Data processing method, device and equipment

Abstract

The embodiment of the invention discloses a data processing method, device and equipment. The method comprises the steps of receiving problem data to be processed; based on the problem data, generating prompt information corresponding to the problem data by using a preset instruction template, the preset instruction template comprising information for generating a plurality of different reasoning thinking chains; inputting the prompt information into a large language model, and generating a plurality of different reasoning thinking chains corresponding to the problem data; and performing validity evaluation processing on the plurality of different inference thinking chains corresponding to the question data through a large language model to obtain corresponding evaluation results, and determining answer data corresponding to the question data based on the evaluation results.

Description

Technical Field

[0001] This document relates to the field of computer technology, and in particular to a data processing method, device and equipment. Background Art

[0002] With advances in deep learning technology, large language models have become powerful tools for processing and understanding natural language, demonstrating impressive performance in a variety of applications, including text generation, privacy detection and protection, translation, and question-answering systems. However, these models still face challenges in handling complex reasoning tasks, particularly in problem-solving scenarios that require intermediate reasoning steps. Therefore, a better data processing solution is needed to enhance problem-solving capabilities while also providing greater transparency into the decision-making process of large language models, thereby enhancing user understanding and trust in their output. Summary of the Invention

[0003] The purpose of the embodiments of this specification is to provide a better data processing solution to improve problem-solving capabilities, while providing more transparency for the decision-making process of the large language model and enhancing users' understanding and trust in the output of the large language model.

[0004] In order to implement the above technical solution, the embodiments of this specification are implemented as follows: An embodiment of this specification provides a data processing method, which includes: receiving question data to be processed; based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning thinking chains; inputting the prompt information into a large language model to generate multiple different reasoning thinking chains corresponding to the question data; performing validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data through the large language model to obtain corresponding evaluation results, and determining answer data corresponding to the question data based on the evaluation results.

[0005] An embodiment of the present specification provides a data processing device, which includes: a question receiving module, which receives question data to be processed; an instruction generating module, which uses a preset instruction template to generate prompt information corresponding to the question data based on the question data, and the preset instruction template includes information for generating multiple different reasoning thinking chains; a multiple thinking chain generating module, which inputs the prompt information into a large language model to generate multiple different reasoning thinking chains corresponding to the question data; an answer determining module, which performs validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data through the large language model to obtain corresponding evaluation results, and determines answer data corresponding to the question data based on the evaluation results.

[0006] An embodiment of the present specification provides a data processing device, which includes: a processor; and a memory arranged to store computer-executable instructions, wherein the executable instructions, when executed, cause the processor to: receive question data to be processed; based on the question data, use a preset instruction template to generate prompt information corresponding to the question data, wherein the preset instruction template includes information for generating multiple different reasoning chains; input the prompt information into a large language model to generate multiple different reasoning chains corresponding to the question data; perform validity evaluation processing on the multiple different reasoning chains corresponding to the question data through the large language model to obtain corresponding evaluation results, and determine answer data corresponding to the question data based on the evaluation results.

[0007] An embodiment of this specification also provides a storage medium, which is used to store computer-executable instructions, and the executable instructions implement the following process when executed by a processor: receiving question data to be processed; based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning thinking chains; inputting the prompt information into a large language model to generate multiple different reasoning thinking chains corresponding to the question data; performing validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data through the large language model to obtain corresponding evaluation results, and determining answer data corresponding to the question data based on the evaluation results.

[0008] An embodiment of this specification also provides a computer program product, including a computer program, which implements the following process when executed by a processor: receiving question data to be processed; based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning thinking chains; inputting the prompt information into a large language model to generate multiple different reasoning thinking chains corresponding to the question data; performing validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data through the large language model to obtain corresponding evaluation results, and determining answer data corresponding to the question data based on the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some of the embodiments described in this specification. Those skilled in the art can also derive other drawings based on these drawings without inventive work. Figure 1This is a schematic diagram of the structure of a data processing system in this specification; Figure 2 This is a schematic diagram of a data processing process in this specification; Figure 3 This is a diagram of a question-and-answer page for this manual; Figure 4 This is a schematic diagram of an instruction template for this manual; Figure 5 This is a schematic diagram of a data processing process including an effectiveness evaluation process in this specification; Figure 6 This is a schematic diagram of a data processing process based on a large language model in this specification; Figure 7 A schematic diagram of a data processing process including a confidence processing process in this specification; Figure 8 This is a schematic diagram of a data processing process including correction of the reasoning thought chain in this specification; Figure 9 This is a schematic diagram of a data processing process including reinforcement learning processing in this specification; Figure 10 This is a schematic diagram of a data processing device in this specification; Figure 11 This is a schematic diagram of a data processing device in this specification. DETAILED DESCRIPTION

[0010] The embodiments of this specification provide a data processing method, apparatus, and device.

[0011] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0012] The embodiments of this specification provide a mechanism for improving the reasoning capabilities of large language models. With the advancement of deep learning technology, large language models have become powerful tools for processing and understanding natural language, and have demonstrated good performance in various applications such as text generation, translation, and question-answering systems. However, these models still face challenges when handling complex reasoning tasks, especially in problem-solving scenarios that require intermediate reasoning steps.

[0013] Typically, the Chain of Thought (CoT) approach improves the performance of large language models on complex tasks by requiring them to generate intermediate reasoning steps. This approach simulates the human thought process and improves the performance of large language models on specific tasks by explicitly generating problem-solving steps. However, this approach has limitations: it typically only generates a single chain of reasoning, lacking diversity, which limits the model's performance when faced with multiple-solution problems or those requiring multi-angle considerations. Fine-tuning large language models can also improve reasoning capabilities for complex tasks, but the effectiveness of fine-tuning depends on the availability of annotated data. For tasks with scarce data, performance gains may be limited. Frequently updating large language models with the emergence of new data and tasks can be challenging. To overcome this, this embodiment proposes a Divergent Chain of Thought (DCoT) approach to improve the reasoning capabilities of large models. The core idea of ​​the DCoT approach is to generate multiple different chains of reasoning within a single reasoning step (or a single reasoning task, i.e., a reasoning task corresponding to the same problem data). The final answer data is then selected by evaluating the chains of reasoning. This approach draws on the theories of divergent and convergent thinking in psychology. By generating multiple possible solutions (divergent thinking) and then selecting a superior solution (convergent thinking), it improves problem-solving skills. Furthermore, by generating multiple chains of reasoning, it provides greater transparency to the large language model's decision-making process, enhancing users' understanding and trust in the model's output. For detailed processing, please refer to the following examples.

