Data processing method, problem processing method, computing device, and storage medium

Abstract

Embodiments of the present disclosure provide a data processing method, a problem processing method, a computing device, and a storage medium. The data processing method comprises: determining initial data, and using a data processing model to perform data reasoning on the initial data to obtain an initial reasoning result; when it is determined on the basis of the initial reasoning result that the initial data satisfies a data reasoning condition, on the basis of the initial reasoning result, using the data processing model to perform data update on the initial data to obtain updated data, wherein the data volume of the updated data is less than that of the initial data, and the data processing model is a machine learning model; using the data processing model to perform data reasoning on the updated data to obtain an updated reasoning result; and when it is determined on the basis of the updated reasoning result that the updated data does not satisfy the data reasoning condition, determining the updated reasoning result as a target data processing result.

Description

Data processing method, problem processing method, computing device, and storage medium

[0001] The present disclosure claims priority to Chinese Patent Application No. 202410897972.9, filed on July 4, 2024, with the Chinese Patent Office, entitled "Data processing method, problem processing method, computing device, and storage medium", the entire contents of which are incorporated herein by reference. TECHNICAL FIELD

[0002] Embodiments of the present disclosure relate to the field of computer technology, and in particular, to a data processing method. One or more embodiments of the present disclosure also relate to a problem processing method, a computing device, a computer-readable storage medium, and a computer program product. BACKGROUND

[0003] With the continuous development of artificial intelligence technology, neural network models can be used to process various types of data to meet the needs of actual applications.

[0004] In the prior art, data to be processed can be input into a neural network model, and data reasoning is performed using the neural network model to obtain a corresponding data reasoning result. However, the neural network model has a low reasoning efficiency when processing complex and large amounts of data. Therefore, how to improve the data reasoning efficiency of the neural network model is a technical problem that needs to be solved. SUMMARY

[0005] In view of this, embodiments of the present disclosure provide a data processing method. One or more embodiments of the present disclosure also relate to a problem processing method, a data processing device, a problem processing device, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defect of low data reasoning efficiency of the neural network model in the prior art.

[0006] According to a first aspect of embodiments of the present disclosure, a data processing method is provided, comprising:

[0007] determining initial data and performing data reasoning on the initial data using a data processing model to obtain an initial reasoning result;

[0008] in a case where it is determined, according to the initial reasoning result, that the initial data satisfies a data reasoning condition, performing data updating on the initial data using the data processing model according to the initial reasoning result to obtain updated data, wherein a data quantity of the updated data is less than a data quantity of the initial data, and the data processing model is a machine learning model;

[0009] perform data inference on the update data by using the data processing model to obtain an update inference result;

[0010] In a case where it is determined, according to the update inference result, that the update data does not satisfy the data inference condition, the update inference result is determined as a target data processing result.

[0011] According to a second aspect of the embodiments of the present disclosure, a data processing apparatus is provided, including:

[0012] A first result determination module is configured to determine initial data, and perform data inference on the initial data by using a data processing model to obtain an initial inference result;

[0013] A data update module is configured to, in a case where it is determined, according to the initial inference result, that the initial data satisfies a data inference condition, perform data update on the initial data by using the data processing model according to the initial inference result to obtain update data, wherein a data quantity of the update data is less than a data quantity of the initial data, and the data processing model is a machine learning model;

[0014] A second result determination module is configured to perform data inference on the update data by using the data processing model to obtain an update inference result;

[0015] A third result determination module is configured to, in a case where it is determined, according to the update inference result, that the update data does not satisfy the data inference condition, determine the update inference result as a target data processing result.

[0016] According to a third aspect of the embodiments of the present disclosure, a problem processing method is provided, including:

[0017] Determining initial problem text, and performing problem inference on the initial problem text by using a data processing model to obtain an initial problem inference result;

[0018] In a case where it is determined, according to the initial problem inference result, that the initial problem text satisfies a problem inference condition, performing problem update on the initial problem text by using the data processing model according to the initial problem inference result to obtain update problem text, wherein a data quantity of the update problem text is less than a data quantity of the initial problem text, and the data processing model is a machine learning model;

[0019] Performing problem inference on the update problem text by using the data processing model to obtain an update problem inference result;

[0020] In a case where it is determined, according to the update question reasoning result, that the update question text does not satisfy the question reasoning condition, the update question reasoning result is determined as the target question processing result.

[0021] According to a fourth aspect of the embodiments of the present disclosure, a question processing apparatus is provided, comprising:

[0022] A first result determination module is configured to determine an initial question text, and perform question reasoning on the initial question text by using a data processing model to obtain an initial question reasoning result.

[0023] A question update module is configured to, in a case where it is determined, according to the initial question reasoning result, that the initial question text satisfies a question reasoning condition, perform question update on the initial question text by using the data processing model according to the initial question reasoning result to obtain an update question text, wherein a data quantity of the update question text is less than a data quantity of the initial question text, and the data processing model is a machine learning model.

[0024] A second result determination module is configured to perform question reasoning on the update question text by using the data processing model to obtain an update question reasoning result.

[0025] A third result determination module is configured to, in a case where it is determined, according to the update question reasoning result, that the update question text does not satisfy the question reasoning condition, determine the update question reasoning result as a target question processing result.

[0026] According to a fifth aspect of the embodiments of the present disclosure, a computing device is provided, comprising:

[0027] a memory and a processor;

[0028] The memory is configured to store computer programs / instructions, and the processor is configured to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the above data processing method or question processing method.

