an execution path based on the problem constraints and the callable resources;

[0049] A calling module, used for calling resources according to the execution path to generate an answer to the question;

[0050] A verification module, used to verify the answer to the question according to the task objective and the question constraints;

[0051] The feedback module is used to send the verified answers to the questions to the client.

[0052] In a possible design, the analysis module is specifically used to:

[0053] Input the question content into a preset natural language processing model for parsing to obtain the task objective;

[0054] Performing semantic analysis on the question content to obtain multiple words;

[0055] Calculating similarity between the vocabulary and the task target to obtain similarity data;

[0056] A plurality of keywords whose similarity data is greater than a preset similarity threshold are selected from the vocabulary.

[0057] In a possible design, the constraint module is specifically used to:

[0058] According to the task objective and the keywords, relevant information is obtained from a preset information library by matching, wherein the relevant information includes environmental information and task guidance information;

[0059] The task objectives, keywords, environmental information and task guidance information are input into a preset multi-dimensional learning model to generate problem constraints.

[0060] In a possible design, the query module is also used to:

[0061] Get historical search logs;

[0062] Matching the task goal and the problem constraint with the historical search log to obtain historical similar problems and available resources and search success rate data corresponding to the historical similar problems;

[0063] Calculating based on the available resources, the success rate data and a preset weighted value to obtain confidence data;

[0064] Comparing the confidence data with a preset confidence threshold;

[0065] When the confidence data is less than the preset confidence threshold, a problem warning is directly generated and problem feedback is performed;

[0066] When the confidence data is greater than or equal to the preset confidence threshold, a query is performed in a preset resource database according to the problem constraints to obtain available resources.

[0067] In one possible design, a module is generated, specifically for:

[0068] Inputting the problem constraints and the callable resources into a preset analysis model to obtain multiple executable paths;

[0069] Performing a weighted analysis on the resource utilization efficiency, response time and execution cost of each executable path to obtain path comprehensive data;

[0070] A path with the highest value is selected from a plurality of executable paths according to the path comprehensive data as an execution path.

[0071] In one possible design, the calling module is specifically used to:

[0072] Calling resources from the preset resource database according to the sequence of steps in the execution path;

[0073] Obtaining the data returned by the resource and integrating it to obtain resource information;

[0074] Performing data cleaning and information extraction on the resource information to obtain target information;

[0075] The target information is sorted based on a preset answer format to obtain an answer to the question.

[0076] In a possible design, the verification module is specifically used to:

[0077] Checking the answer to the question based on the task goal and the problem constraints to determine whether the answer to the question satisfies the problem constraints and the task goal;

[0078] When the answer to the question satisfies the question constraints and the task objectives, the answer to the question passes verification;

[0079] When the answer to the question does not satisfy the question constraints and / or the task objectives, the question to be processed is input into a preset analysis model to improve the question content, and the step of analyzing the question content of the question to be processed to obtain the task objectives and keywords is returned.

[0080] In a third aspect, the present application provides a self-verifying intelligent reply device, comprising: a processor, and a memory communicatively connected to the processor;

[0081] Memory stores computer-executable instructions;

[0082] The processor executes the computer-executable instructions stored in the memory, so that the device performs any one of the self-verification intelligent reply methods in the first aspect.

[0083] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement a self-verifying intelligent reply method as described in any one of the first aspects.

[0084] The self-verifying intelligent reply method, device, equipment and storage medium provided by the present application receive pending questions sent by the client; analyze the content of the pending questions to obtain task objectives and keywords; generate problem constraints through task objectives, keywords and preset information base; query the preset resource database to obtain callable resources according to the problem constraints; generate an execution path based on the problem constraints and callable resources; call resources according to the execution path to generate the answer to the question; verify the answer to the question according to the task objectives and problem constraints; and send the verified answer to the client. By extracting the task objectives, keywords and generating problem constraints, querying and generating the optimal filter in the resource database for resource calling, the efficiency and accuracy of answer generation are improved; and self-verification is performed after the answer is generated to ensure that the answer provided is verified and reliable information, effectively improving the user experience and the credibility of the answer. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0086] Figure 1 This is a flowchart of the self-verification intelligent reply method provided in the embodiment of the present application. Figure 1 ;

[0087] Figure 2 This is a flowchart of the self-verification intelligent reply method provided in the embodiment of the present application. Figure 2 ;

[0088] Figure 3This is a flowchart of the self-verification intelligent reply method provided in the embodiment of the present application. Figure 3 ;