[0014] The data processing method provided in one or more embodiments of this specification can be applied to the implementation environment of data processing. Figure 1 , the implementation environment includes at least: The client 100 and the server 200 may include one or more different large language models, among which: The client 100 can run on a terminal device, which can be a mobile phone, a personal computer, a tablet computer, an e-book reader, a wearable device, a device that interacts with information based on AR (Augmented Reality) or VR (Virtual Reality), a laptop computer, etc. The terminal device can install the client 100, which can be an application, a browser, or a subroutine installed in an application, etc.

[0015] The server 200 can run on a server, which can be one or more servers, a server cluster consisting of several servers, or a cloud server on a cloud computing platform. The server can be installed with the server 200, which can be an application or a subroutine within an application. One or more different large language models can be integrated into the server 200, or the server 200 can call one or more different large language models to perform corresponding operations.

[0016] In this implementation environment, the server 200 can receive question data to be processed, and can use a preset instruction template to generate prompt information corresponding to the question data based on the question data. The preset instruction template includes information for generating multiple different reasoning thinking chains. Afterwards, the prompt information can be input into the large language model to generate multiple different reasoning thinking chains corresponding to the question data. The large language model then performs validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data to obtain corresponding evaluation results, and determines the answer data corresponding to the question data based on the evaluation results, thereby generating multiple different reasoning thinking chains in a single reasoning step (or a single reasoning task), and selecting the final answer data by evaluating the reasoning thinking chains, thereby improving problem-solving capabilities. At the same time, by generating multiple reasoning thinking chains, more transparency is provided for the decision-making process of the large language model, enhancing the user's understanding and trust in the output of the large language model.

[0017] like Figure 2 As shown, an embodiment of this specification provides a method for processing data, and the execution subject of the method can be a terminal device or a server, etc., wherein the terminal device can be a mobile terminal device such as a mobile phone, a tablet computer, or a computer device such as a laptop or a desktop computer, or an IoT device (specifically such as a smart watch, a car-mounted device, etc.), etc., wherein the server can be an independent server, or a server cluster composed of multiple servers, etc. The server can be a background server in fields such as medical diagnosis, legal consulting, financial analysis and scientific research (especially in the field of problem solving that requires multi-step reasoning), or a background server of an application, etc. In this embodiment, the execution subject is taken as an example to explain in detail. For the case where the execution subject is a terminal device, please refer to the following server case processing, which will not be repeated here. The method can specifically include the following steps: In step S202, question data to be processed is received.

[0018] Among them, the question data can be data of any question, for example, the question data can be "Who will be the president of Country A in 2025?", or the question data can be "What film and television works has the director of K Theater starred in?", etc., which can be set according to actual conditions, and the embodiments of this manual do not limit this.

[0019] In implementation, when users or technicians need to know the answer data corresponding to a question data, they can obtain the question and answer page through the terminal device, such as Figure 3 As shown, the question-and-answer page may include an input box for question data, an output box for answer data, an OK button, and a Cancel button. A user or technician may enter the question data to be queried in the question data input box on the question-and-answer page. After the input is completed, the user or technician may click the OK button. At this time, the terminal device of the user or technician may obtain the question data in the question data input box and may send the question data to the server. The server may receive the question data entered by the user or technician and may use the question data as question data to be processed.

[0020] It should be noted that the question data input by the user in the embodiments of this specification may be data for simpler questions. For example, the question data may be the answer data corresponding to the question data that can be obtained after a single jump. The question data may be data for more complex questions. For complex questions, it is difficult for a large language model to directly give the correct answer, so step-by-step reasoning is required to obtain the correct answer. Based on this, the question data for more complex questions may be data that can only obtain the answer data corresponding to the question data after multiple jumps (such as 2 jumps, 3 jumps, etc.) (for example, the question data is "What film and television works has the director of K Grand Theater starred in?", then to answer the question corresponding to the question data, it is necessary to first determine who is the director of K Grand Theater? Then, determine what film and television works he has starred in? In this way, answering the above question requires 2 jumps to obtain the answer data corresponding to the question data).

[0021] In step S204, based on the above question data, a preset instruction template is used to generate prompt information corresponding to the question data, and the preset instruction template includes information for generating multiple different reasoning thought chains.

[0022] Among them, the instruction template (i.e., Prompt Template) can be used to guide the large language model to generate multiple different reasoning chains. The instruction template can require the large language model to generate multiple different reasoning chains for a single reasoning step (i.e., the same question data or corresponding prompt information) and finally give the final answer data. The form of the instruction template can be varied, for example Figure 4As shown, the instruction template includes a prompt and a response. The prompt takes the form: [Question] XXXX [Options] XXXX [Number of Answers] k, and the response takes the form: [Answer 1 (answer data 1)] CoT1 [Answer 2 (answer data 2)] ... [Answer k (answer data k)] CoTk [FinalAnswer (final answer data)] XXX, where k represents the number of chains to be generated. Prompts can be used in large language models. Prompts are text input to the large language model to guide it in generating output that meets the requirements. Prompts can be questions, instructions, or context, helping the large language model better understand and respond to various queries. Prompts can include explicit instruction prompts, context-supplementing prompts, and example-guided prompts. Prompts provide detailed background information and examples (such as those in the instruction template above) to help the large language model generate more relevant and accurate output. Chain-of-Thought (CoT) is a novel language model reasoning approach that requires a large language model to generate a step-by-step reasoning process before providing a final answer. This approach mimics human thinking by breaking down complex problems into a series of simpler steps, thereby improving the large language model's understanding and reasoning capabilities. It should be noted that the Chain-of-Thought reasoning approach in this embodiment can be a Divergent Chain of Thought (DCoT), a design further developed based on the aforementioned Chain-of-Thought approach. The core concept of the DCoT approach is to generate multiple different chains of reasoning within a single reasoning step (or single reasoning task), and then select the final answer data by evaluating the chains. This approach draws on the theories of divergent and convergent thinking in psychology. By generating multiple possible solutions (divergent thinking) and then selecting a superior solution (convergent thinking), it not only improves problem-solving capabilities but also enhances the reasoning capabilities of the large model.