[0029] According to a sixth aspect of the embodiments of the present disclosure, a computer readable storage medium is provided, which stores computer programs / instructions, which, when executed by a processor, implement the steps of the above data processing method or question processing method.

[0030] According to a seventh aspect of the embodiments of the present disclosure, a computer program product is provided, comprising computer programs / instructions, which, when executed by a processor, implement the steps of the above data processing method or question processing method.

[0031] The data processing method provided in one or more embodiments of the present disclosure, in the process of data processing by using a data processing model, first, the initial data is subjected to first-stage data reasoning by using the data processing model, and an initial reasoning result is obtained; and in the case that it is determined according to the initial reasoning result that the initial data satisfies a data reasoning condition, the initial data is updated according to the initial reasoning result, so as to obtain updated data with a smaller data quantity than the initial data; then, the updated data with a smaller data quantity is subjected to second-stage data reasoning by using the data processing model, so as to quickly obtain an updated reasoning result. Since the data processing model can quickly reason the updated data with a smaller data quantity, although multiple stages of reasoning are needed, the overall data reasoning efficiency is higher than the data reasoning efficiency of the data processing model for data with a larger data quantity, thereby improving the data reasoning efficiency of the data processing model and avoiding the problem of low reasoning efficiency of the neural network model in processing data with a larger data quantity and being more complex. In the case that it is determined according to the updated reasoning result that the updated data does not satisfy the data reasoning condition, it is determined that the data reasoning is ended, and the updated reasoning result is determined as a target data processing result, thereby meeting the requirement of data processing of various types of data by using the neural network model in actual application. BRIEF DESCRIPTION OF DRAWINGS

[0032] FIG. 1 is an application schematic diagram of a data processing method according to an embodiment of the present disclosure;

[0033] FIG. 2 is a flowchart of a data processing method according to an embodiment of the present disclosure;

[0034] FIG. 3 is a process flowchart of a data processing method according to an embodiment of the present disclosure;

[0035] FIG. 4 is a flowchart of a problem processing method according to an embodiment of the present disclosure;

[0036] FIG. 5 is a structural schematic diagram of a data processing apparatus according to an embodiment of the present disclosure;

[0037] FIG. 6 is a structural schematic diagram of a problem processing apparatus according to an embodiment of the present disclosure;

[0038] FIG. 7 is a structural block diagram of a computing device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0039] In the following description, a lot of specific details are set forth in order to provide a thorough understanding of the present disclosure. However, the present disclosure can be practiced in many different ways beyond the specific embodiments described herein, and it is understood that it can be practiced with or without many of the details described below. Therefore, the present disclosure is not limited to the specific implementations disclosed herein but only by the claims and their equivalents.

[0040] The terminology used in this disclosure one or more embodiments is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure one or more embodiments. As used in this disclosure one or more embodiments and the accompanying claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this disclosure one or more embodiments, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0041] It will be understood that, although the terms first, second, etc. can be used herein to describe various information, these terms are not intended to denote a temporal sequence, but are used only to distinguish one type of information from another. For example, without departing from the scope of this disclosure one or more embodiments, first can be termed second, and, similarly, second can be termed first. Depending on the context, the word "if' as used herein can be interpreted to mean "when" or "in response to determining."

[0042] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this disclosure one or more embodiments are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0043] In this disclosure one or more embodiments, a large model refers to a deep learning model with a large number of model parameters, usually containing hundreds of millions, tens of billions, hundreds of billions, thousands of billions or even tens of billions of model parameters. The large model can also be called a foundation model. Through large-scale unlabeled corpus pre-training, a pre-trained model with hundreds of millions of parameters is produced. This model can adapt to a wide range of downstream tasks and has good generalization ability. For example, large language models (LLM) and multi-modal pre-training models.

[0044] In practical applications, a large model can be applied to different tasks by fine-tuning a pre-trained model with a small amount of samples. The large model can be widely applied in natural language processing (NLP) and computer vision, and can be applied to computer vision tasks such as visual question answering (VQA), image captioning (IC), image generation, and natural language processing tasks such as text-based sentiment classification, text summarization generation, and machine translation. The main application scenarios of the large model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, and the like.

[0045] First, the technical terms related to one or more embodiments of the present disclosure are explained.

[0046] RAG (Retrieval-augmented Generation): refers to retrieval-augmented generation, which is a retrieval-augmented generation model combining retrieval and generation techniques. The retrieval-augmented generation model combines a language model and information retrieval techniques. Specifically, when the model needs to generate text or answer a question, it will first retrieve relevant information from a large document set, and then use the retrieved information to guide the generation of text, thereby improving the quality and accuracy of the prediction.

[0047] Markov Chain (MC): is a stochastic process with Markov property in probability theory and mathematical statistics, which exists in discrete index set and state space. The main idea is that the state space undergoes a random process from one state to another. This process requires the property of "no memory": the probability distribution of the next state can only be determined by the current state, and the previous events in the time series are irrelevant.

[0048] CoT (Chain of Thought): refers to the large model thought chain, which is a high-level prompt engineering designed to enhance the performance of large language models in various complex reasoning tasks. By allowing the large model to gradually participate in the process of decomposing a complex problem into step-by-step sub-problems and solving them one by one, the performance of the large model can be significantly improved, and the intermediate steps of this series of reasoning are called the thought chain (Chain of Thought).

[0049] tools: refers to tools, which are various external tools that can be called by the large model.