[0089] Figure 4 This is a flowchart of the self-verification intelligent reply method provided in the embodiment of the present application. Figure 4 ;

[0090] Figure 5 This is a flowchart of the self-verification intelligent reply method provided in the embodiment of the present application. Figure 5 ;

[0091] Figure 6 This is a flowchart of the self-verification intelligent reply method provided in the embodiment of the present application. Figure 6 ;

[0092] Figure 7 This is a flowchart of the self-verification intelligent reply method provided in the embodiment of the present application. Figure 7 ;

[0093] Figure 8 It is a structural diagram of a self-verifying intelligent reply device provided in an embodiment of the present application;

[0094] Fig. 9 It is a schematic diagram of the hardware structure of the self-verifying intelligent reply device provided in the embodiment of the present application.

[0095] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0096] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0098] At present, AI systems represented by GPT can answer almost any question and respond to various tasks and needs, but it is difficult to guarantee the accuracy of the answers. Especially when faced with complex questions or tasks, various large models and intelligent agent systems are difficult to give practical and usable results, so the user experience in actual applications is poor.

[0099] Problems in real-world applications usually contain many implicit goals and constraints beyond the problem statement. The same problem requires different processing and answers in different scenarios. Existing methods often ignore these constraints, resulting in many errors in the execution and answer verification stages. Take the simple question "The weather is not very good recently" as an example: in ordinary question-and-answer scenarios, you can reply to the queried weather results or directly generate some emotional replies. In agricultural services, you need to analyze the current solar terms and weather forecast results to give impacts and suggestions. In travel scenarios, you not only need to make decisions based on information from multiple parties, but you may also need to call other tools to do some ticket refund operations. Even in the same travel scenario, different applications will have different requirements for response time. In telephone customer service applications, you need to reply to users in a timely manner and appease their emotions, and at the same time call another service in the background to perform complex queries, executions, and other operations. In the operation background operation application, you can extend the response time to do more analysis and verification to achieve the best results. Existing systems will generate answers immediately but ignore many constraints. Some questions require missing resources, so the processing should be terminated and feedback should be given immediately. The correctness of the answer also needs to be verified by various constraints.

[0100] Existing smart replies can give answers based on questions, but they are not verified to be right or wrong. When faced with actual complex tasks, the high error rate leads to poor performance. Some smart reply models are based on pattern matching based on training data in principle, which is only suitable for questions that can directly match existing knowledge in the training data. For questions that lack corresponding knowledge or require complex analysis, they will directly generate wrong or even illusory answers.

[0101] The self-verification intelligent reply method provided in this application is intended to solve the above technical problems in the prior art.

[0102] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0103] Figure 1 Schematic diagram of a self-verification intelligent reply method according to an embodiment of the present invention Figure 1 ,like Figure 1 As shown, this embodiment provides a self-verification intelligent reply method, and the process includes the following steps:

[0104] Step S101: receiving unprocessed issues sent by the client.

[0105] Specifically, the pending questions are questions that need to be answered and are input by the user through the client, and are the original data for subsequent steps.

[0106] Step S102: Analyze the content of the problem to be processed to obtain task objectives and keywords.

[0107] Specifically, by parsing the question content to extract task objectives and keywords, we can accurately understand the user's intentions and improve our ability to understand the question by identifying the core content of the question.

[0108] Step S103: Generate problem constraints through task objectives, keywords and preset information base.

[0109] Specifically, by generating problem constraints, subsequent steps can be performed within a clear framework, which not only improves the accuracy of problem handling but also ensures that reasonable judgments and decisions can be made in complex environments.

[0110] Step S104: query the preset resource database to obtain available resources according to the problem constraints.

[0111] Specifically, the preset resource database contains various available resource information, such as data sets, algorithms, API interfaces, etc. Based on the problem constraints, query operations are performed in the resource database. Through precise queries, resources that match the problem constraints are identified to ensure that the acquired resources are relevant and available, which improves the accuracy of resource calls and optimizes the overall processing efficiency.

[0112] Step S105: Generate an execution path based on the problem constraints and callable resources.

[0113] Specifically, the optimal execution path is selected from multiple possible solutions through problem constraints and available resources, ensuring the efficiency and economy of resource calling.

[0114] Step S106: Generate an answer to the question by calling resources according to the execution path.

[0115] Specifically, by strictly executing path nodes to call resources, the final output is an answer that meets user needs, which improves the accuracy and efficiency of answer generation and ensures the quality of the final output and user satisfaction.

[0116] Step S107: Verify the answer to the question according to the task objectives and problem constraints.