[0023] In implementation, corresponding instruction templates can be pre-set according to actual conditions, different instruction templates can be set according to different industry fields or scenarios, different instruction templates can be set according to different needs, and different instruction templates can be set according to different types of users, etc. The specific settings can be based on actual conditions, and this specification embodiment does not limit this. Then, based on the received question data, relevant information such as the current industry field, scenario, need, and user type can be determined, and the corresponding instruction template (such as Figure 4 ), the command template is combined with the received question data to generate prompt information corresponding to the question data. For example, the prompt information generated for the question data might be: "What film and television works has the director of K Grand Theater appeared in?" This is a question that requires k different reasoning processes (i.e., reasoning chains). The output format is as follows: [Answer Data 1] Reasoning Chain CoT1; [Answer Data 2] Reasoning Chain CoT2; [Answer Data 3] Reasoning Chain CoT3...; [Answer Data k] Reasoning Chain CoTk.

[0024] In step S206, the prompt information is input into the large language model to generate multiple different reasoning chains corresponding to the above question data.

[0025] Among them, large language models (LLM) refer to a type of complex artificial intelligence models trained with large-scale data sets. They usually adopt deep learning architecture, especially the Transformer architecture widely used in the field of natural language processing (NLP). Such models have extremely high parameter counts, usually reaching tens of billions or even more, which gives them powerful language understanding and generation capabilities. By learning language rules and patterns from massive data, large language models can perform well in various natural language processing tasks, including but not limited to question answering, text generation, translation, summarization, dialogue interaction, sentiment analysis, etc., and can show high flexibility and creativity in these tasks. Due to their strong generalization ability and high adaptability to language structure, large language models have become an important pillar of modern natural language processing technology, including the GPT series of large language models (specifically GPT-3, GPT-4), Tongyi Qianwen, etc.

[0026] In implementation, a large language model architecture can be constructed based on actual conditions (for example, a large language model architecture can be constructed based on a Transformer module, etc.). Subsequently, a large number of training samples can be obtained and used to train the large language model, resulting in a trained large language model. In another embodiment, a pre-trained large language model can be directly obtained, and then a certain amount of sample data can be obtained, which can be used to fine-tune the pre-trained large language model, ultimately resulting in a fine-tuned large language model.

[0027] After obtaining the prompt information through the above method, the prompt information can be input into the above-mentioned large language model, and the problem data can be analyzed through the large language model. Therefore, for complex problems, a complex problem data can be gradually decomposed into step-by-step sub-problem data, thereby obtaining a reasoning process determined to determine the answer data of the problem data. During the decomposition process, a complex problem data can be decomposed into sub-problem data of finer-grained steps, or a complex problem data can be decomposed into sub-problem data of coarser-grained steps. The reasoning process (i.e., the reasoning thinking chain corresponding to the problem data) can also be directly given. Through the above method (i.e., the divergent thinking chain DCoT method), multiple different reasoning thinking chains can be generated for the same problem data. For simple problems, the reasoning process can be given directly, or its reasoning process can be determined from historical data, or the reasoning process can be given by example, etc., thereby obtaining multiple different reasoning thinking chains corresponding to the problem data. Each reasoning thinking chain is an independent thinking path for the problem data, attempting to solve the problem data from multiple different angles.

[0028] In step S208, the validity evaluation process of the multiple different reasoning chains corresponding to the above question data is performed through the large language model to obtain corresponding evaluation results, and the answer data corresponding to the question data is determined based on the evaluation results.

[0029] The large language model can be the same as the large language model used to generate the inference chain, or it can be a different large language model from the large language model used to generate the inference chain. The specific setting can be based on actual circumstances. The validity assessment process can be a process for evaluating whether the inference chain is correct and reasonable.

[0030] In implementation, the above-mentioned large language model or another pre-trained large language model can be used to perform validity evaluation on each of the multiple different reasoning chains corresponding to the above-mentioned question data. Specifically, it can be evaluated whether the reasoning logic of a certain reasoning chain is contradictory or reasonable, whether a certain reasoning chain is consistent with the facts, etc., thereby achieving validity evaluation of each of the multiple different reasoning chains corresponding to the question data and obtaining corresponding evaluation results. If each reasoning chain is wrong or unreasonable, the reasoning chain can be deleted, and only the reasoning chain with a correct and reasonable evaluation result can be retained. The most likely correct reasoning chain can be selected from the correct and reasonable reasoning chains, and the answer data corresponding to the reasoning chain can be used as the answer data corresponding to the question data. Among them, the validity evaluation of the multiple different reasoning chains corresponding to the above-mentioned question data by the large language model can be specifically to construct corresponding prompt information Prompt based on the multiple different reasoning chains corresponding to the above-mentioned question data, and then the prompt information Prompt can be input into the large language model to obtain corresponding evaluation results.

[0031] The embodiment of this specification provides a data processing method, which receives question data to be processed and can generate prompt information corresponding to the question data based on the question data using a preset instruction template. The preset instruction template includes information for generating multiple different reasoning thinking chains. After that, the prompt information can be input into a large language model to generate multiple different reasoning thinking chains corresponding to the question data. The large language model is then used to perform validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data to obtain corresponding evaluation results, and the answer data corresponding to the question data is determined based on the evaluation results. In this way, based on the divergent thinking and convergent thinking theories in psychology, The divergent thinking chain method improves problem-solving ability by generating multiple possible solutions (divergent thinking) and then selecting a better solution (convergent thinking). Moreover, it can generate multiple different independent reasoning thinking chains in a single reasoning step (or a single reasoning task), and then select the final answer data by evaluating the reasoning thinking chains, thereby simulating the human way of thinking of considering multiple possibilities when solving problems, thereby increasing the diversity and comprehensiveness of problem solving, improving problem-solving ability and the accuracy of answers. At the same time, by generating multiple reasoning thinking chains, it provides more transparency for the decision-making process of the large language model, enhancing users' understanding and trust in the output of the large language model.

[0032] In practical applications, the multiple different reasoning chains corresponding to the problem data include a sequence of reasoning chains consisting of multiple reasoning steps. Based on this, in the above step S208, the effectiveness evaluation of the multiple different reasoning chains corresponding to the above problem data is performed through a large language model, and the specific processing methods for obtaining the corresponding evaluation results can be varied. The following provides an optional processing method, which can specifically include the following steps S20802 and S20804. Based on this, in the above step S208 Figure 2 Based on this, the method specifically includes the following steps: Figure 5 shown.

[0033] In step S20802, one reasoning step and question data are selected from the reasoning thinking chain of a sequence composed of multiple reasoning steps to constitute the first object to be evaluated, and / or, multiple reasoning steps and question data are selected from the reasoning thinking chain of a sequence composed of multiple reasoning steps to constitute the second object to be evaluated.

[0034] Among them, the reasoning thinking chain of a sequence composed of multiple reasoning steps can be a reasoning thinking chain of a sequence composed of step 1-step 2-step 3-…-step n. Specifically, the problem data can be decomposed into multiple different reasoning steps, namely step 1-step 2-step 3-…-step n, and the sequence of decomposed reasoning steps is used as a reasoning thinking chain for the problem data.