[0050] Reduction: refers to the process of transforming a complex problem into one or more known, simpler problems. This transformation allows the original problem to be solved indirectly by solving these simple problems.

[0051] Adopting process: refers to a series of steps or strategies adopted when performing data reasoning, such as a series of steps or strategies adopted when solving mathematical problems, which includes selecting appropriate theorems, formulas, algorithms or methods to gradually approach the solution of the problem; for example, when solving a complex integral problem, a series of substitutions, partial integrals, etc. may be required; these can be considered as "adopting process".

[0052] Markov thought chain: a model for describing and inferring system behavior, i.e. building LLM based on Markov chain and large model thought chain.

[0053] Reasoning task: refers to the task of inferring unknown information from known information and reasoning rules.

[0054] ToT (Tree of Thoughts): a new type of language model reasoning framework. The thought tree provides a more structured prompting framework for LLM reasoning by breaking down complex problems into smaller, easier-to-solve problems.

[0055] PoT: refers to the thought program reasoning, which is based on CoT and converts the natural description language of thought chain into thought program, which can perform better in reasoning.

[0056] With the continuous development of artificial intelligence technology, neural network models can be used to process various types of data to meet the needs of practical applications; for example, with the rapid development of artificial intelligence and natural language processing technology, large language models have become a popular research field in recent years; however, large models still perform poorly in logical reasoning, i.e. large models have certain challenges and limitations in the field of reasoning (especially mathematical logic reasoning).

[0057] In practical applications, the reasoning method of large models is divided into single-step reasoning and multi-step reasoning, where single-step reasoning can solve simple problems and has relatively high reasoning efficiency; multi-step reasoning has stronger reasoning logic and higher reasoning accuracy. However, multi-step reasoning has a major flaw: as the reasoning process deepens, it is prone to hallucinations and low reasoning efficiency.

[0058] To address the above problems, the present disclosure provides a scheme for logical reasoning using CoT, PoT, ToT, etc. However, this scheme has low decoding efficiency, decoding error accumulation, and low utilization of training data.

[0059] Based on this, in the present disclosure, a data processing method is provided, and one or more embodiments of the present disclosure also relate to a problem processing method, a data processing device, a problem processing device, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail in the following embodiments.

[0060] Referring to FIG. 1, FIG. 1 shows an application schematic diagram of a data processing method according to an embodiment of the present disclosure. Based on FIG. 1, a user can send a question to a server 104 through a terminal 102, and the server 104 is a server to which the data processing method of one or more embodiments of the present disclosure is applied. After receiving the question, the server 104 inputs the question into an LLM large model, uses the large model to perform one-step reasoning on the current question, obtains a first-stage reasoning result after one-step reasoning, and then uses the question and the reasoning result to obtain a new question. The large model is used to perform one-step reasoning on the new question, obtain a second-stage reasoning result, and again obtain a question consistent with the answer to the new question but simpler based on the reasoning result and the new question. This process is repeated until the answer can be obtained through the question and answer until the answer corresponding to the user question is obtained. Then, the answer is sent to the terminal 102. In the above steps, the long sequence reasoning process in the original complex question is converted into multiple short sequence reasoning processes, which significantly reduces the reasoning cost of the large model and improves the reasoning efficiency of the large model.

[0061] Referring to FIG. 2, FIG. 2 shows a flowchart of a data processing method according to an embodiment of the present disclosure, which specifically includes the following steps.

[0062] Step 202: determining initial data and using a data processing model to perform data reasoning on the initial data to obtain an initial reasoning result.

[0063] The initial data can be understood as data that needs to be processed by a data processing model. The initial data can be any type of data, for example, text data, code data, language data, or video data, image data, etc. That is, the initial data can be initial question text, initial code data, initial voice data, initial image data, or initial video data, etc., which is not limited here.

[0064] The data processing model can be understood as a neural network model capable of data processing. The data processing model can be an LLM or a machine learning model. It should be noted that the data processing model is different when the data processing method is applied to different scenarios. For example, when the data processing method is applied to a question reasoning scenario, the initial data can be an initial question text. The initial reasoning result can be an initial question reasoning result. The data processing model can be a question processing model, which is used for question reasoning and question updating of the initial question text. Based on this, the question reasoning result (e.g., question answer) of the question processing model is obtained by inputting the initial question text into the question processing model, thereby meeting the demand for question reasoning using a neural network model in actual applications.

[0065] For another example, when the data processing method is applied to a code optimization scenario, the initial data can be initial code data. The initial reasoning result can be an initial code optimization result for the initial code data. The data processing model can be a code optimization model, which is used for code optimization and code updating of the initial code data. Based on this, the optimized code data (i.e., target data processing result) output by the code optimization model is obtained by inputting the initial code data into the code optimization model, thereby meeting the demand for code optimization using a neural network model in actual applications.

[0066] The initial reasoning result can be understood as a processing result obtained by performing first-stage data reasoning on the initial data using the data processing model. The first-stage data reasoning can be understood as data reasoning on initial sub-data in the initial data. For example, when the initial data is an initial question text, the initial sub-data can be a sub-question constituting the initial question text. Based on this, the initial reasoning result can be a question reasoning result obtained by performing question reasoning on the sub-question in the initial question text using the data processing model. For another example, when the initial data is initial code data, the initial sub-data can be one or more code statements in the initial code data. Based on this, the initial reasoning result can be a code optimization result obtained by performing code optimization on the one or more code statements in the initial code data using the code optimization model.