[0117] Specifically, through strict verification steps, we ensure that the answers fed back to users are accurate and relevant, reduce the possibility of incorrect feedback, and improve the overall user experience by providing high-quality answers and fast response times.

[0118] Step S108: Send the verified answer to the question to the client.

[0119] The self-verifying intelligent reply method provided in the embodiment of the present invention improves the efficiency and accuracy of answer generation by extracting task objectives, keywords and generating question constraints, querying and generating the optimal filter in the resource database for resource call; and performs self-verification after generating the answer to ensure that the answer provided is verified and reliable information, effectively improving the user experience and the credibility of the answer.

[0120] Figure 2 Schematic diagram of the self-verification intelligent reply method provided in the embodiment of the present application Figure 2 .like Figure 2 As shown, this embodiment describes in detail the process of analyzing the problem content of the problem to be processed in the above embodiment to obtain the task objectives and keywords. The specific implementation method of the process includes the following steps:

[0121] Step S201: Input the question content into a preset natural language processing model for parsing to obtain the task objective.

[0122] Specifically, the question content entered by the user is input into the preset natural language processing model for parsing. The semantic understanding ability of the natural language processing model is used to analyze the sentence structure and semantics of the user input, accurately identify the core intention and task objectives of the question, and thus understand the user's real needs.

[0123] Step S202: semantically analyzing the question content to obtain multiple words.

[0124] Specifically, semantic analysis methods such as part-of-speech tagging and dependency parsing are used to decompose the question content into smaller language units, such as words or phrases, and then categorize and annotate them. These words can reflect the core elements of the question. By deeply understanding the components of the question, we can identify the words that may be important in the question, which makes it easier to establish semantic associations between the words and the question content.

[0125] Step S203: Calculate the similarity between the vocabulary and the task target to obtain similar data.

[0126] Specifically, similarity algorithms (such as cosine similarity, Jaccard similarity, etc.) are used to analyze the relevance between each word and the task target. For example, words are converted into points in the vector space through vector representation (Word Embeddings), and then the distance or angle between the points is calculated. By calculating the similarity between the word and the task target, the relevance of each word to the core task is quantified, which helps to screen out the most representative keywords.

[0127] Step S204: Filter out multiple keywords whose similar data are greater than a preset similarity threshold from the vocabulary.

[0128] Specifically, obtaining keywords through threshold screening can effectively reduce the interference of irrelevant information, improve the efficiency and accuracy of resource query, and ensure that the selected keywords are highly relevant to the task objectives.

[0129] The embodiment of the present invention provides an accurate information basis for subsequent processing by extracting task objectives and keywords from the question content. The screening process not only improves the accuracy of keyword extraction, but also enhances the ability to understand user intentions, thereby improving the quality of the overall reply.

[0130] In some optional embodiments, the method further comprises:

[0131] The problem to be processed is input into a preset complexity analysis model to obtain an output result, which is the problem to be processed or multiple sub-problems; the complexity analysis model is used to perform complexity analysis on the problem to be processed, check whether the problem to be processed exceeds a preset complexity threshold, and split the problem to be processed into multiple sub-problems when it is greater than or equal to the preset complexity threshold.

[0132] When the output question is a pending question, directly execute step S102;

[0133] When the output question is multiple sub-questions, determine whether there is a correlation between the multiple sub-questions;

[0134] If there is a correlation between the sub-problems, the sub-problems with a correlation are sorted according to their correlation order to obtain one or more problem sequences, and the sub-problems are input into the recursive analysis model according to the order in the problem sequence; the remaining sub-problems without a correlation are randomly input into the recursive analysis model; until the sub-problem answer of each sub-problem is obtained, the answers to all sub-problems are summarized to obtain the problem answer, and step S107 is continued.

[0135] If there is no correlation between the sub-questions, randomly select sub-questions and input them into the recursive analysis model until the sub-question answer of each sub-question is obtained, and the answer to all sub-questions is summarized to obtain the question answer, and then continue to execute step S107.

[0136] The recursive analysis model is used to execute steps S102 to S106 to obtain the first-level answer to the sub-problem as the problem to be processed, and store the first-level answer in the preset information library, and then input the sub-problem into the complexity analysis model to obtain the output result; when the output problem is the sub-problem, the process ends and the first-level answer is used as the answer to the sub-problem; when the output problem is still multiple problems, the sub-problem is used as the root node, and the multiple output problems are used as branch nodes of the root node, until the complexity of the problem at the final level is less than the preset complexity threshold and reaches the end of the path, and the answers to the leaf nodes at each level are summarized to obtain the answer to the sub-problem. The output answer of each level will be stored in the preset information library, and the information in the preset information library will be updated and supplemented to accumulate and utilize the existing solution experience.