[0035] In implementation, Figure 6 As shown, through the large language model, multiple different reasoning chains can be generated for the question data. For a reasoning chain consisting of a sequence of multiple reasoning steps, any one of the reasoning steps can be selected to form the first object to be evaluated with the question data. For example, step 2 can be selected to form the first object to be evaluated with the question data, i.e., (q1, CoT12), where q1 represents the question data and CoT12 represents step 2 in the above reasoning chain. Alternatively, step 3 can be selected to form the first object to be evaluated with the question data, i.e., (q1, CoT13). In addition, for a reasoning chain that does not contain multiple reasoning steps, the question data and the reasoning chain can be directly combined to form the object to be evaluated. In addition, for the reasoning thinking chain of the sequence composed of multiple reasoning steps, multiple reasoning steps and problem data can be selected from the reasoning thinking chain of the sequence composed of multiple reasoning steps to constitute the second object to be evaluated. For example, step 1 and step 2 can be selected from the reasoning thinking chain of the sequence composed of the above multiple reasoning steps and the problem data to constitute the second object to be evaluated, that is, (q1, CoT11, CoT12), or step 1, step 2 and step 3 can be selected from the reasoning thinking chain of the sequence composed of the above multiple reasoning steps and the problem data to constitute the second object to be evaluated, that is, (q1, CoT11, CoT12, CoT13), etc.

[0036] In step S20804, the first object and / or the second object is evaluated for effectiveness using the large language model to obtain corresponding evaluation results.

[0037] In implementation, Figure 6 As shown, a corresponding prompt message Prompt can be constructed based on the first object, and then the prompt message Prompt can be input into the large language model. The first object is evaluated for effectiveness by the large language model to obtain a corresponding evaluation result. The specific implementation method of the effectiveness evaluation process can be found in the aforementioned related content and will not be repeated here.

[0038] In actual applications, in the above step S208, the effectiveness of the multiple different reasoning thinking chains corresponding to the above problem data are evaluated through the large language model, and the specific processing methods for obtaining the corresponding evaluation results can be diverse. An optional processing method is provided below, which may specifically include the following: the effectiveness of the reasoning logic of each of the multiple different reasoning thinking chains corresponding to the problem data is evaluated through the large language model to determine whether each reasoning thinking chain is valid in terms of reasoning logic.

[0039] In implementation, first, the logical consistency of the reasoning thinking chain can be checked through the large language model, and a corresponding prompt information Prompt can be constructed for each reasoning thinking chain in multiple different reasoning thinking chains corresponding to the problem data, and the prompt information Prompt can be input into the large language model to guide the large language model to evaluate the logical consistency of each reasoning thinking chain through the prompt information Prompt, so as to check whether each reasoning thinking chain has contradictions or unreasonable inferences in the reasoning logic, thereby realizing the validity evaluation processing of the reasoning logic of each reasoning thinking chain, and determining whether each reasoning thinking chain is valid in the reasoning logic, that is, if it is determined through the validity evaluation processing of the reasoning logic that a certain reasoning thinking chain does not have contradictions or unreasonable inferences in the reasoning logic, then the reasoning thinking chain is determined to be valid in the reasoning logic; otherwise, it can be determined that the reasoning thinking chain has contradictions or unreasonable inferences in the reasoning logic and is invalid.

[0040] In actual applications, the multiple different reasoning thinking chains corresponding to the problem data include reasoning thinking chains for objective facts. Based on this, in the above step S208, the effectiveness evaluation of the multiple different reasoning thinking chains corresponding to the above problem data is performed through the large language model, and the specific processing methods for obtaining the corresponding evaluation results can be various. An optional processing method is provided below, which may specifically include the following: through the large language model, using pre-set benchmark data, the factual correctness evaluation of the reasoning thinking chains for objective facts in the multiple different reasoning thinking chains corresponding to the problem data is performed to determine whether the reasoning thinking chains for objective facts are correct.

[0041] In practice, since objective facts are real and not subject to human will, benchmark data can be pre-set. This benchmark data can be implemented in a variety of ways. For example, a designated knowledge base or reliable data source can be pre-set, and the aforementioned benchmark data can be provided by the designated knowledge base or reliable data source. Alternatively, designated benchmark data can be set according to actual needs based on expert experience, etc., and the specific setting can be based on actual conditions. In order to verify the content involving objective facts in the reasoning chain, the designated knowledge base or reliable data source can be called through the large language model, and the factual correctness of the reasoning chain for objective facts in multiple different reasoning chains corresponding to the problem data can be evaluated through the benchmark data in the designated knowledge base or reliable data source to determine whether the reasoning chain for objective facts is correct.

[0042] In actual applications, in the above step S208, the effectiveness of multiple different reasoning thinking chains corresponding to the above problem data are evaluated through a large language model, and the specific processing methods for obtaining the corresponding evaluation results can be diverse. An optional processing method is provided below, which may specifically include the following: compare the multiple different reasoning thinking chains corresponding to the problem data through a large language model to determine whether there are erroneous reasoning thinking chains among the multiple different reasoning thinking chains corresponding to the problem data.

[0043] During implementation, it is also possible to identify errors or inconsistencies in multiple different reasoning chains by comparing different reasoning chains, which may include logical errors, factual errors, or answers irrelevant to the problem data. Specifically, a corresponding prompt message Prompt can be constructed based on two or more reasoning chains among the multiple different reasoning chains, and the prompt message Prompt can be input into the large language model. The prompt message Prompt can be used to guide the large language model to compare two or more reasoning chains among the multiple different reasoning chains. Through differential analysis between each other, the erroneous reasoning chains such as obvious inconsistencies or logical contradictions between the multiple different reasoning chains corresponding to the problem data can be determined.

[0044] In practical applications, in the above step S208, the effectiveness evaluation of multiple different reasoning chains corresponding to the above problem data is performed through a large language model, and the specific processing methods for obtaining the corresponding evaluation results can be varied. The following provides an optional processing method, which can specifically include the following steps S20806 and S20808. Based on this, in the above Figure 2 Based on this, the method specifically includes the following steps: Figure 7 shown.

[0045] In step S20806, a confidence score is performed on each of the multiple different reasoning chains corresponding to the question data using the large language model to obtain a confidence score value for each reasoning chain.

[0046] In step S20808, based on the confidence score of each reasoning chain, it is determined whether there is an untrustworthy reasoning chain among the multiple different reasoning chains corresponding to the question data.