[0067] In one or more embodiments provided by the present disclosure, considering that the neural network model can have a low reasoning efficiency when processing relatively complex and large data, the data processing method provided by the present disclosure can perform multi-stage data reasoning on the initial data. Each stage can perform data reasoning on initial sub-data in the initial data until the data reasoning on the initial data is completed. The specific implementation manner is as follows.

[0068] The data processing model is used to perform data reasoning on the initial data to obtain an initial reasoning result, including:

[0069] The initial data is input into the data processing model, and the data processing model is used to perform data extraction on the initial data to determine a plurality of initial sub-data contained in the initial data.

[0070] The initial sub-data is determined from the plurality of initial sub-data, and semantic reasoning is performed on the initial sub-data to obtain the initial reasoning result.

[0071] Wherein, the initial sub-data can be understood as a sub-data constituting the initial data, in the case of the initial data being an initial question text, the initial sub-data can be a sub-question text in the initial question text, for example, the initial question text can be "1+2*3 equals how much", and the corresponding plurality of sub-question texts are "1+2 equals how much", "2*3 equals how much"; the initial question text can be "Who is the father of A's wife", and the corresponding plurality of sub-question texts are "Who is A's wife", "Who is the father of the wife"; in the case of the initial data being initial video data, the initial sub-data can be one or more video frames in the initial video data; in the case of the initial data being initial image data or initial audio data, the initial sub-data can be a local image region in the initial image data or a small piece of audio data in the initial audio data; in the case of the initial data being initial code data, the initial sub-data can be one or more code statements in the initial code data, for example, the initial code data can be "int main(){

[0072] Correspondingly, the initial sub-data can be one or more code statements in the initial code data, for example, the initial sub-data can be "int a=5;、int b=10;" two code statements.

[0073] The to-be-reasoned sub-data can be understood as any one of the initial sub-data that needs to be reasoned by using the data processing model, or the initial sub-data that needs to be reasoned in the first stage according to a data reasoning rule (such as a mathematical operation rule or a code running rule). For example, the first sub-question in the plurality of initial sub-questions, or the sub-question that needs to be reasoned in the first stage according to the mathematical operation rule, wherein, in the case that the initial question text is “1+2×3 is equal to how much”, the plurality of sub-question texts are “1+2 is equal to how much” and “2×3 is equal to how much”, the sub-question that needs to be reasoned in the first stage according to the mathematical operation rule can be understood as the sub-question “2×3 is equal to how much” determined by the data processing model according to the mathematical operation rule, which is the sub-question that needs to be reasoned in the first stage.

[0074] The initial reasoning result can be understood as a result obtained by reasoning the to-be-reasoned sub-data by using the data processing model. For example, in the case that the to-be-reasoned sub-data is the sub-question “2×3 is equal to how much”, the initial question reasoning result (i.e., the initial reasoning result) can be “2×3 is equal to 6”; in the case that the to-be-reasoned sub-data is the two code statements “int a=5;、int b=10;” that need to be optimized, the initial code optimization result (i.e., the initial reasoning result) can be “int a=5;、int b=10; can be written as a code statement int a=5,b=10;”.

[0075] The data processing method provided by the present disclosure is described and explained below in the application of the data processing method in the question reasoning scenario, wherein the initial data is an initial question text, the data processing model is a question processing model, the initial sub-data is an initial sub-question, the to-be-reasoned sub-data is a to-be-reasoned sub-question, and the initial reasoning result is an initial question reasoning result. Based on this, the data processing method provided by the present disclosure can be applied to a data processing node. After determining the initial question text, the data processing node inputs the initial question text into the trained question processing model, uses the question processing model to extract questions from the initial question text, thereby identifying a plurality of initial sub-questions contained in the initial question text; then, a to-be-reasoned sub-question is selected from the plurality of initial sub-questions, and the to-be-reasoned sub-question is reasoned in the first stage, thereby obtaining an initial question reasoning result for the to-be-reasoned sub-question.

[0076] It should be noted that the data processing node can be a server, a server, a virtual machine, a container, a cloud server, a client, a mobile terminal, or the like, or the data processing node can be an application program, a software system, or the like, which is not limited here.

[0077] In the above embodiments, in the process of using the data processing model to perform data inference on the initial data to obtain an initial inference result, the data processing model is used to determine to-be-inferred sub-data from a plurality of initial sub-data included in the initial data, and data inference is performed on the to-be-inferred sub-data to obtain the initial inference result. Since the data processing model can quickly and accurately perform data inference on relatively simple and small-amount sub-data, when the data processing model is used to perform multi-stage data inference on initial sub-data, efficient data inference on initial data can be achieved, and the problem of low inference efficiency of a neural network model in processing relatively complex and large-amount data is avoided.

[0078] In one or more embodiments provided in the present disclosure, the data processing method can be applied to a data processing node receiving initial data sent by a client based on a user, and the initial data can be subsequently processed. The specific implementation manner is as follows.

[0079] The initial data is determined, including:

[0080] The initial data sent by the client is received, wherein the initial data is generated by the client based on a data processing operation performed by the user on a data processing page.

[0081] The client can be a mobile terminal, a smart device, or other server devices, or the client can be an application program, a software system, or the like.

[0082] The data processing page can be understood as a user interaction interface displayed in the client and used to implement data processing of the user. The data processing page can be a web page, an application program interface, or the like.

