Question processing method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the field of artificial intelligence and the field of financial science and technology, and relates to an artificial intelligence-based question processing method, which comprises the following steps of: performing text conversion on question voice input by a user to obtain a first text; preprocessing the first text to obtain a second text; performing intention recognition on the second text based on a large language model to obtain a target intention; screening a target thinking chain from all thinking chains based on the target intention and the processing complexity of the second text; performing execution and reasoning processing according to the target thinking chain through a large language model, and outputting answer data corresponding to the target intention; and performing feedback processing on the user based on the answer data. The invention further provides a question processing device based on artificial intelligence, computer equipment and a storage medium. In addition, answer data may be stored in a blockchain. The method can be applied to a dialogue processing scene in the financial field, and based on the use of the large language model mounted with the thinking chain library, the answering accuracy of the questioning voice is effectively improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and the field of financial technology, and in particular to artificial intelligence-based question processing methods, devices, computer equipment and storage media. Background Art

[0002] With the rapid development of artificial intelligence technology, intelligent question-answering systems for human-computer interactive dialogues have been widely used in the financial field. For example, they can be used in insurance plan communication, home purchase plan communication, and so on.

[0003] At present, the intelligent question-answering systems used by financial enterprises, such as insurance enterprises, are usually intelligent question-answering systems based on large language models. The process of processing the question data input by users by such intelligent question-answering systems is usually as follows: the intelligent question-answering system uses the large language model to extract abstract features from the existing information to generate vector expressions, and stores them in the database; then the large language model is used to extract features from the question data raised by the user to generate vector expressions, and the content matching the question is searched in the database by vector recall, and finally the large language model is used to process the question and search results to obtain the corresponding answer. However, this question processing method has the following problems: due to the diverse formats of the questions input by users in the usage scenarios, the vector recall accuracy of the large language model is relatively low for complex questions, resulting in low accuracy of the intelligent question-answering system's answers to the question data. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a question processing method, apparatus, computer device and storage medium based on artificial intelligence to solve the technical problem that the existing question processing method has low accuracy in answering question data due to the various formats of questions input by users in usage scenarios and the low vector recall accuracy of the large language model for complex questions.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a question processing method based on artificial intelligence, which adopts the following technical solutions:

[0006] Performing text conversion processing on the received question voice input by the user to obtain a first text;

[0007] Preprocessing the first text to obtain a second text;

[0008] Performing intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is loaded with a thought chain library generated based on a preset construction method, and the thought chain library includes a certain number of thought chains;

[0009] determining a processing complexity corresponding to the second text;

[0010] Based on the target intention and the processing complexity, a target thinking chain is selected from all thinking chains contained in the thinking chain library;

[0011] Through the large language model, execution and reasoning processing are performed according to the target thought chain, and answer data corresponding to the target intention is output;

[0012] Feedback processing is performed on the user based on the answer data.

[0013] Furthermore, the step of selecting a target thinking chain from all thinking chains contained in the thinking chain library based on the target intention and the processing complexity specifically includes:

[0014] Acquire the intention identification information corresponding to each of the thought chains;

[0015] Based on the intention identification information, acquiring a first thought chain matching the target intention from all the thought chains;

[0016] Determining whether the number of the first thought chain includes multiple ones;

[0017] If so, filtering out a second thinking chain that matches the processing complexity from the first thinking chain based on a preset filtering rule;

[0018] The second thought chain is used as the target thought chain.

[0019] Furthermore, the step of selecting a second chain of thought matching the processing complexity from the first chain of thought based on a preset screening rule specifically includes:

[0020] Determining whether the processing complexity is high complexity;

[0021] If the processing complexity is high complexity, a third thinking chain with the largest chain length is selected from all the first thinking chains, and the third thinking chain is used as the second thinking chain;

[0022] If the processing complexity is not high complexity, determining whether the processing complexity is low complexity;

[0023] If the processing complexity is low complexity, a fourth thinking chain with the smallest chain length is selected from all the first thinking chains, and the fourth thinking chain is used as the second thinking chain;

[0024] If the processing complexity is not low complexity, the processing complexity is determined to be medium complexity, and the usage frequency value of the fifth thinking chain is obtained; wherein the fifth thinking chain is all other thinking chains in the first thinking chain except the third thinking chain and the fourth thinking chain;

[0025] A sixth chain of thinking with the highest frequency of use is selected from the fifth chain of thinking, and the sixth chain of thinking is used as the second chain of thinking.

[0026] Furthermore, the step of determining the processing complexity corresponding to the second text specifically includes:

[0027] Performing sentence processing on the second text to obtain corresponding sentences, and counting a first number of the sentences;

[0028] Performing word segmentation processing on the second text to obtain corresponding words, and counting a second number of the words;

[0029] Based on the first number and the second number, a complexity score of the second text is calculated based on a preset calculation strategy;

[0030] A query process is performed on a preset mapping data table based on the complex score, so as to query the processing complexity corresponding to the complex score from the mapping data table.

[0031] Furthermore, the step of calculating the complexity score of the second text based on the first number and the second number and based on a preset calculation strategy specifically includes:

[0032] Determine the preset weighting algorithm;

[0033] Determine a first weight corresponding to the number of sentences based on the weight obtaining algorithm;

[0034] Determine a second weight corresponding to the number of words based on the weight obtaining algorithm;

[0035] Based on the first weight and the second weight, calling a preset score calculation formula to calculate the first quantity and the second quantity to obtain a corresponding calculation result;

[0036] The calculation result is used as the complexity score of the second text.