[0047] In practical applications, the specific processing methods for determining the answer data corresponding to the question data based on the evaluation result in the above step S208 can be various. The following provides an optional processing method, which can specifically include the following steps S20810 to S20814. Based on this, in the above step S208 Figure 2 Based on this, the method specifically includes the following steps: Figure 8 shown.

[0048] In step S20810, if the evaluation result indicates that there are reasoning chains that do not meet the validity evaluation conditions among the multiple different reasoning chains corresponding to the problem data, the reasoning chains that do not meet the validity evaluation conditions are corrected through the large language model to obtain multiple corrected different reasoning chains.

[0049] Among them, the existence of a chain of reasoning thinking that does not meet the validity evaluation conditions may include determining that the chain of reasoning thinking is invalid in terms of reasoning logic (there are contradictions or unreasonableness in the reasoning logic, etc.) through the validity evaluation processing of the reasoning logic of the chain of reasoning thinking, determining that the chain of reasoning thinking based on objective facts is incorrect through the factual correctness evaluation of the chain of reasoning thinking based on objective facts, determining that there are errors including logical errors, factual errors or answers irrelevant to the problem data through comparison of multiple different chains of reasoning thinking, and determining that the chain of reasoning thinking is unreliable through confidence scoring processing. One or more of the following.

[0050] During implementation, if the evaluation result indicates that there is a reasoning thinking chain that does not meet the validity evaluation conditions among the multiple different reasoning thinking chains corresponding to the problem data, the large language model can be used to attempt to correct the reasoning thinking chain that does not meet the validity evaluation conditions. The correction of the reasoning thinking chain that does not meet the validity evaluation conditions can be achieved in a variety of different ways. For example, the correction of the reasoning thinking chain that does not meet the validity evaluation conditions can be achieved through backtracking modification or regeneration. Among them, the backtracking modification method can be a correction method of backtracking to the wrong starting step in the reasoning thinking chain and trying to generate a more reasonable subsequent step. The regeneration method can be a complete regeneration of the reasoning thinking chain that does not meet the validity evaluation conditions. This method is especially for situations where the reasoning thinking chain that does not meet the validity evaluation conditions is difficult to correct.

[0051] In step S20812, the validity of the corrected multiple different reasoning chains is evaluated through the large language model to obtain a re-evaluation result.

[0052] The effectiveness evaluation process in step S20812 can refer to the relevant content of the effectiveness evaluation process in the aforementioned content, which will not be repeated here.

[0053] In step S20814, if the re-evaluation result indicates that the multiple different reasoning thinking chains corresponding to the question data all meet the validity evaluation conditions, the answer data corresponding to the question data is determined based on the corrected multiple different reasoning thinking chains; otherwise, the correction processing of the reasoning thinking chains that do not meet the validity evaluation conditions through the large language model is repeatedly performed until the obtained evaluation result indicates that the multiple different reasoning thinking chains corresponding to the question data all meet the validity evaluation conditions.

[0054] In implementation, if the re-evaluation result indicates that the multiple different reasoning chains corresponding to the problem data all meet the validity evaluation conditions, the answer data corresponding to the problem data can be determined through the corrected multiple different reasoning chains and the reasoning chains that meet the validity evaluation conditions, wherein the specific processing method for determining the answer data corresponding to the problem data through the corrected multiple different reasoning chains and the reasoning chains that meet the validity evaluation conditions can be found in the aforementioned related content and will not be repeated here. If the re-evaluation result indicates that there are still reasoning chains that do not meet the validity evaluation conditions among the multiple different reasoning chains corresponding to the problem data, the correction processing of the reasoning chains that do not meet the validity evaluation conditions by the large language model in step S20810, as well as the processing of the above-mentioned steps S20812 and S20814, can be repeated until the obtained evaluation result indicates that the multiple different reasoning chains corresponding to the problem data all meet the validity evaluation conditions. At this time, if Figure 6 As shown, the answer data corresponding to the question data can be determined through multiple different corrected reasoning chains and reasoning chains that meet the validity evaluation conditions. The correction processing of the above reasoning chains can be regarded as a feedback loop. The large language model improves the accuracy of its reasoning through continuous comparison, evaluation and correction.

[0055] In practical applications, the specific processing methods for determining the answer data corresponding to the question data based on the evaluation result in the above step S208 can be various. The following provides an optional processing method, which can specifically include the following steps S20816 and S20818. Based on this, in the above step S208 Figure 2 Based on this, the method specifically includes the following steps: Figure 9 shown.

[0056] In step S20816, if the evaluation result indicates that the multiple different reasoning thinking chains corresponding to the problem data all meet the validity evaluation conditions, then each of the multiple different reasoning thinking chains corresponding to the problem data is used as a strategy in reinforcement learning, and each strategy is evaluated through a pre-trained reward model to determine the reasoning thinking chain among the multiple different reasoning thinking chains corresponding to the problem data that can generate accurate answer data.

[0057] In implementation, reinforcement learning can be introduced. After generating multiple reasoning thinking chains in the above manner, the reinforcement learning framework can be used to evaluate and select a better reasoning path, thereby determining the corresponding reasoning thinking chain and its answer data. Specifically, if the evaluation result indicates that the multiple different reasoning thinking chains corresponding to the problem data all meet the validity evaluation conditions, then each of the multiple different reasoning thinking chains corresponding to the problem data is used as a strategy in reinforcement learning, and each reasoning thinking chain is evaluated through a pre-trained reward model. The large language model determines the reasoning thinking chain that can generate accurate answer data among the multiple different reasoning thinking chains corresponding to the problem data through the evaluation results of the reward model.

[0058] In step S20818, based on the determined chain of reasoning, answer data corresponding to the question data is determined.

[0059] The embodiment of this specification provides a data processing method, which receives question data to be processed and can generate prompt information corresponding to the question data based on the question data using a preset instruction template. The preset instruction template includes information for generating multiple different reasoning thinking chains. After that, the prompt information can be input into a large language model to generate multiple different reasoning thinking chains corresponding to the question data. The large language model is then used to perform validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data to obtain corresponding evaluation results, and the answer data corresponding to the question data is determined based on the evaluation results. In this way, based on the divergent thinking and convergent thinking theories in psychology, The divergent thinking chain method improves problem-solving ability by generating multiple possible solutions (divergent thinking) and then selecting a better solution (convergent thinking). Moreover, it can generate multiple different independent reasoning thinking chains in a single reasoning step (or a single reasoning task), and then select the final answer data by evaluating the reasoning thinking chains, thereby simulating the human way of thinking of considering multiple possibilities when solving problems, thereby increasing the diversity and comprehensiveness of problem solving, improving problem-solving ability and the accuracy of answers. At the same time, by generating multiple reasoning thinking chains, it provides more transparency for the decision-making process of the large language model, enhancing users' understanding and trust in the output of the large language model.