[0083] The data processing operation can be an operation performed by the user based on the data processing page, such as inputting data, clicking a button, or the like to trigger a data processing control in the data processing page.

[0084] In the above example, the client can display a data processing page to the user, and the user can perform a data processing operation based on a data processing control (such as a button, an input box, or the like) in the data processing page, so as to trigger the client to generate initial data and send the initial data to the data processing node to instruct the data processing node to perform a data processing operation on the initial data using a data processing model.

[0085] In the above embodiments, the user performs a data processing operation on initial data by using the client and the data processing page, thereby meeting the user's need to use a data processing model to perform data processing on initial data.

[0086] Step 204: in the case where it is determined according to the initial reasoning result that the initial data satisfies a data reasoning condition, performing data updating on the initial data according to the initial reasoning result by using the data processing model to obtain updated data, wherein the data quantity of the updated data is less than the data quantity of the initial data, and the data processing model is a machine learning model.

[0087] The data reasoning condition can be understood as a condition for which data processing model needs to be used again for data reasoning. The data reasoning condition can be set according to an actual application scenario. For example, satisfying the data reasoning condition can be understood as not completing the data reasoning operation for all initial sub-data contained in the initial data, or the data updating can be performed based on the data reasoning result (for example, the initial reasoning result, the updated reasoning result). For another example, not satisfying the data reasoning condition can be understood as completing the data reasoning operation for all initial sub-data contained in the initial data, or the data updating cannot be performed based on the data reasoning result (for example, the initial reasoning result, the updated reasoning result).

[0088] The updated data can be understood as data obtained after the data updating on the initial data. For example, in the case where the initial data is "1+2*3 equals how much", and the initial reasoning result is "2*3 equals 6", the updated data can be "1+6 equals how much". For another example, in the case where the initial reasoning result is "int a=5; int b=10; can be written as a code statement int a=5,b=10;", the updated data can be "int a=5,b=10;"

[0089] In one or more embodiments provided in the present disclosure, in the process of performing multi-stage data reasoning, it can be checked at each stage whether to continue the reasoning, so as to determine whether the reasoning is completed in time. The specific implementation manner is as follows.

[0090] The determining, according to the initial reasoning result, that the initial data satisfies a data reasoning condition comprises:

[0091] determining initial sub-data contained in the initial data by using the data processing model;

[0092] performing semantic analysis on the initial sub-data and the initial reasoning result, and in the case where it is determined that the initial sub-data are multiple, and there is to-be-updated sub-data associated with the initial reasoning result in the multiple initial sub-data, determining that the initial data satisfies the data reasoning condition;

[0093] The to-be-updated sub-data can be understood as sub-data that needs to be updated using the initial inference result. For example, in a case where the plurality of sub-problem texts are "1+2 equals how much", "2*3 equals how much", and the initial inference result is "2*3 equals 6", the large model can determine that the sub-problem "2*3 equals how much" is to-be-updated sub-data from the plurality of sub-problem texts, and the to-be-updated sub-data can be updated based on the initial inference result.

[0094] Continuing with the above example, after determining the initial problem inference result "2*3 equals 6" using the data processing model, the initial problem inference result and the initial problem are input into the large model again for verification, so as to determine whether to continue problem inference (i.e., problem reduction). When it is determined that the initial problem includes a plurality of sub-problems, and there is a to-be-updated sub-problem "2*3 equals how much" corresponding to "2*3 equals 6" in the plurality of sub-problems, it is determined that problem reduction can still be continued.

[0095] In the above embodiment, in the process of multi-stage data inference, it is determined whether the inference is completed in time by checking whether to continue the inference at each stage, so as to avoid wasting of computing resources due to repeated inference.

[0096] In one or more embodiments provided by the present disclosure, the initial data is an initial problem text, the initial inference result is an initial problem inference result, and the data processing model is a problem processing model.

[0097] In a case where it is determined that the initial data satisfies the data inference condition according to the initial inference result, data of the initial data is updated according to the initial inference result using the data processing model, to obtain updated data.

[0098] In a case where it is determined that the initial problem text satisfies the problem inference condition according to the initial problem inference result, the initial problem text is updated according to the initial problem inference result using the problem processing model, to obtain an updated problem text.

[0099] The problem inference condition can be problem reduction.

[0100] For explanation and description of this embodiment, reference can be made to the corresponding or corresponding explanation in one or more other embodiments.

[0101] In one or more embodiments provided by the present disclosure, considering that the neural network model may have a low inference efficiency in processing relatively complex and large amount of data, the data processing method provided by the present disclosure can process the initial data through multi-stage data inference, and in the processing of each stage, the current stage of inference data (for example, the initial data) is updated using the data inference result to obtain data with a smaller amount, and the specific implementation manner is as follows.

[0102] The data updating of the initial data using the data processing model according to the initial inference result to obtain updated data comprises steps one and two.

[0103] Step one: using the data processing model, determining a plurality of initial sub-data contained in the initial data, and determining, from the plurality of initial sub-data, to-be-updated sub-data associated with the initial inference result.

[0104] Specifically, the step of using the data processing model to determine a plurality of initial sub-data contained in the initial data, and determining, from the plurality of initial sub-data, to-be-updated sub-data associated with the initial inference result comprises:

[0105] inputting the initial inference result and the initial data into the data processing model, using the data processing model to perform data extraction on the initial data, and determining the plurality of initial sub-data contained in the initial data;

[0106] performing semantic analysis on each initial sub-data and the initial inference result to obtain the to-be-updated sub-data associated with the initial inference result.