[0037] Furthermore, before the step of performing intent recognition on the second text based on the preset large language model to obtain a target intent corresponding to the second text, the method further includes:

[0038] Determine whether a thinking chain library construction request triggered by a specified user is received;

[0039] If yes, display the preset build page;

[0040] Receiving thought chain data input by the designated user on the construction page;

[0041] Building a corresponding thinking chain library based on the thinking chain data;

[0042] The thought chain library is stored and processed.

[0043] Furthermore, after the step of performing feedback processing on the user based on the answer data, the method further includes:

[0044] Obtain pre-collected performance test data;

[0045] Performing a performance evaluation on the large language model based on the performance test data to obtain corresponding model performance data;

[0046] Acquire basic model performance data corresponding to the large language model;

[0047] Comparing and analyzing the model performance data with the basic model performance data to obtain corresponding comparison and analysis results;

[0048] The large language model is adjusted based on the comparison and analysis results.

[0049] In order to solve the above technical problems, the embodiment of the present application further provides a question processing device based on artificial intelligence, which adopts the following technical solution:

[0050] A conversion module, used for performing text conversion processing on the received question voice input by the user to obtain a first text;

[0051] A preprocessing module, used for preprocessing the first text to obtain a second text;

[0052] A recognition module, configured to perform intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is loaded with a thought chain library generated based on a preset construction method, and the thought chain library includes a certain number of thought chains;

[0053] A determination module, used to determine a processing complexity corresponding to the second text;

[0054] A screening module, used for screening out a target thinking chain from all thinking chains contained in the thinking chain library based on the target intention and the processing complexity;

[0055] A first processing module is used to perform execution and reasoning processing according to the target thought chain through the large language model, and output answer data corresponding to the target intention;

[0056] The second processing module is used to perform feedback processing on the user based on the answer data.

[0057] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0058] Performing text conversion processing on the received question voice input by the user to obtain a first text;

[0059] Preprocessing the first text to obtain a second text;

[0060] Performing intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is loaded with a thought chain library generated based on a preset construction method, and the thought chain library includes a certain number of thought chains;

[0061] determining a processing complexity corresponding to the second text;

[0062] Based on the target intention and the processing complexity, a target thinking chain is selected from all thinking chains contained in the thinking chain library;

[0063] Through the large language model, execution and reasoning processing are performed according to the target thought chain, and answer data corresponding to the target intention is output;

[0064] Feedback processing is performed on the user based on the answer data.

[0065] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0066] Performing text conversion processing on the received question voice input by the user to obtain a first text;

[0067] Preprocessing the first text to obtain a second text;

[0068] Performing intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is loaded with a thought chain library generated based on a preset construction method, and the thought chain library includes a certain number of thought chains;

[0069] determining a processing complexity corresponding to the second text;

[0070] Based on the target intention and the processing complexity, a target thinking chain is selected from all thinking chains contained in the thinking chain library;

[0071] Through the large language model, execution and reasoning processing are performed according to the target thought chain, and answer data corresponding to the target intention is output;

[0072] Feedback processing is performed on the user based on the answer data.

[0073] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0074] The present application first performs text conversion processing on the question voice input by the user to obtain the first text; then preprocesses the first text to obtain the second text; then performs intent recognition on the second text based on the preset large language model to obtain the target intent corresponding to the second text; subsequently determines the processing complexity corresponding to the second text; further based on the target intent and the processing complexity, selects the target thinking chain from all the thinking chains contained in the thinking chain library; and executes and infers the target thinking chain through the large language model, outputs the answer data corresponding to the target intent; finally, based on the answer data, the user is fed back. After the text conversion and preprocessing of the question voice input by the user, the large language model with the thinking chain library generated based on the preset construction method will first perform intent recognition on the second text to obtain the target intent, and then intelligently selects the matching target thinking chain from all the thinking chains contained in the thinking chain library according to the processing complexity and target intent corresponding to the second text, and finally executes and infers the obtained target thinking chain through the large language model, outputs the answer data corresponding to the target intent and feeds it back to the user. The present application uses an optimized large language model mounted with a thinking chain library generated based on a preset construction method to perform response processing for user input question voice, which can achieve accurate and controllable answer data output for user input question voice, and effectively improves the accuracy of the answer to the question voice. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0076] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0077] Figure 2 A flowchart of an embodiment of an artificial intelligence-based question processing method according to the present application;

[0078] Figure 3 is a structural schematic diagram of an embodiment of an artificial intelligence-based question processing device according to the present application;

[0079] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0080] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0081] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0082] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0083] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0084] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0085] The terminal devices 101, 102, and 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, and the like.

[0086] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0087] It should be noted that the artificial intelligence-based question processing method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based question processing device is generally set in the server / terminal device.