[0060] Furthermore, by guiding the large language model to compare and evaluate multiple generated reasoning chains and select the optimal answer data, this internal comparison mechanism strengthens the large language model's decision-making process, enabling it to self-correct and thus improve the accuracy of the answer data. Furthermore, by generating multiple reasoning chains, the large language model possesses stronger self-correction capabilities, able to identify and correct incorrect answer data even without external feedback, thereby improving the accuracy, robustness, and generalization of the large language model.

[0061] The above is a data processing method provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a data processing device, such as Figure 10 shown.

[0062] The data processing device includes: a question receiving module 1001, an instruction generating module 1002, a multi-thinking chain generating module 1003 and an answer determining module 1004, wherein: The question receiving module 1001 receives question data to be processed; The instruction generation module 1002 generates prompt information corresponding to the question data using a preset instruction template based on the question data, wherein the preset instruction template includes information for generating multiple different reasoning chains; The multi-thinking chain generating module 1003 inputs the prompt information into the large language model to generate multiple different reasoning thinking chains corresponding to the question data; The answer determination module 1004 performs validity evaluation processing on multiple different reasoning chains corresponding to the question data through a large language model, obtains corresponding evaluation results, and determines the answer data corresponding to the question data based on the evaluation results.

[0063] In the embodiment of this specification, the multiple different reasoning chains corresponding to the question data include a sequence of reasoning chains consisting of multiple reasoning steps. The answer determination module 1004 includes: An evaluation object determination unit selects one reasoning step from the reasoning chain of the sequence consisting of the plurality of reasoning steps and the question data to constitute a first object to be evaluated, and / or selects multiple reasoning steps from the reasoning chain of the sequence consisting of the plurality of reasoning steps and the question data to constitute a second object to be evaluated; An evaluation unit performs validity evaluation processing on the first object and / or the second object through a large language model to obtain corresponding evaluation results.

[0064] In the embodiment of this specification, the answer determination module 1004 uses a large language model to evaluate the validity of the reasoning logic of each of the multiple different reasoning chains corresponding to the question data to determine whether each reasoning chain is valid in terms of reasoning logic.

[0065] In the embodiment of this specification, the multiple different reasoning thinking chains corresponding to the question data include reasoning thinking chains based on objective facts. The answer determination module 1004, through a large language model and using pre-set benchmark data, performs factual correctness evaluation on the reasoning thinking chains based on objective facts in the multiple different reasoning thinking chains corresponding to the question data to determine whether the reasoning thinking chains based on objective facts are correct.

[0066] In the embodiment of this specification, the answer determination module 1004 compares the multiple different reasoning chains corresponding to the question data through a large language model to determine whether there is an incorrect reasoning chain among the multiple different reasoning chains corresponding to the question data.

[0067] In the embodiment of this specification, the answer determination module 1004 includes: a confidence determination unit that performs confidence scoring processing on each of the multiple different reasoning chains corresponding to the question data using a large language model to obtain a confidence score value for each reasoning chain; Based on the confidence score value of each reasoning chain, it is determined whether there is an untrustworthy reasoning chain among the multiple different reasoning chains corresponding to the problem data.

[0068] In the embodiment of this specification, the answer determination module 1004 includes: a correction unit, which, if the evaluation result indicates that there is a reasoning chain that does not meet the validity evaluation condition among the multiple different reasoning chains corresponding to the problem data, corrects the reasoning chain that does not meet the validity evaluation condition by using the large language model to obtain multiple corrected different reasoning chains; a re-evaluation unit, which performs validity evaluation processing on the revised multiple different reasoning chains through a large language model to obtain a re-evaluation result; The first answer determination unit determines the answer data corresponding to the question data based on the corrected multiple different reasoning thinking chains if the re-evaluation result indicates that the multiple different reasoning thinking chains corresponding to the question data all meet the validity evaluation conditions; otherwise, repeatedly executes the correction processing of the reasoning thinking chains that do not meet the validity evaluation conditions through the large language model until the obtained evaluation result indicates that the multiple different reasoning thinking chains corresponding to the question data all meet the validity evaluation conditions.

[0069] In the embodiment of this specification, the answer determination module 1004 includes: an evaluation unit, if the evaluation result indicates that the multiple different reasoning chains corresponding to the question data all meet the validity evaluation condition, then using each of the multiple different reasoning chains corresponding to the question data as a strategy in reinforcement learning, evaluating each strategy through a pre-trained reward model, and determining a reasoning chain among the multiple different reasoning chains corresponding to the question data that can generate accurate answer data; The second answer determination unit determines answer data corresponding to the question data based on the determined reasoning chain.

[0070] The embodiment of this specification provides a data processing device, which receives question data to be processed and can generate prompt information corresponding to the question data based on the question data using a preset instruction template. The preset instruction template includes information for generating multiple different reasoning thinking chains. After that, the prompt information can be input into a large language model to generate multiple different reasoning thinking chains corresponding to the question data. The large language model is then used to perform validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data to obtain corresponding evaluation results, and the answer data corresponding to the question data is determined based on the evaluation results. In this way, based on the divergent thinking and convergent thinking theories in psychology, The divergent thinking chain method improves problem-solving ability by generating multiple possible solutions (divergent thinking) and then selecting a better solution (convergent thinking). Moreover, it can generate multiple different independent reasoning thinking chains in a single reasoning step (or a single reasoning task), and then select the final answer data by evaluating the reasoning thinking chains, thereby simulating the human way of thinking of considering multiple possibilities when solving problems, thereby increasing the diversity and comprehensiveness of problem solving, improving problem-solving ability and the accuracy of answers. At the same time, by generating multiple reasoning thinking chains, it provides more transparency for the decision-making process of the large language model, enhancing users' understanding and trust in the output of the large language model.

[0071] Furthermore, by guiding the large language model to compare and evaluate multiple generated reasoning chains and select the optimal answer data, this internal comparison mechanism strengthens the large language model's decision-making process, enabling it to self-correct and thus improve the accuracy of the answer data. Furthermore, by generating multiple reasoning chains, the large language model possesses stronger self-correction capabilities, able to identify and correct incorrect answer data even without external feedback, thereby improving the accuracy, robustness, and generalization of the large language model.

[0072] The above is a data processing device provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a data processing device, such as Figure 11 shown.