[0107] In the above example, after determining the initial problem inference result "2*3 equals 6" using the data processing model, the initial problem inference result and the initial problem are input into the large model again for semantic analysis, and from the plurality of sub-questions contained in the initial problem, the to-be-updated sub-question "2*3 equals how much" corresponding to "2*3 equals 6" is determined, so that subsequent problem reduction can be continued.

[0108] In the above embodiment, in the process of using the data processing model to determine a plurality of initial sub-data contained in the initial data, and determining, from the plurality of initial sub-data, to-be-updated sub-data associated with the initial inference result, the data processing model can be used to perform semantic analysis on the plurality of initial sub-data contained in the initial data, so as to determine the to-be-updated sub-data associated with the initial inference result, thereby facilitating subsequent accurate data updating of the to-be-updated sub-data using the initial inference result, avoiding semantic deviation of the initial data caused by updating errors, and ensuring the accuracy of the target data processing result.

[0109] Step two: using the data processing model, updating the to-be-updated sub-data in the initial data based on the initial inference result to obtain the updated data.

[0110] Specifically, using the data processing model, updating the to-be-updated sub-data in the initial data based on the initial inference result to obtain the updated data includes:

[0111] using the data processing model, replacing the to-be-updated sub-data in the initial data based on the initial inference result to obtain the updated data; or

[0112] using the data processing model, performing semantic rewriting on the to-be-updated sub-data in the initial data based on the initial inference result to obtain the updated data.

[0113] In the above example, the initial problem inference result can be "2x3 equals 6" or "6". Based on this, when the data processing model is used to determine the to-be-updated sub-problem "2x3 equals how much" corresponding to the initial problem inference result "6", the to-be-updated sub-problem "2x3 equals how much" in the initial problem text "1+2x3 equals how much" can be directly replaced based on the initial problem inference result to obtain the replaced problem text "1+6 equals how much", thereby realizing problem reduction.

[0114] In the case of the initial problem inference result being "2x3 equals 6", the data processing model can be used to perform semantic rewriting on the to-be-updated sub-data "2x3 equals how much" in the initial problem text "1+2x3 equals how much" based on the initial problem inference result "2x3 equals 6" to obtain the semantic rewritten problem text "1+6 equals how much", thereby realizing problem reduction.

[0115] In the above embodiments, the to-be-inferred data (e.g., initial data) at the current stage is updated using the data inference result to obtain data with a smaller number, thereby improving the data processing efficiency of the subsequent data processing model.

[0116] Step 206: using the data processing model to perform data inference on the updated data to obtain an updated inference result.

[0117] The data processing model is used to perform data inference on the updated data to obtain an updated inference result, which can be referred to the above step of using the data processing model to perform data inference on the initial data to obtain an initial inference result, and will not be described in detail here.

[0118] Continuing with the above example, the updated question text can be "1+6 equals how much", which is input into the large model for semantic reasoning to obtain the corresponding reasoning result "1+6 equals 7".

[0119] In a case where it is determined, according to the updated reasoning result, that the updated data does not satisfy the data reasoning condition, the updated reasoning result is determined as the target data processing result.

[0120] The target data processing result is understood as a processing result for the initial data. For example, in a case where the initial data is an initial question text, the target data processing result can be a question answer corresponding to the question text; in a case where the initial data is initial code data, the target data processing result can be code data obtained after optimization of the initial code data. It should be noted that the initial reasoning result can be a reasoning result for a certain initial sub-data in the initial data, the updated reasoning result can be a reasoning result for a certain updated sub-data in the updated data, and the target data processing can be a processing result for the entire initial data. In a case where the updated reasoning result is determined to not satisfy the data reasoning condition, the updated reasoning result can be determined as the target data processing result.

[0121] Continuing with the above example, in a case where the updated question reasoning result (i.e., the updated reasoning result) corresponding to the updated question text is "1+6 equals 7", the updated question reasoning result and the updated question are input into the large model again for semantic analysis. In a case where it is determined that the updated question only contains a sub-question "1+6 equals how much" and the sub-question is associated with the updated question reasoning result, it is determined that the multi-stage data reasoning is completed, so that the updated question reasoning result "the answer equals 7" is taken as the final answer (i.e., the target data processing result), or in a case where the updated question reasoning result is "1+6 equals 7", the "1+6 equals 7" is converted into a question answer "1+2*3 equals 7" (i.e., the target data processing result).

[0122] In one or more embodiments provided in the present disclosure, in the process of multi-stage data reasoning, it is necessary to check whether to continue reasoning at each stage, so as to determine whether the reasoning is completed in time. The specific implementation manner is as follows.

[0123] The determining, according to the updated reasoning result, that the updated data does not satisfy the data reasoning condition comprises:

[0124] determining an updated sub-data contained in the updated data by using the data processing model;

[0125] performing semantic analysis on the update sub-data and the update inference result, and determining that the update data does not satisfy the data inference condition in a case where it is determined that the update sub-data is one and the update sub-data is associated with the update inference result.

[0126] The update sub-data can be understood as sub-data contained in the update data. For example, in a case where the update data is an update question text, the update sub-data can be understood as a sub-question text contained in the update question text.

[0127] In the above example, in a case where it is determined that the update question inference result (i.e., the update inference result) corresponding to the update question text is "1+6 equals 7", the update question inference result and the update question are input again into the large model to perform semantic analysis, and in a case where it is determined that the update question only contains one sub-question "1+6 equals how many" and the sub-question is associated with the update question inference result, it is determined that the multi-stage data inference is completed.