[0137] Specifically, the complexity analysis model can be a time complexity model, a space complexity model, an algorithm complexity model, etc., which can be selected according to the actual application scenario. There is no specific restriction here. In actual applications, the complexity analysis model can also comprehensively consider the task objectives, keywords and other information of the problem to analyze and calculate the complexity of the problem more accurately. The complexity of the problem is analyzed by the complexity analysis model to determine whether it exceeds the preset complexity threshold, and the problem is divided into smaller sub-problems when it exceeds the preset complexity threshold. Through recursive analysis until the problem complexity of the final node of all sub-problems is lower than the preset complexity threshold, it is ensured that the leaf node problems at each level are within the processable complexity range; and by finally summarizing the answers to each leaf node to obtain the answers to the sub-problems, and then summarizing the answers to the sub-problems to form a complete answer to the problem to be processed.

[0138] This implementation is suitable for scenarios where complex problems need to be dealt with. By differentiating complex problems, it ensures that each sub-problem is within a controllable complexity range, thereby improving the efficiency and feasibility of overall problem solving.

[0139] Figure 3 Schematic diagram of the self-verification intelligent reply method provided in the embodiment of the present application Figure 3 .like Figure 3 As shown, this embodiment describes in detail the process of generating problem constraints through task objectives, keywords and preset information base in the above embodiment. The specific implementation method of this process includes the following steps:

[0140] Step S301: matching and obtaining relevant information from a preset information library according to the task objectives and keywords, the relevant information including environmental information and task guidance information.

[0141] Specifically, keyword search technology is used to retrieve environmental information and task guidance information related to task objectives and keywords in a preset information base (such as a knowledge graph, database, or document collection). For example, if the user asks a question about the weather forecast, the relevant environmental information may include geographic location, current season, etc.; the task guidance information may be an API interface or website link for obtaining weather data. By matching, information closely related to the problem can be quickly located, providing the necessary context and resource guidance for subsequent problem constraint generation.

[0142] Step S302: Input the task objectives, keywords, environmental information and task guidance information into a preset multi-dimensional learning model to generate problem constraints.

[0143] Specifically, a machine learning model (such as a neural network, support vector machine, or other classifier) ​​is used to comprehensively consider task objectives, keywords, environmental information, and task guidance information to perform feature extraction and pattern recognition. The model is able to handle complex relationships and diverse sources of information to generate more accurate and comprehensive problem constraints. By converting information into an input format that the model can understand, and analyzing their interactions and importance through the model's training weights. Ultimately, the model outputs the constraints of the problem, which clarify the scope, limitations, and requirements of the problem, and provide clear guidance for subsequent resource queries and path generation, thereby helping to improve the accuracy and relevance of the answer.

[0144] The embodiment of the present invention generates problem constraints so that subsequent steps can be performed within a clear framework. This approach not only improves the accuracy of problem handling, but also ensures that reasonable judgments and decisions can be made in complex environments.

[0145] Figure 4 Schematic diagram of the self-verification intelligent reply method provided in the embodiment of the present application Figure 4 .like Figure 4 As shown, this embodiment describes in detail the process of the above embodiment before querying the preset resource database to obtain the callable resources according to the problem constraints. The specific implementation of the process includes the following steps:

[0146] Step S401: Obtain historical search logs.

[0147] Specifically, historical search logs are obtained by collecting and organizing the system's past search records, including information such as previously solved questions, generated answers, and user feedback on the answers. Historical search logs can provide a rich historical data foundation to help identify past cases similar to the current problem, thereby better understanding the context of new problems and improving the efficiency and quality of problem solving.

[0148] Step S402: Match the task objectives and problem constraints with the historical search logs to obtain historical similar problems and available resources and search success rate data corresponding to the historical similar problems.

[0149] Specifically, the current task objectives and problem constraints can be compared with the records in the historical retrieval log through text similarity algorithms (such as cosine similarity, Jaccard similarity, etc.) to find the most similar one or more historical problems. At the same time, the available resources corresponding to these historical problems (such as API interfaces, database table names, etc.) and their retrieval success rate data are collected.

[0150] Step S403: Calculate based on available resources, success rate data and preset weighted values ​​to obtain confidence data.