[0088] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0089] Continue to refer Figure 2 , showing a flow chart of an embodiment of an artificial intelligence-based question processing method according to the present application. According to different needs, the order of the steps in the flowchart can be changed, and some steps can be omitted. The artificial intelligence-based question processing method provided in the embodiment of the present application can be applied to any scenario that requires dialogue interaction, and the artificial intelligence-based question processing method can be applied to products in these scenarios, for example, insurance plan communication, home purchase plan communication, and treatment plan communication in financial insurance scenarios or digital medical scenarios. The artificial intelligence-based question processing method comprises the following steps:

[0090] Step S201 : performing text conversion processing on the received question voice input by the user to obtain a first text.

[0091] In this embodiment, the electronic device (e.g. Figure 1The server / terminal device shown in the figure) can obtain the question voice through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, Wi-Fi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (Ultra Wi-Fi Band) connection, and other wireless connection methods currently known or to be developed in the future. Among them, the above-mentioned question voice is the question data in a voice format that the user inputs through voice to interact with the large language model. The question voice can be a question in any format, such as questions with different sentence patterns or questions containing redundant information; the question voice can also be a complex question. Exemplarily, the content of the question voice may include: "Because I accompanied the customer to drink together and drank gastric bleeding, is it considered a work injury?" In addition, the question voice can be converted into a first text by using automatic speech recognition (ASR) technology.

[0092] Step S202: preprocess the first text to obtain a second text.

[0093] In this embodiment, the above-mentioned preprocessing may include text cleaning processing, and the content of the text cleaning processing may include: removing irrelevant information such as noise, special characters, HTML tags, etc. in the first text.

[0094] Step S203, performing intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is mounted with a thinking chain library generated based on a preset construction method, and the thinking chain library includes a certain number of thinking chains.

[0095] In this embodiment, after the second text is input into the large language model, the second text is deeply analyzed by utilizing the deep learning architecture of the large language model to understand its meaning and context, and then the user's intention is obtained according to the text content recognition of the second text, such as querying information, expressing emotions, requesting services, etc. Among them, the thinking chain library is integrated as an optional module of the large language model to ensure that the large language model can support the mounting of the thinking chain library. In addition, the above-mentioned preset construction method may include constructing a thinking chain library through manual autonomous editing, or may also include a thinking chain extracted from historical cases by a large language model, and the thinking chain is manually quality-checked to construct a thinking chain library, and the preset construction method is different from the existing method of generating thinking chains by the large language model itself. In addition, there is no specific limitation on the value of the above-mentioned certain number, and it can be set according to actual construction needs.

[0096] Step S204: determine the processing complexity corresponding to the second text.

[0097] In this embodiment, the specific implementation process of determining the processing complexity corresponding to the second text will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0098] Step S205, based on the target intention and the processing complexity, a target thinking chain is screened out from all thinking chains contained in the thinking chain library.

[0099] In this embodiment, the specific implementation process of selecting the target thinking chain from all the thinking chains contained in the thinking chain library based on the target intention and the processing complexity will be described in further detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0100] Step S206, through the large language model, execution and reasoning processing are performed according to the target thinking chain, and answer data corresponding to the target intention is output.

[0101] In this embodiment, after selecting the target thinking chain that matches the target intent and the processing complexity from all the thinking chains contained in the thinking chain library, the subsequent large language model will execute and reason according to the target thinking chain, and output the answer data corresponding to the target intent, so as to output accurate and controllable answer data for the question voice input by the user, and based on the use of the thinking chain library, it can ensure that each round of decision-making in the conversation can achieve accurate and controllable answers.

[0102] Step S207: performing feedback processing on the user based on the answer data.

[0103] In this embodiment, after obtaining the above answer data, the answer data can be further subjected to grammar detection and fluency optimization processing to obtain target answer data, and the target answer data will be used to perform feedback processing on the user later. The target answer data is data in text format. If the generated target answer data needs to be output to the user in voice form, the text is further converted into corresponding answer voice data using text-to-speech technology (TTS), and the answer voice data is returned to the user, thereby efficiently, accurately and naturally completing the dialogue interaction processing corresponding to the question voice input by the user.

[0104] After the text conversion and preprocessing of the question voice input by the user, the present application will first use a large language model mounted with a thinking chain library generated based on a preset construction method to perform intent recognition on the second text to obtain the target intent, and then intelligently screen out the matching target thinking chain from all the thinking chains contained in the thinking chain library according to the processing complexity and target intent corresponding to the second text, and finally execute and reason according to the obtained target thinking chain through the large language model, output the answer data corresponding to the target intent and feed it back to the user. The present application uses an optimized large language model mounted with a thinking chain library generated based on a preset construction method to perform response processing for the question voice input by the user, which can realize the output of accurate and controllable answer data for the question voice input by the user, and effectively improves the accuracy of the answer to the question voice.

[0105] In some optional implementations, step S205 includes the following steps:

[0106] Acquire the intention identification information corresponding to each of the thought chains.

[0107] In this embodiment, in the process of constructing the thinking chain library, for each thinking chain in the thinking chain library, the intention identification information corresponding to each thinking chain is generated according to the problem type information corresponding to each thinking chain. There may be more than one intention label information corresponding to each thinking chain.

[0108] Based on the intention identification information, a first thought chain matching the target intention is obtained from all the thought chains.