[0128] In the above embodiments, in the process of performing multi-stage data inference, it is determined whether the inference is completed in time by checking whether to continue the inference at each stage, so as to avoid wasting of computing resources due to repeated inference.

[0129] In one or more embodiments provided in the disclosure, after the data processing node to which the data processing method is applied receives initial data sent by a user based on a client and processes the initial data, the target data processing result can be sent to the client for display. The specific implementation manner is as follows.

[0130] After the update inference result is determined as the target data processing result in a case where it is determined that the update data does not satisfy the data inference condition according to the update inference result, the method further includes:

[0131] The target data processing result is sent to the client, so that the target data processing result is displayed on the data processing page based on the client.

[0132] The target data processing result is displayed on the data processing page based on the client can be understood as that the target data processing result is displayed to the user based on the data processing page, or the target data processing result is displayed to another object based on the data processing page. The other object can be an object that needs to obtain the target data processing result. For example, the user can be a teacher, and the other object can be a student.

[0133] In the above embodiment, the target data processing result is displayed by using the client and the data processing page, thereby meeting the demand of the user for data processing of the initial data by using the data processing model.

[0134] In one or more embodiments provided in the present disclosure, during the process of multi-stage data reasoning, it is necessary to check whether to continue reasoning at each stage, so as to determine whether the reasoning is completed in time. The specific implementation manner is as follows.

[0135] After the data processing model is used to perform data reasoning on the update data and an update reasoning result is obtained, the method further includes:

[0136] In a case where it is determined, according to the update reasoning result, that the update data satisfies the data reasoning condition, second update data is obtained by using the data processing model to perform data update on the update data according to the update reasoning result, wherein a data quantity of the second update data is less than a data quantity of the update data.

[0137] The second update data is determined as the update data, and the step of using the data processing model to perform data reasoning on the update data and obtaining an update reasoning result is continuously executed until it is determined, according to the update reasoning result, that the update data does not satisfy the data reasoning condition.

[0138] Taking the above example, the initial question can be "1+1+2*3 equals how much", the update question can be "1+1+6 equals how much", and the update question reasoning result of the sub-question "1+1 equals how much" in the update question can be "1+1 equals 2". Based on this, when the data processing model determines that the update question reasoning result "1+1 equals 2", the update question reasoning result and the update question are input into the large model again for checking, so as to determine whether to continue the question reasoning (i.e. question reduction). When it is determined that the update question includes multiple sub-questions "1+1 equals how much" and "1+6 equals how much", and there is a to-be-updated sub-question "1+1 equals how much" corresponding to "1+1 equals 2" in the multiple sub-questions, it is determined that the question reduction can be continued.

[0139] The data processing method provided in one or more embodiments of the present disclosure, in the process of data processing by using a data processing model, first, the initial data is subjected to first-stage data reasoning by using the data processing model, and an initial reasoning result is obtained; and in the case where it is determined according to the initial reasoning result that the initial data satisfies a data reasoning condition, the initial data is updated according to the initial reasoning result, so as to obtain updated data with a smaller data quantity than the initial data; then, the updated data with a smaller data quantity is subjected to second-stage data reasoning by using the data processing model, so as to quickly obtain an updated reasoning result. Since the data processing model can quickly reason the updated data with a smaller data quantity, although multiple stages of reasoning are required, the overall data reasoning efficiency is higher than the data reasoning efficiency of the data processing model for data with a larger data quantity, thereby improving the data reasoning efficiency of the data processing model and avoiding the problem of low reasoning efficiency of the neural network model in processing data with a larger data quantity and being more complex. In the case where it is determined according to the updated reasoning result that the updated data does not satisfy the data reasoning condition, it is determined that the data reasoning is ended, and the updated reasoning result is determined as a target data processing result, thereby meeting the requirement of data processing of various types of data by using the neural network model in actual applications.

[0140] The data processing method provided in one or more embodiments of the present disclosure, in the process of data reasoning, is further described below with reference to FIG. 3. FIG. 3 is a flowchart of a processing process of a data processing method according to an embodiment of the present disclosure. As can be seen from FIG. 3, the data processing method provided in the present disclosure can combine Markov chain with CoT, PoT or ToT, and the core idea is to convert the problem into multiple independent problem reasoning and sub-problem generation. That is, the multiple-step reasoning problem is converted into a process of multiple independent problem reasoning and sub-problem generation, and the problem reasoning process is completed through multiple iterations. Specifically, the following steps are included.

[0141] Step 302: input a problem 1 provided by a user into an LLM for reasoning.

[0142] Specifically, the problem 1 can be understood as a problem that needs to be reasoned by using the LLM to obtain an answer. The problem 1 can be an original problem input by the user. For example, the problem 1 can be “Given that sin a = 1 / 2 and a is an angle in the first quadrant, solve tan a?”.

[0143] Step 304: determine a reasoning result 1 obtained by the LLM in reasoning the problem 1, and reduce the problem 1 based on the reasoning result 1 to obtain a problem 2.

[0144] Specifically, the data processing method provided by the present disclosure can complete the reasoning process of the problem in the form of multi-stage iteration. In this process, the original large model LLM uses the information of previous reasoning to perform reasoning at the current step (i.e., the current stage). Thus, the problem reasoning step for the original problem is converted into a more simplified process in which the large model only needs to complete one-step reasoning based on the current problem to complete the entire task reasoning.