[0151] Specifically, a confidence value is calculated based on the quality of available resources and the retrieval success rate data, combined with preset weighted values ​​(such as the importance of resources, frequency of use, etc.), through a certain mathematical model (such as linear regression, logistic regression, etc.). The confidence value represents the confidence level that the current problem to be processed can be successfully answered. The ability to answer the problem to be processed is reflected in a quantitative way, which helps to discover potential errors in advance and take corresponding measures to improve the accuracy of the answer.

[0152] Step S404: Compare the confidence data with a preset confidence threshold.

[0153] Specifically, the calculated confidence value is compared with a pre-set confidence threshold to determine whether the current question has sufficient historical support for reliable resource query. If the confidence value is lower than the confidence threshold, it means that there is a large degree of uncertainty in the system's answer to the question to be processed. By setting a threshold to determine whether further processing or manual intervention is needed, it is ensured that the answer finally provided to the user is accurate and reliable enough.

[0154] Step S405: When the confidence data is less than a preset confidence threshold, a problem warning is directly generated and problem feedback is performed.

[0155] Specifically, it avoids unreliable resource queries under low confidence conditions, generates incorrect answers, and prompts the need for further manual intervention or redefinition of the problem, so as to avoid errors that lead to a decline in user experience.

[0156] Step S406: When the confidence data is greater than or equal to a preset confidence threshold, query the preset resource database to obtain available resources according to the problem constraints.

[0157] Specifically, if the confidence value meets the requirements, the system will search for relevant information or services in the preset resource database according to the established problem constraints, ensuring that subsequent operations will only be performed when there is sufficient confidence, calling historically effective resources, and improving the efficiency and accuracy of problem solving.

[0158] The embodiment of the present invention improves the intelligence and response speed of answers and effectively reduces the occurrence rate of erroneous queries by making full use of historical data to analyze the processing reliability of current questions before performing resource queries.

[0159] Figure 5 Schematic diagram of the self-verification intelligent reply method provided in the embodiment of the present application Figure 5 .like Figure 5 As shown, this embodiment describes in detail the process of generating an execution path based on problem constraints and callable resources in the above embodiment. The specific implementation of the process includes the following steps:

[0160] Step S501: Input problem constraints and available resources into a preset large analysis model to obtain multiple executable paths.

[0161] Specifically, the preset analysis model is a rule-based machine learning model that can generate multiple possible execution paths based on the input information. Each path represents a possible resource call and operation sequence. The model uses its analysis capabilities to generate multiple executable paths. By analyzing the calculation and analysis of the analysis model, multiple possible solution paths are generated, providing multiple options for solving problems, ensuring that different problem requirements and constraints can be flexibly responded to.

[0162] Step S502: Perform a weighted analysis on the resource utilization efficiency, response time and execution cost of each executable path to obtain path comprehensive data.

[0163] Specifically, a weighted analysis is performed on each executable path, including but not limited to calculating the resource utilization efficiency (such as CPU, memory, etc.), expected response time, and execution cost (monetary cost or other forms of resource consumption) of the path in actual applications. By comprehensively analyzing the advantages and disadvantages of each path and quantifying the overall effectiveness of each path, clear quantitative indicators are provided for path selection, which facilitates comparison and decision-making, ensuring that the selected path is optimal in multiple dimensions.

[0164] Step S503: selecting the path with the highest value from the multiple executable paths as the execution path according to the path comprehensive data.

[0165] Specifically, ensure that the selected execution path is the optimal solution in terms of resource usage, time efficiency, and cost-effectiveness under current conditions. By selecting the optimal execution path, the efficiency and quality of problem solving can be improved while reducing unnecessary waste of resources.

[0166] The embodiment of the present invention not only improves the overall decision-making ability by selecting the optimal execution path from multiple possible solutions, but also ensures the efficiency and economy of resource calling, thereby optimizing the user problem solving process.

[0167] Figure 6 Schematic diagram of the self-verification intelligent reply method provided in the embodiment of the present application Figure 6 .like Figure 6 As shown, this embodiment describes in detail the process of generating the answer to the question by calling resources according to the execution path in the above embodiment, and the specific implementation method of the process includes the following steps:

[0168] Step S601: calling resources from a preset resource database according to the sequence of steps in the execution path.

[0169] Specifically, according to the selected optimal execution path, relevant resources are called step by step. The process may include API interfaces, database queries or other forms of external service requests. Each step of resource calling is strictly carried out in the order of steps in the execution path to ensure the logic and coherence of the entire process. By calling resources in an orderly manner, it can be ensured that each link can obtain the necessary information support, maximize resource utilization efficiency and result accuracy, and improve the speed of solving the entire problem.