[0109] In this embodiment, the above-mentioned target intent can be matched with the intent identification information of each of the above-mentioned thinking chains respectively to determine the matching degree between the above-mentioned target intent and each of the above-mentioned thinking chains, and then a designated thinking chain with a matching degree greater than a preset matching threshold is screened out from all the thinking chains, and the designated thinking chain is used as the first thinking chain that matches the target intent. Among them, a similarity measurement algorithm can be used to calculate the matching degree between the target intent and each of the above-mentioned thinking chains. And there is no specific limitation on the selection of the similarity measurement algorithm, which can be set according to actual business usage requirements.

[0110] Determine whether the number of the first thought chain includes multiple ones.

[0111] In this embodiment, the number of the first thought chain may include one or more. If the number of the first thought chain is one, the first thought chain is directly used as the target thought chain.

[0112] If so, a second thinking chain matching the processing complexity is screened out from the first thinking chain based on a preset screening rule.

[0113] In this embodiment, the specific implementation process of filtering out the second thinking chain that matches the processing complexity from the first thinking chain based on the preset filtering rules will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0114] The second thought chain is used as the target thought chain.

[0115] The present application can adaptively and accurately screen out the target thought chain for processing the answer processing of the question voice input by the user from all the thought chains contained in the thought chain library according to the target intention corresponding to the question voice and the processing complexity of the question voice, thereby realizing more effective use of the thought chain technology to improve the performance of the large language model in question feedback processing, thereby improving the processing efficiency of the large language model in feedback processing of the question voice input by the user, and ensuring the data accuracy of the generated answer data corresponding to the target intention.

[0116] In some optional implementations of this embodiment, the filtering out the second thought chain matching the processing complexity from the first thought chain based on a preset filtering rule includes the following steps:

[0117] It is determined whether the processing complexity is high complexity.

[0118] In this embodiment, the above processing complexity may include high complexity, medium complexity and low complexity.

[0119] If the processing complexity is high complexity, a third thinking chain with the largest chain length is selected from all the first thinking chains, and the third thinking chain is used as the second thinking chain.

[0120] In this embodiment, if the processing complexity is high complexity, it is determined that the processing task corresponding to the question voice belongs to a high complexity task, and then the third thought chain with the largest chain length is intelligently screened out from all the first thought chains, and the third thought chain is used as the target thought chain for processing the answer to the question voice input by the user. For high complexity tasks, by using the thought chain with the largest chain length, more reasoning steps and more detailed explanations can be provided, thereby effectively improving the data accuracy of the answer data generated by the large language model corresponding to the target intent.

[0121] If the processing complexity is not high complexity, it is determined whether the processing complexity is low complexity.

[0122] In this embodiment, if the processing complexity is not high complexity, the processing complexity may be medium complexity or low complexity.

[0123] If the processing complexity is low complexity, a fourth thinking chain with the smallest chain length is selected from all the first thinking chains, and the fourth thinking chain is used as the second thinking chain.

[0124] In this embodiment, if the processing complexity is low complexity, it is determined that the processing task corresponding to the question voice is a low complexity task, and then the fourth thought chain with the smallest chain length is intelligently screened out from all the first thought chains, and the fourth thought chain is used as the target thought chain for processing the answer to the question voice input by the user. For low complexity tasks, by using the thought chain with the smallest chain length, because the question input by the user is relatively simple, it does not require too many intermediate steps, thereby effectively improving the performance of the large language model in question feedback processing.

[0125] If the processing complexity is not low complexity, the processing complexity is determined to be medium complexity, and the usage frequency value of the fifth thinking chain is obtained.

[0126] In this embodiment, the fifth chain of thought is all the other chains of thought in the first chain of thought except the third chain of thought and the fourth chain of thought. The usage frequency value can be obtained by calculating the quotient between the total number of times the fifth chain of thought is used in one month and 30.

[0127] A sixth chain of thinking with the highest frequency of use is selected from the fifth chain of thinking, and the sixth chain of thinking is used as the second chain of thinking.

[0128] In this embodiment, the usage frequency values ​​of all the obtained fifth thought chains can be numerically compared to determine the designated usage frequency value with the highest value, and then the sixth thought chain corresponding to the designated usage frequency value can be used as the second thought chain.

[0129] After obtaining the first chain of thought that matches the target intent from all the chains of thought based on the intent identification information, if it is detected that the number of the first chains of thought includes multiple chains, the application will intelligently further select the second chain of thought for processing the answer processing of the question voice input by the user from the first chain of thought according to the processing complexity of the question voice, thereby ensuring the accuracy of the obtained second chain of thought. And it realizes more effective use of the chain of thought technology to improve the performance of the large language model in question feedback processing, thereby improving the processing efficiency of the large language model in the feedback processing of the question voice input by the user, and ensuring the data accuracy of the generated answer data corresponding to the target intent.

[0130] In some optional implementations, step S204 includes the following steps:

[0131] The second text is segmented into sentences to obtain corresponding sentences, and a first number of the sentences is counted.

[0132] In this embodiment, the voice pause position information corresponding to the question voice input by the user can be obtained, and the second text can be segmented based on the voice pause position information to obtain the corresponding sentences.

[0133] The second text is segmented to obtain corresponding words, and a second number of the words is counted.

[0134] In this embodiment, the second text may be segmented by using a general segmentation tool, such as a ji eba tool, so as to segment the second text into words.