[0073] The data processing device may provide a terminal device or a server, etc. for the above embodiment.

[0074] Data processing devices can vary significantly depending on their configuration or performance. They may include one or more processors 1101 and memory 1102. Memory 1102 may store one or more applications or data. Memory 1102 may be either ephemeral or persistent. Applications stored in memory 1102 may include one or more modules (not shown), each of which may include a series of computer-executable instructions for the data processing device. Furthermore, processor 1101 may be configured to communicate with memory 1102, executing the series of computer-executable instructions in memory 1102 on the data processing device. The data processing device may also include one or more power supplies 1103, one or more wired or wireless network interfaces 1104, one or more input / output interfaces 1105, and one or more keyboards 1106.

[0075] Specifically, in this embodiment, the data processing device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer-executable instructions for the data processing device, and the one or more programs are configured to be executed by one or more processors, including computer-executable instructions for performing the following: Receive problem data to be processed; Based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning chains; Inputting the prompt information into a large language model to generate multiple different reasoning chains corresponding to the question data; The effectiveness of multiple different reasoning chains corresponding to the question data is evaluated through a large language model to obtain corresponding evaluation results, and answer data corresponding to the question data is determined based on the evaluation results.

[0076] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from the other embodiments. In particular, the data processing device embodiment is generally similar to the method embodiment, so its description is relatively simple. For relevant portions, refer to the description of the method embodiment.

[0077] The embodiment of this specification provides a data processing device, which receives question data to be processed and can generate prompt information corresponding to the question data based on the question data using a preset instruction template. The preset instruction template includes information for generating multiple different reasoning thinking chains. After that, the prompt information can be input into a large language model to generate multiple different reasoning thinking chains corresponding to the question data. The large language model is then used to perform validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data to obtain corresponding evaluation results, and the answer data corresponding to the question data is determined based on the evaluation results. In this way, based on the divergent thinking and convergent thinking theories in psychology, The divergent thinking chain method improves problem-solving ability by generating multiple possible solutions (divergent thinking) and then selecting a better solution (convergent thinking). Moreover, it can generate multiple different independent reasoning thinking chains in a single reasoning step (or a single reasoning task), and then select the final answer data by evaluating the reasoning thinking chains, thereby simulating the human way of thinking of considering multiple possibilities when solving problems, thereby increasing the diversity and comprehensiveness of problem solving, improving problem-solving ability and the accuracy of answers. At the same time, by generating multiple reasoning thinking chains, it provides more transparency for the decision-making process of the large language model, enhancing users' understanding and trust in the output of the large language model.

[0078] Furthermore, based on the above Figures 2 to 9 One or more embodiments of this specification further provide a storage medium for storing computer-executable instruction information. In a specific embodiment, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer-executable instruction information stored in the storage medium is executed by a processor, the following process can be implemented: Receive problem data to be processed; Based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning chains; Inputting the prompt information into a large language model to generate multiple different reasoning chains corresponding to the question data; The effectiveness of multiple different reasoning chains corresponding to the question data is evaluated through a large language model to obtain corresponding evaluation results, and answer data corresponding to the question data is determined based on the evaluation results.

[0079] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from the other embodiments. In particular, the aforementioned storage medium embodiment is generally similar to the method embodiment, so its description is relatively simple. For relevant portions, refer to the description of the method embodiment.

[0080] The embodiment of this specification provides a storage medium, which receives question data to be processed and can generate prompt information corresponding to the question data based on the question data using a preset instruction template. The preset instruction template includes information for generating multiple different reasoning thinking chains. After that, the prompt information can be input into a large language model to generate multiple different reasoning thinking chains corresponding to the question data. The large language model is then used to perform validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data to obtain corresponding evaluation results, and the answer data corresponding to the question data is determined based on the evaluation results. In this way, based on the divergent thinking and convergent thinking theories in psychology, The divergent thinking chain approach improves problem-solving ability by generating multiple possible solutions (divergent thinking) and then selecting a better solution (convergent thinking). Moreover, it can generate multiple different independent reasoning chains in a single reasoning step (or a single reasoning task), and then select the final answer data by evaluating the reasoning chains, thereby simulating the human way of thinking of considering multiple possibilities when solving problems, thereby increasing the diversity and comprehensiveness of problem solving, improving problem-solving ability and the accuracy of answers. At the same time, by generating multiple reasoning chains, it provides more transparency for the decision-making process of the large language model, enhancing users' understanding and trust in the output of the large language model.

[0081] Furthermore, based on the above Figures 2 to 9 One or more embodiments of this specification further provide a computer program product, including a computer program. When the computer program in the computer program product is executed by a processor, it can implement the following process: Receive problem data to be processed; Based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning chains; Inputting the prompt information into a large language model to generate multiple different reasoning chains corresponding to the question data; The effectiveness of multiple different reasoning chains corresponding to the question data is evaluated through a large language model to obtain corresponding evaluation results, and answer data corresponding to the question data is determined based on the evaluation results.

[0082] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between the other embodiments. In particular, the aforementioned computer program product embodiment is generally similar to the method embodiment, so its description is relatively simple. For relevant portions, reference can be made to the description of the method embodiment.

[0083] The embodiment of this specification provides a computer program product, which receives question data to be processed and can generate prompt information corresponding to the question data based on the question data using a preset instruction template. The preset instruction template includes information for generating multiple different reasoning thinking chains. After that, the prompt information can be input into a large language model to generate multiple different reasoning thinking chains corresponding to the question data. The large language model is then used to perform validity evaluation processing on the multiple different reasoning thinking chains corresponding to the question data to obtain corresponding evaluation results, and the answer data corresponding to the question data is determined based on the evaluation results. In this way, based on the divergent thinking and convergent thinking theories in psychology, The divergent thinking chain method improves problem-solving ability by generating multiple possible solutions (divergent thinking) and then selecting a better solution (convergent thinking). Moreover, it can generate multiple different independent reasoning thinking chains in a single reasoning step (or a single reasoning task), and then select the final answer data by evaluating the reasoning thinking chains, thereby simulating the human way of thinking of considering multiple possibilities when solving problems, thereby increasing the diversity and comprehensiveness of problem solving, improving problem-solving ability and the accuracy of answers. At the same time, by generating multiple reasoning thinking chains, it provides more transparency for the decision-making process of the large language model, enhancing users' understanding and trust in the output of the large language model.