[0170] Step S602: Acquire the data returned by the resource and integrate it to obtain resource information.

[0171] Specifically, for each resource call, the returned data will be received and processed. This data may be structured (such as JSON, XML format) or unstructured (such as text). These scattered data fragments need to be integrated to form a unified and complete resource information set. Through data integration, a complete view is formed to ensure that subsequent processing is based on a comprehensive information foundation.

[0172] Step S603: Perform data cleaning and information extraction on the resource information to obtain target information.

[0173] Specifically, data is cleaned by removing redundant or irrelevant data, correcting erroneous or inconsistent information, and other processes. At the same time, key content directly related to problem solving needs to be extracted from the cleaned data. Data cleaning and information extraction help improve the quality and relevance of data, ensuring that the extracted information is accurate and directly related to problem solving.

[0174] Step S604: Arrange the target information based on a preset answer format to obtain an answer to the question.

[0175] Specifically, according to the preset answer template or format requirements, the extracted target information is organized into a complete answer, ensuring the formatting and readability of the answer, making it easy to understand and use, while maintaining the integrity and accuracy of the information. Through formatting and organization, the answer can be made clearer and easier to understand, improving the user experience. At the same time, it is also convenient for subsequent answer verification.

[0176] The embodiment of the present invention optimizes and verifies each link in the process of calling resources to generate answers, and finally outputs answers that meet user needs, thereby improving the accuracy and efficiency of answer generation and ensuring the quality of the final output and user satisfaction.

[0177] Figure 7 Schematic diagram of the self-verification intelligent reply method provided in the embodiment of the present application Figure 6 .like Figure 7 As shown, this embodiment describes in detail the process of verifying the answer to the question according to the task goal and the problem constraint in the above embodiment, and the specific implementation method of the process includes the following steps:

[0178] Step S701: Check the answer to the question based on the task objectives and problem constraints to determine whether the answer to the question satisfies the problem constraints and task objectives.

[0179] Specifically, the generated answers are compared with the original task objectives and problem constraints, including checking whether the answer content completely covers all the key points required by the task objectives and whether it meets the various restrictions specified in the problem constraints (such as keyword matching, logical consistency, etc.). By using clear standards and conditions for verification, the quality of the answers is ensured and the answers are consistent with user needs and problem context.

[0180] Step S702: When the answer to the question satisfies the question constraints and task objectives, the answer to the question passes verification.

[0181] Specifically, if the answer to the question passes all the check items, the answer to the question is considered valid. The judgment process helps to screen out high-quality answers and avoid providing inaccurate or incomplete information to users.

[0182] Step S703: When the answer to the question does not satisfy the question constraints and / or task objectives, the question to be processed is input into a preset analysis model to improve the question content, and the step of analyzing the question content of the question to be processed to obtain the task objectives and keywords is returned.

[0183] Specifically, if the answer does not meet the problem constraints and / or task objectives, the problem to be processed is re-entered into the preset analysis model. The model is used to improve and re-analyze the problem content to obtain more accurate task objectives and keywords. The iterative improvement method is used to continuously optimize the problem understanding and answer generation process. Through the feedback loop mechanism, the entire process can be self-corrected and improved when encountering difficulties, thereby continuously improving problem-solving capabilities.

[0184] The embodiments of the present invention ensure that the answers fed back to users are accurate and relevant through strict verification steps, quickly identify and correct answers that do not meet the requirements during the verification process, and reduce the possibility of incorrect feedback. By providing high-quality answers and fast response times, the overall user experience is improved. At the same time, through iterative analysis and improvement of question content, continuous learning and improvement, the ability to solve future problems is improved. In this way, not only the quality of the answers is ensured, but also the intelligence and adaptability of the replies are improved through continuous self-improvement and adjustment.

[0185] Figure 8 This is a schematic diagram of the structure of a self-verifying intelligent reply device provided by an embodiment of the present invention. Figure 8 As shown, the self-verifying intelligent reply device 80 includes: an acquisition module 801, an analysis module 802, a constraint module 803, a query module 804, a generation module 805, a call module 806, a verification module 807, and a feedback module 808.