[0135] Based on the first number and the second number, a complexity score of the second text is calculated based on a preset calculation strategy.

[0136] In this embodiment, the specific implementation process of calculating the complexity score of the second text based on the first number and the second number based on a preset calculation strategy will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0137] A query process is performed on a preset mapping data table based on the complex score, so as to query the processing complexity corresponding to the complex score from the mapping data table.

[0138] In this embodiment, the mapping data table is a data table that stores a one-to-one correspondence between score ranges and complexities and is pre-constructed according to an actual complexity statistical analysis request. Specifically, the specified score range corresponding to the complexity score can be determined from the mapping data table first, and then the specified complexity corresponding to the specified score range can be queried from the mapping database, and the specified complexity can be used as the processing complexity corresponding to the complexity score. The complexity content can include high complexity, medium complexity and low complexity.

[0139] The present application performs sentence segmentation and word segmentation on the second text, and then calculates the complexity score of the second text according to the first number of sentences and the second number of words obtained and the calculation strategy. Then, based on the use of a mapping data table, it is possible to quickly and accurately query the processing complexity corresponding to the second text, thereby ensuring the data accuracy of the obtained processing complexity.

[0140] In some optional implementations, the step of calculating the complexity score of the second text based on the first number and the second number and based on a preset calculation strategy includes the following steps:

[0141] Determine the preset weighting algorithm.

[0142] In this embodiment, there is no specific limitation on the selection of the above-mentioned weight calculation algorithm. For example, any one of the various algorithms such as hierarchical analysis method, entropy method, principal component analysis method, independence weight method, factor analysis method, information weight method, etc. can be used, or two or three algorithms can be randomly selected from a variety of algorithms and combined as the above-mentioned weight selection algorithm.

[0143] A first weight corresponding to the number of sentences is determined based on the weight obtaining algorithm.

[0144] In this embodiment, the weight generation process may be performed on the number of sentences by using the weight calculation algorithm described above to generate a first weight corresponding to the number of sentences.

[0145] A second weight corresponding to the number of words is determined based on the weight obtaining algorithm.

[0146] In this embodiment, the weight generation process can be performed on the number of words by using the above-mentioned weight obtaining algorithm to obtain a second weight corresponding to the number of words.

[0147] Based on the first weight and the second weight, a preset score calculation formula is called to calculate the first quantity and the second quantity to obtain a corresponding calculation result.

[0148] In this embodiment, the score calculation formula may specifically adopt a weighted sum formula, and the first weight, the second weight, the first quantity, and the second quantity may be substituted into corresponding positions in the score calculation formula for calculation, thereby obtaining a corresponding calculation result.

[0149] The calculation result is used as the complexity score of the second text.

[0150] The present application can effectively ensure the rationality and accuracy of the data of the first weight corresponding to the number of sentences and the second weight corresponding to the number of words by using a weight obtaining algorithm. Then, based on the obtained first weight and second weight, the first number and the second number are calculated and processed using a score calculation formula, so that the complex score of the second text can be generated quickly and accurately, the efficiency of complex score generation is improved, and the accuracy of the obtained complex score is guaranteed.

[0151] In some optional implementations of this embodiment, before step S203, the electronic device may further perform the following steps:

[0152] Determine whether a thinking chain library construction request triggered by a specified user is received.

[0153] In this embodiment, the designated user is a staff member related to the operation and maintenance of the large language model. The thought chain library construction request is a processing request generated according to the actual business demand of manually constructing a thought chain triggered by the designated user.

[0154] If yes, show the default build page.

[0155] In this embodiment, the above-mentioned construction page is a pre-built editable page for assisting relevant users in building a thinking chain library. Among them, a construction reminder information box can also be set in the above-mentioned construction page, and the preset thinking chain construction guidance information is displayed in the construction reminder information box. The above-mentioned thinking chain construction guidance information is a guidance information for guiding relevant users on how to construct and generate standardized thinking chains according to the actual construction generation requirements of the thinking chain. In addition, the construction of the thinking chain library requires the generation of multiple thinking chains. The thinking chain consists of four parts: Prompt, SOP, knowledge, and tools, where Prompt is a model prompt instruction; SOP is a standard operating procedure for solving problems under a certain type of intent query; SOP execution depends on external knowledge and tools (API). For the generation of each thinking chain, a data input section corresponding to the four parts of Prompt, SOP, knowledge, and tools is pre-built. After the relevant user completes the data input of a thinking chain, the relevant user can complete the construction of the thinking chain by clicking Submit. After that, the relevant data of the next thinking chain input by the relevant user is received and the construction of the corresponding thinking chain is completed. Finally, all the generated thinking chains are integrated to obtain a thinking chain library. In addition, QA knowledge, document knowledge, graph knowledge, and table knowledge can be used to enhance the knowledge retrieval of the knowledge in the thinking chain to improve the knowledge richness of the thinking chain library.

[0156] Receive the thought chain data input by the designated user on the construction page.

[0157] In this embodiment, the user can input specific thinking chain data corresponding to the thinking chain to be constructed in the construction page according to the actual thinking chain generation requirements.

[0158] A corresponding thinking chain library is constructed based on the thinking chain data.