[0084] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0085] In the 1990s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using physical hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly performed using software called a "logic compiler." This is similar to the software compilers used during program development. Before compilation, the original code must be written in a specific programming language, called a Hardware Description Language (HDL). There are many types of HDL, including ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that simply by programming a method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0086] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the memory control logic. Those skilled in the art will also appreciate that, in addition to implementing the controller purely in computer-readable program code, the controller can also be implemented in the form of logic gates, switches, an application-specific integrated circuit, a programmable logic controller, an embedded microcontroller, etc. by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the means for implementing the various functions included therein can also be considered as structures within the hardware component. Alternatively, the means for implementing the various functions can be considered both a software module implementing the method and a structure within the hardware component.

[0087] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0088] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0089] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Thus, one or more embodiments of this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] The embodiments of this specification are described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable fraud case serial and parallel device to produce a machine, so that the instructions executed by the processor of the computer or other programmable fraud case serial and parallel device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0091] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable fraud case serial and parallel device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, the instruction device being implemented in the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0093] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0094] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0095] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0096] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0097] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Thus, one or more embodiments of this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0098] One or more embodiments of this specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of this specification may also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communications network. In distributed computing environments, program modules may be located in local and remote computer storage media, including storage devices.

[0099] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0100] The foregoing is merely an example of the present invention and is not intended to limit this document. Various modifications and variations are possible within the scope of this document. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this document are intended to be included within the scope of the claims of this document.

Claims

1. A data processing method, the method comprising: Receive problem data to be processed; Based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning chains; Inputting the prompt information into a large language model to generate multiple different reasoning chains corresponding to the question data; The effectiveness of multiple different reasoning chains corresponding to the question data is evaluated through a large language model to obtain corresponding evaluation results, and answer data corresponding to the question data is determined based on the evaluation results.

2. The method according to claim 1, wherein the multiple different reasoning chains corresponding to the question data include a sequence of reasoning chains consisting of multiple reasoning steps, and wherein the effectiveness evaluation of the multiple different reasoning chains corresponding to the question data using the large language model is performed to obtain corresponding evaluation results, including: Selecting one reasoning step from the chain of reasoning thoughts of the sequence formed by the plurality of reasoning steps and the question data to constitute a first object to be evaluated, and / or selecting multiple reasoning steps from the chain of reasoning thoughts of the sequence formed by the plurality of reasoning steps and the question data to constitute a second object to be evaluated; The first object and / or the second object are evaluated for effectiveness using a large language model to obtain corresponding evaluation results.

3. The method according to claim 1, wherein the large language model is used to perform validity evaluation on multiple different reasoning chains corresponding to the question data to obtain corresponding evaluation results, including: The validity evaluation process of the reasoning logic of each of the multiple different reasoning thinking chains corresponding to the problem data is performed through a large language model to determine whether each reasoning thinking chain is valid in terms of reasoning logic.

4. The method according to claim 1, wherein the multiple different reasoning chains corresponding to the question data include reasoning chains based on objective facts, and wherein the effectiveness evaluation of the multiple different reasoning chains corresponding to the question data using the large language model is performed to obtain corresponding evaluation results, including: Through the large language model and using pre-set benchmark data, the factual correctness of the reasoning thinking chains for objective facts in multiple different reasoning thinking chains corresponding to the problem data is evaluated to determine whether the reasoning thinking chains for objective facts are correct.

5. The method according to claim 1, wherein the large language model is used to perform validity evaluation on multiple different reasoning chains corresponding to the question data to obtain corresponding evaluation results, including: The large language model is used to compare multiple different reasoning chains corresponding to the question data to determine whether there is an incorrect reasoning chain among the multiple different reasoning chains corresponding to the question data.

6. The method according to claim 1, wherein the large language model is used to perform validity evaluation on multiple different reasoning chains corresponding to the question data to obtain corresponding evaluation results, including: Performing confidence scoring processing on each of the multiple different reasoning chains corresponding to the question data using a large language model to obtain a confidence score value for each reasoning chain; Based on the confidence score value of each reasoning chain, it is determined whether there is an untrustworthy reasoning chain among the multiple different reasoning chains corresponding to the problem data.

7. The method according to any one of claims 1 to 6, wherein determining the answer data corresponding to the question data based on the evaluation result comprises: If the evaluation result indicates that there is a reasoning thinking chain that does not meet the validity evaluation conditions among the multiple different reasoning thinking chains corresponding to the problem data, the reasoning thinking chain that does not meet the validity evaluation conditions is corrected by the large language model to obtain multiple corrected different reasoning thinking chains; Performing validity evaluation on the revised multiple different reasoning chains using a large language model to obtain a re-evaluation result; If the re-evaluation result indicates that the multiple different reasoning chains corresponding to the question data all meet the validity evaluation conditions, then the answer data corresponding to the question data is determined based on the corrected multiple different reasoning chains; otherwise, the correction process of the reasoning chains that do not meet the validity evaluation conditions is repeatedly performed through the large language model until the obtained evaluation result indicates that the multiple different reasoning chains corresponding to the question data all meet the validity evaluation conditions.

8. The method according to any one of claims 1 to 6, wherein determining the answer data corresponding to the question data based on the evaluation result comprises: If the evaluation result indicates that the multiple different reasoning chains corresponding to the question data all meet the validity evaluation conditions, then each of the multiple different reasoning chains corresponding to the question data is used as a strategy in reinforcement learning, and each strategy is evaluated using a pre-trained reward model to determine the reasoning chain that can generate accurate answer data among the multiple different reasoning chains corresponding to the question data; Based on the determined chain of reasoning, answer data corresponding to the question data is determined.

9. A data processing device, comprising: A problem receiving module receives problem data to be processed; An instruction generation module, based on the question data, uses a preset instruction template to generate prompt information corresponding to the question data, wherein the preset instruction template includes information for generating multiple different reasoning chains; A multi-thinking chain generation module inputs the prompt information into a large language model to generate multiple different reasoning thinking chains corresponding to the question data; The answer determination module uses a large language model to perform validity evaluation on multiple different reasoning chains corresponding to the question data, obtains corresponding evaluation results, and determines the answer data corresponding to the question data based on the evaluation results.

10. A data processing device, comprising: processor; as well as a memory arranged to store computer-executable instructions which, when executed, cause the processor to: Receive problem data to be processed; Based on the question data, using a preset instruction template to generate prompt information corresponding to the question data, the preset instruction template includes information for generating multiple different reasoning chains; Inputting the prompt information into a large language model to generate multiple different reasoning chains corresponding to the question data; The effectiveness of multiple different reasoning chains corresponding to the question data is evaluated through a large language model to obtain corresponding evaluation results, and answer data corresponding to the question data is determined based on the evaluation results.

Images