[0186] The acquisition module 801 is used to receive the pending issues sent by the client;

[0187] An analysis module 802 is used to analyze the content of the problem to be processed to obtain task objectives and keywords;

[0188] The constraint module 803 is used to generate problem constraints through task objectives, keywords and a preset information base;

[0189] A query module 804 is used to query a preset resource database to obtain available resources according to the problem constraints;

[0190] A generation module 805, for generating an execution path based on problem constraints and callable resources;

[0191] A calling module 806 is used to call resources according to the execution path to generate an answer to the question;

[0192] Verification module 807, used to verify the answer to the question according to the task objectives and problem constraints;

[0193] The feedback module 808 is used to send the verified answers to the questions to the client.

[0194] In one possible design, the analysis module 802 is specifically configured to:

[0195] Input the question content into the preset natural language processing model for parsing to obtain the task goal;

[0196] Semantically analyzing the question content to obtain multiple words;

[0197] Calculate the similarity between vocabulary and task objectives to obtain similar data;

[0198] Filter out multiple keywords from the vocabulary whose similar data is greater than a preset similarity threshold.

[0199] In a possible design, the constraint module 803 is specifically configured to:

[0200] According to the task objectives and keywords, relevant information is matched from the preset information library, and the relevant information includes environmental information and task guidance information;

[0201] The task objectives, keywords, environmental information and task guidance information are input into the preset multi-dimensional learning model to generate problem constraints.

[0202] In one possible design, the query module 804 is further configured to:

[0203] Get historical search logs;

[0204] Match the task objectives and problem constraints with historical search logs to obtain historical similar problems and the available resources and search success rate data corresponding to historical similar problems;

[0205] Calculate based on available resources, success rate data and preset weighted values ​​to obtain confidence data;

[0206] Comparing the confidence data with a preset confidence threshold;

[0207] When the confidence data is less than the preset confidence threshold, a problem warning is directly generated and problem feedback is given;

[0208] When the confidence data is greater than or equal to a preset confidence threshold, a query is performed in the preset resource database according to the problem constraints to obtain available resources.

[0209] In one possible design, the generating module 805 is specifically configured to:

[0210] Input the problem constraints and available resources into the preset analysis model to obtain multiple executable paths;

[0211] Perform weighted analysis on the resource utilization efficiency, response time, and execution cost of each executable path to obtain comprehensive path data;

[0212] According to the path synthesis data, the path with the highest value is selected from multiple executable paths as the execution path.

[0213] In one possible design, calling module 806 is specifically used to:

[0214] Calling resources from a preset resource database according to the sequence of steps in the execution path;

[0215] Get the data returned by the resource and integrate it to get the resource information;

[0216] Perform data cleaning and information extraction on resource information to obtain target information;

[0217] The target information is sorted based on the preset answer format to get the answer to the question.

[0218] In one possible design, the verification module 807 is specifically configured to:

[0219] Check the answer to the question based on the task objectives and problem constraints to determine whether the answer meets the problem constraints and task objectives;

[0220] When the answer to a question satisfies the problem constraints and task objectives, the answer to the question is verified;

[0221] When the answer to the question does not meet the question constraints and / or task objectives, the question to be processed is input into a preset analysis model to improve the question content, and the step of analyzing the question content of the question to be processed to obtain the task objectives and keywords is returned.

[0222] The self-verification intelligent reply device provided in this embodiment can be used to execute the above-mentioned self-verification intelligent reply method. Its implementation principle and technical effects are similar, and this embodiment will not be repeated here.

[0223] Fig. 9 A schematic diagram of the hardware structure of a self-verifying intelligent reply device provided by an embodiment of the present invention is shown in FIG. Fig. 9 As shown, the self-verifying intelligent reply device 90 includes: at least one processor 901 and a memory 902. Optionally, the self-verifying intelligent reply device 90 also includes a communication component 903. The processor 901, the memory 902 and the communication component 903 are connected via a bus 904.

[0224] In the specific implementation process, at least one processor 901 executes the computer execution instructions stored in the memory 902, so that at least one processor 901 executes the above self-verification intelligent reply method.

[0225] The communication component 903 can exchange data with the server.

[0226] The specific implementation process of the processor 901 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0227] In the above Fig. 9 In the illustrated embodiment, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0228] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage.

[0229] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0230] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above self-verification intelligent reply method is implemented.

[0231] The computer-readable storage medium mentioned above can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0232] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0233] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0234] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0235] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0236] If the function 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 technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0237] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0238] It should be further noted that, although the various steps in the flowchart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0239] It should be understood that the above-mentioned device embodiments are only illustrative, and the device of the present application can also be implemented in other ways. For example, the division of units / modules in the above-mentioned embodiments is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units, modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed.