[0159] In this embodiment, the corresponding thinking chain can be constructed based on the thinking chain data input by the designated user, and then all the obtained thinking chains are integrated to obtain the corresponding thinking chain library. In addition, after the construction of the thinking chain library is completed, the logical relationship and reasoning process in the thinking chain in the thinking chain library can be further checked to see if they are reasonable, and the thinking chain can be corrected and optimized based on expert opinions or user feedback.

[0160] The thought chain library is stored and processed.

[0161] In this embodiment, there is no specific limitation on the storage method of the thinking chain library, which can be set according to actual usage requirements, for example, blockchain storage, local database storage, cloud server storage, etc.

[0162] Through this application, based on the construction page and the use of thinking chain data input by the user, it is possible to automatically, quickly and accurately complete the construction process of a thinking chain library that meets the user's personal construction needs, which can effectively reduce the workload of building a thinking chain library, simplify the processing flow of building a thinking chain library, and improve the processing intelligence and efficiency of building a thinking chain library.

[0163] In some optional implementations of this embodiment, the above-mentioned thinking chain library is not limited to manual writing, and can also use the thinking chain extracted from historical cases by the large language model, and obtain it through manual quality inspection. The specific implementation process may include: 1) First, collect a large amount of historical case data previously processed by the large language model, including case descriptions, case processing results, relevant processing basis, etc. Then pre-process the historical case data, including text cleaning, word segmentation, removal of stop words, etc., so that the large language model can better understand and process these historical case data. 2) Use historical case data to train the large language model, so that the large language model learns how to extract key information from case data, understand the logical relationship and reasoning process of the case. 3) Design extraction rules for extracting thinking chains from texts generated by the large language model. The extraction rules can be set based on factors such as language structure and logical reasoning. 4) Evaluate and optimize the extracted thinking chains. Manual evaluation can be used, or some automated indicators can be designed to measure the quality of the thinking chain. According to the evaluation results, fine-tune or optimize the large language model to improve the accuracy and completeness of thinking chain extraction. For each historical case, let the large language model generate relevant text description or analysis, and then use the above extraction rules to extract the chain of thought from the text.

[0164] In some optional implementations of this embodiment, after step S207, the electronic device may further perform the following steps:

[0165] Get pre-collected performance test data.

[0166] In this embodiment, the performance test data is pre-collected speech data for performance testing of the large language model. The same performance test data is used in advance to perform performance testing on the initial large language model without the above-mentioned thinking chain library, and the basic model performance data corresponding to the initial large language model is collected.

[0167] A performance evaluation is performed on the large language model based on the performance test data to obtain corresponding model performance data.

[0168] In this embodiment, the large language model is a model after the initial large language model is mounted with a thought chain library generated based on a preset construction method. The performance test data is input into the large language model, and the model performance data corresponding to the preset key performance indicators generated by the large language model during the data processing of the performance test data is collected.

[0169] Basic model performance data corresponding to the large language model is obtained.

[0170] In this embodiment, key performance indicators for reflecting the model performance of the large language model are determined in advance, such as accuracy, F1 score, recall rate, etc., and then before the large language model is mounted on the above-mentioned thinking chain library, the large language model at this time is called the initial large language model, and the above-mentioned performance test data is used to perform a performance test on the initial large language model, and the baseline values ​​of the key performance indicators are recorded to obtain the corresponding model performance data.

[0171] The model performance data is compared and analyzed with the basic model performance data to obtain corresponding comparison and analysis results.

[0172] In this embodiment, the model performance data is compared and analyzed with the basic model performance data to evaluate whether the model effect of the large language model is improved after the thinking chain library is mounted. The above comparison and analysis results include: the model performance data is better than the basic model performance data, or the model performance data is worse than the basic model performance data.

[0173] The large language model is adjusted based on the comparison and analysis results.

[0174] In this embodiment, if the model performance data obtained by the large language model after mounting the thinking chain library is improved compared with the basic model performance data, that is, the model performance data is better than the basic model performance data, then it is determined that mounting the thinking chain library is an effective model optimization strategy for the large language model, and the large language model will not be subsequently processed. In addition, if the model performance data obtained by the large language model after mounting the thinking chain library shows a performance decline or an insignificant performance improvement compared with the basic model performance data, that is, the model performance data is inferior to the basic model performance data, then it is determined that mounting the thinking chain library is an effective model optimization strategy, then it is determined that mounting the thinking chain library is an invalid model optimization strategy for the large language model, and then the large language model will be adjusted based on the adjustment strategy. Specifically, if the performance of the large language model is not significantly improved after mounting the thinking chain library, try to increase the length or complexity of the thinking chain in the thinking chain library; if the performance of the large language model declines after mounting the thinking chain library, try to consider reducing the length or complexity of the chain.

[0175] After obtaining the large language model with the thinking chain library mounted on it, the present application will also intelligently use the same performance test data to perform performance evaluation on the large language model, and compare and analyze the obtained model performance data with the basic model performance data to monitor the model performance changes of the large language model before and after the thinking chain library is mounted, and then adaptively adjust the large language model according to the obtained comparison and analysis results, which is conducive to more effectively utilizing the thinking chain technology to improve the performance of the large language model in conversational speech processing tasks.

[0176] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0177] It should be emphasized that in order to further ensure the privacy and security of the above answer data, the above answer data can also be stored in a blockchain node.

[0178] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0179] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0180] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0181] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0182] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings 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 their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0183] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a question processing device based on artificial intelligence, and the embodiment of the device is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0184] like Figure 3As shown, the artificial intelligence-based question processing device 300 described in this embodiment includes: a conversion module 301, a preprocessing module 302, a recognition module 303, a determination module 304, a screening module 305, a first processing module 306 and a second processing module 307. Among them:

[0185] The conversion module 301 is used to convert the received question voice input by the user into text to obtain a first text;

[0186] A preprocessing module 302, configured to preprocess the first text to obtain a second text;

[0187] The recognition module 303 is used to perform intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is mounted with a thought chain library generated based on a preset construction method, and the thought chain library includes a certain number of thought chains;

[0188] A determination module 304, configured to determine a processing complexity corresponding to the second text;

[0189] A screening module 305 is used to screen out a target thinking chain from all thinking chains contained in the thinking chain library based on the target intention and the processing complexity;

[0190] The first processing module 306 is used to perform execution and reasoning processing according to the target thought chain through the large language model, and output answer data corresponding to the target intention;

[0191] The second processing module 307 is used to perform feedback processing on the user based on the answer data.

[0192] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the question processing method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0193] In some optional implementations of this embodiment, the screening module 305 includes:

[0194] A first acquisition submodule is used to acquire the intention identification information corresponding to each thought chain;

[0195] A second acquisition submodule is used to acquire a first thought chain matching the target intention from all the thought chains based on the intention identification information;

[0196] A first judgment submodule is used to judge whether the number of the first thought chain includes multiple ones;

[0197] a screening submodule, for, if yes, screening out a second thinking chain matching the processing complexity from the first thinking chain based on a preset screening rule;

[0198] The first determining submodule is used to take the second thinking chain as the target thinking chain.

[0199] In some optional implementations of this embodiment, the screening submodule includes:

[0200] A judging unit, used to judge whether the processing complexity is high complexity;

[0201] A first screening unit is configured to screen out a third thinking chain with the largest chain length from all the first thinking chains if the processing complexity is high complexity, and use the third thinking chain as the second thinking chain;

[0202] A second judgment unit, configured to judge whether the processing complexity is low complexity if the processing complexity is not high complexity;

[0203] a second screening unit, configured to screen out a fourth thinking chain with the smallest chain length from all the first thinking chains if the processing complexity is low complexity, and use the fourth thinking chain as the second thinking chain;

[0204] an acquisition unit, configured to determine that the processing complexity is of medium complexity if the processing complexity is not of low complexity, and acquire a usage frequency value of a fifth chain of thought; wherein the fifth chain of thought is all other chains of thought in the first chains of thought except the third chain of thought and the fourth chain of thought;

[0205] The third screening unit is used to screen out a sixth thinking chain with the highest usage frequency from the fifth thinking chain, and use the sixth thinking chain as the second thinking chain.

[0206] In some optional implementations of this embodiment, the determining module 304 includes:

[0207] a sentence segmentation module, configured to perform sentence segmentation processing on the second text to obtain corresponding sentences, and to count a first number of the sentences;

[0208] A word segmentation submodule, configured to perform word segmentation processing on the second text to obtain corresponding words, and to count a second number of the words;

[0209] A calculation submodule, configured to calculate a complexity score of the second text based on the first number and the second number and based on a preset calculation strategy;

[0210] The query submodule is used to query a preset mapping data table based on the complex score, so as to query the processing complexity corresponding to the complex score from the mapping data table.

[0211] In some optional implementations of this embodiment, the calculation submodule includes:

[0212] A first determining unit, used to determine a preset weight obtaining algorithm;

[0213] A second determining unit, configured to determine a first weight corresponding to the number of sentences based on the weight obtaining algorithm;

[0214] A third determining unit, configured to determine a second weight corresponding to the number of words based on the weight obtaining algorithm;

[0215] a calculation unit, configured to calculate the first quantity and the second quantity based on the first weight and the second weight and to obtain a corresponding calculation result by calling a preset score calculation formula;

[0216] The fourth determining unit is configured to use the calculation result as the complexity score of the second text.

[0217] In some optional implementations of this embodiment, the question processing device based on artificial intelligence further includes:

[0218] A judgment module is used to judge whether a thinking chain library construction request triggered by a specified user is received;

[0219] Display module, used to display the preset construction page if yes;

[0220] A receiving module, used for receiving the thought chain data input by the designated user on the construction page;

[0221] A construction module, used for constructing a corresponding thinking chain library based on the thinking chain data;

[0222] The storage module is used to store and process the thought chain library.

[0223] In some optional implementations of this embodiment, the question processing device based on artificial intelligence further includes:

[0224] A first acquisition module is used to acquire pre-collected performance test data;

[0225] A module, configured to perform a performance evaluation on the large language model based on the performance test data to obtain corresponding model performance data;

[0226] A second acquisition module, used to acquire basic model performance data corresponding to the large language model;

[0227] An analysis module, used to compare and analyze the model performance data with the basic model performance data to obtain corresponding comparison and analysis results;

[0228] An adjustment module is used to adjust the large language model based on the comparison and analysis results.

[0229] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0230] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0231] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0232] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the question processing method based on artificial intelligence, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0233] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions or process data stored in the memory 41, such as computer-readable instructions for running the artificial intelligence-based question processing method.