[0240] In addition, unless otherwise specified, each functional unit / module in each embodiment of the present application may be integrated into one unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The above-mentioned integrated unit / module may be implemented in the form of hardware or in the form of a software program module.

[0241] If the integrated unit / module is implemented in the form of hardware, the hardware may be a digital circuit, an analog circuit, etc. The physical implementation of the hardware structure includes but is not limited to transistors, memristors, etc. If not otherwise specified, the processor may be any appropriate hardware processor, such as a CPU, a GPU, an FPGA, a DSP, and an ASIC, etc. If not otherwise specified, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc.

[0242] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0243] In the above embodiments, the description of each embodiment has its own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0244] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0245] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A self-verification intelligent reply method, characterized in that: include: Receive pending issues sent by the client; Analyze the problem content of the problem to be processed to obtain task objectives and keywords; Generate problem constraints through the task objectives, keywords and preset information base; According to the problem constraints, query the preset resource database to obtain available resources; generating an execution path based on the problem constraints and the callable resources; Calling resources according to the execution path to generate an answer to the question; Verifying the answer to the question according to the task objective and the problem constraints; Send the verified answer to the question to the client.

2. The method according to claim 1, characterized in that Analyze the content of the problem to be solved to obtain task objectives and keywords, including: Input the question content into a preset natural language processing model for parsing to obtain the task objective; Performing semantic analysis on the question content to obtain multiple words; Calculating similarity between the vocabulary and the task target to obtain similarity data; A plurality of keywords whose similarity data is greater than a preset similarity threshold are selected from the vocabulary.

3. The method according to claim 1, characterized in that Generate problem constraints through the task objectives, keywords and preset information base, including: According to the task objective and the keywords, relevant information is obtained from a preset information library by matching, wherein the relevant information includes environmental information and task guidance information; The task objectives, keywords, environmental information and task guidance information are input into a preset multi-dimensional learning model to generate problem constraints.

4. The method according to claim 1, characterized in that: Before querying the preset resource database to obtain the callable resources according to the problem constraints, the method further includes: Get historical search logs; Matching the task goal and the problem constraint with the historical search log to obtain historical similar problems and available resources and search success rate data corresponding to the historical similar problems; Calculating based on the available resources, the success rate data and a preset weighted value to obtain confidence data; Comparing the confidence data with a preset confidence threshold; When the confidence data is less than the preset confidence threshold, a problem warning is directly generated and problem feedback is performed; When the confidence data is greater than or equal to the preset confidence threshold, a query is performed in a preset resource database according to the problem constraints to obtain available resources.

5. The method according to claim 1, characterized in that Generating an execution path based on the problem constraints and the callable resources includes: Inputting the problem constraints and the callable resources into a preset analysis model to obtain multiple executable paths; Performing a weighted analysis on the resource utilization efficiency, response time and execution cost of each executable path to obtain path comprehensive data; A path with the highest value is selected from a plurality of executable paths according to the path comprehensive data as an execution path.

6. The method according to claim 1, characterized in that The resource call is performed according to the execution path to generate the answer to the question, including: Calling resources from the preset resource database according to the sequence of steps in the execution path; Obtaining the data returned by the resource and integrating it to obtain resource information; Performing data cleaning and information extraction on the resource information to obtain target information; The target information is sorted based on a preset answer format to obtain an answer to the question.

7. The method according to claim 1, characterized in that Verifying the answer to the question according to the task objective and the question constraints includes: Checking the answer to the question based on the task goal and the problem constraints to determine whether the answer to the question satisfies the problem constraints and the task goal; When the answer to the question satisfies the question constraints and the task objectives, the answer to the question passes verification; When the answer to the question does not satisfy the question constraints and / or the task objectives, the question to be processed is input into a preset analysis model to improve the question content, and the step of analyzing the question content of the question to be processed to obtain the task objectives and keywords is returned.

8. A self-verifying intelligent reply device, characterized in that: include: The acquisition module is used to receive pending issues sent by the client; An analysis module, used to analyze the problem content of the problem to be processed to obtain task objectives and keywords; A constraint module, used to generate problem constraints through the task objectives, keywords and a preset information library; A query module, used to query a preset resource database to obtain available resources according to the problem constraints; A generation module, used for generating an execution path based on the problem constraints and the callable resources; A calling module, used for calling resources according to the execution path to generate an answer to the question; A verification module, used to verify the answer to the question according to the task objective and the question constraints; The feedback module is used to send the verified answers to the questions to the client.

9. A self-verifying intelligent reply device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

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

Images