[0234] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0235] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned artificial intelligence-based question processing method.

[0236] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0237] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A question processing method based on artificial intelligence, characterized in that: The steps include: Performing text conversion processing on the received question voice input by the user to obtain a first text; Preprocessing the first text to obtain a second text; Performing intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is loaded with a thought chain library generated based on a preset construction method, and the thought chain library includes a certain number of thought chains; determining a processing complexity corresponding to the second text; Based on the target intention and the processing complexity, a target thinking chain is selected from all thinking chains contained in the thinking chain library; Through the large language model, execution and reasoning processing are performed according to the target thought chain, and answer data corresponding to the target intention is output; Feedback processing is performed on the user based on the answer data.

2. The question processing method based on artificial intelligence according to claim 1, characterized in that: The step of selecting a target thinking chain from all thinking chains contained in the thinking chain library based on the target intention and the processing complexity specifically includes: Acquire the intention identification information corresponding to each of the thought chains; Based on the intention identification information, acquiring a first thought chain matching the target intention from all the thought chains; Determining whether the number of the first thought chain includes multiple ones; If so, filtering out a second thinking chain that matches the processing complexity from the first thinking chain based on a preset filtering rule; The second thought chain is used as the target thought chain.

3. The question processing method based on artificial intelligence according to claim 2 is characterized in that: The step of selecting a second thought chain matching the processing complexity from the first thought chain based on a preset screening rule specifically includes: Determining whether the processing complexity is high complexity; If the processing complexity is high complexity, a third thinking chain with the largest chain length is selected from all the first thinking chains, and the third thinking chain is used as the second thinking chain; If the processing complexity is not high complexity, determining whether the processing complexity is low complexity; If the processing complexity is low complexity, a fourth thinking chain with the smallest chain length is selected from all the first thinking chains, and the fourth thinking chain is used as the second thinking chain; If the processing complexity is not low complexity, the processing complexity is determined to be medium complexity, and the usage frequency value of the fifth thinking chain is obtained; wherein the fifth thinking chain is all other thinking chains in the first thinking chain except the third thinking chain and the fourth thinking chain; A sixth chain of thinking with the highest frequency of use is selected from the fifth chain of thinking, and the sixth chain of thinking is used as the second chain of thinking.

4. The question processing method based on artificial intelligence according to claim 1 is characterized in that: The step of determining the processing complexity corresponding to the second text specifically includes: Performing sentence processing on the second text to obtain corresponding sentences, and counting a first number of the sentences; Performing word segmentation processing on the second text to obtain corresponding words, and counting a second number of the words; Based on the first number and the second number, a complexity score of the second text is calculated based on a preset calculation strategy; A query process is performed on a preset mapping data table based on the complex score, so as to query the processing complexity corresponding to the complex score from the mapping data table.

5. The question processing method based on artificial intelligence according to claim 4 is characterized in that: The step of calculating the complexity score of the second text based on the first number and the second number and based on a preset calculation strategy specifically includes: Determine a preset weighting algorithm; Determine a first weight corresponding to the number of sentences based on the weight obtaining algorithm; Determine a second weight corresponding to the number of words based on the weight obtaining algorithm; Based on the first weight and the second weight, calling a preset score calculation formula to calculate the first quantity and the second quantity to obtain a corresponding calculation result; The calculation result is used as the complexity score of the second text.

6. The question processing method based on artificial intelligence according to claim 1, characterized in that: Before the step of performing intent recognition on the second text based on the preset large language model to obtain a target intent corresponding to the second text, the method further includes: Determine whether a thinking chain library construction request triggered by a specified user is received; If yes, display the preset build page; Receiving thought chain data input by the designated user on the construction page; Building a corresponding thinking chain library based on the thinking chain data; The thought chain library is stored and processed.

7. The question processing method based on artificial intelligence according to claim 1 is characterized in that: After the step of performing feedback processing on the user based on the answer data, the method further includes: Obtain pre-collected performance test data; Performing a performance evaluation on the large language model based on the performance test data to obtain corresponding model performance data; Acquire basic model performance data corresponding to the large language model; Comparing and analyzing the model performance data with the basic model performance data to obtain corresponding comparison and analysis results; The large language model is adjusted based on the comparison and analysis results.

8. A question processing device based on artificial intelligence, characterized in that: include: A conversion module, used for performing text conversion processing on the received question voice input by the user to obtain a first text; A preprocessing module, used for preprocessing the first text to obtain a second text; A recognition module, configured to perform intent recognition on the second text based on a preset large language model to obtain a target intent corresponding to the second text; wherein the large language model is loaded with a thought chain library generated based on a preset construction method, and the thought chain library includes a certain number of thought chains; A determination module, used to determine a processing complexity corresponding to the second text; A screening module, used for screening out a target thinking chain from all thinking chains contained in the thinking chain library based on the target intention and the processing complexity; A first processing module is used to perform execution and reasoning processing according to the target thought chain through the large language model, and output answer data corresponding to the target intention; The second processing module is used to perform feedback processing on the user based on the answer data.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the question processing method based on artificial intelligence as claimed in any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the question processing method based on artificial intelligence are implemented as described in any one of claims 1 to 7.

Images