[0145] In this one-step reasoning, the large model + tools or the large model CoT are used to simplify / transform and solve one or more sub-problems (i.e., known conditions) in the original problem.

[0146] Based on this, the execution mode of step 304 is that the LLM large model is used to perform one-step reasoning on the sub-problem in the current problem (i.e., problem 1). After one-step reasoning, the reasoning result 1 (i.e., step-wise solution) corresponding to the sub-problem is obtained.

[0147] Then, a new problem (i.e., problem 2) is obtained based on the original problem and the reasoning result 1. The new problem and the original problem have the same answer.

[0148] The new problem is obtained by problem reduction of problem 1 based on the reasoning result 1.

[0149] The problem reduction refers to replacing and rewriting the known conditions in the original problem based on the simplified conditions (i.e., reasoning result) in the one-step reasoning process by using the text generation capability of the large model, so as to obtain a new problem that is simpler than the original problem.

[0150] For example, the problem reduction is explained and described based on the above problem 1. The original problem is “Given that sin a = 1 / 2 and a is a first quadrant angle, solve tan a?” The reasoning result obtained by one-step reasoning is “because sin a = 1 / 2 and a is a first quadrant angle, a = 30 degrees can be obtained”. That is, the one-step reasoning is performed on the sub-problem “given that sin a = 1 / 2 and a is a first quadrant angle” in the original problem to obtain the corresponding reasoning result.

[0151] Then, based on the reasoning result, the sub-problem “given that sin a = 1 / 2 and a is a first quadrant angle” in the original problem “given that sin a = 1 / 2 and a is a first quadrant angle, solve tan a?” is replaced and rewritten, so as to convert the original problem into a new problem “given that a = 30 degrees, solve tan a?”.

[0152] Further, it needs to be explained that in the process of multi-stage problem reasoning, the large model needs to check whether to continue the problem reasoning at each stage, so as to determine whether the reasoning is completed in time and avoid wasting computing resources due to repeated reasoning; the way in which the large model checks whether to continue the problem reasoning can be referred to the following example:

[0153] For example, the original question can be "1+2×3 equals how much", and the corresponding multiple sub-questions are "1+2 equals how much" and "2×3 equals how much"; after the first stage of reasoning on the original question, it is determined that the reasoning result is "2×3 equals 6".

[0154] After determining that the reasoning result of one-step reasoning by the large model is "2×3 equals 6", the original question and the current reasoning result are input into the large model to check whether to continue the problem reasoning.

[0155] In the process of checking whether to continue the problem reasoning, the large model can determine whether the original question contains multiple sub-questions and whether there is a sub-question in the multiple sub-questions that can be reduced based on the reasoning result; if the original question contains multiple sub-questions and there is a sub-question that can be reduced, it is determined that the problem reasoning needs to be continued.

[0156] For example, the large model determines that the current original question contains two sub-questions "1+2 equals how much" and "2×3 equals how much", and there is a sub-question "2×3 equals how much" that can be reduced based on the reasoning result "2×3 equals 6", and it is determined that the problem reasoning needs to be continued.

[0157] In the case of determining that the problem reasoning needs to be continued, the original question is reduced based on the reasoning result to obtain a new question, and one-step reasoning is continued based on the new question until the data reasoning is completed.

[0158] Step 306: input question 2 into LLM to continue one-step reasoning.

[0159] Step 308: determine the reasoning result 2 obtained by the LLM for question 2, and reduce the question 2 based on the reasoning result 2.

[0160] Specifically, after one-step reasoning on the new question by the large model to obtain the reasoning result 2 (step-wise solution), the new question is reduced again based on the reasoning result 2 to obtain a question that is consistent with the answer of the new question but is simpler.

[0161] The above one-step reasoning and problem reduction steps are repeatedly executed until the most simplified question N is obtained.

[0162] For example, in the case of the original question being "Given that sin a = 1 / 2 and a is an angle in the first quadrant, solve tan a?", the most reduced question N can be "Given that a = 30 degrees, solve tan a?".

[0163] For another example, in the case of the original question being "1 + 2 x 3 equals how much?", the most reduced question N can be "1 + 6 equals how much?".

[0164] Step 310: input the question N into the LLM to continue the question reasoning.

[0165] Step 312: obtain the final reasoning result N.

[0166] Specifically, the above one-step reasoning and question reduction process is repeatedly performed until an answer (i.e., the reasoning result N) is obtained through question solving, and it is determined that the question reasoning is completed.

[0167] It should be noted that the judgment condition of the question reasoning result can be:

[0168] In the process of multi-stage question reasoning, the large model needs to check whether to continue the question reasoning at each stage. In the process of checking whether to continue the question reasoning, if it is determined that the original question only contains one sub-question and the sub-question cannot be reduced, it is determined that the question reasoning is completed.

[0169] For example, the large model determines that the current original question contains "1 + 6 equals how much" as a sub-question, and the sub-question cannot be further reduced, and determines that the question reasoning is completed.

[0170] Based on the above, the data processing method in one or more embodiments of the present disclosure provides an efficient reasoning method based on Markov thought chain. The above describes the application of the Markov thought chain reasoning (MCTR) method in mathematical reasoning tasks. Based on the above, the method regards a question as a state, regards an answer as an action, abstracts the answer process as a state transition process from the current state to the next state, and thus realizes logical reasoning through multiple rounds of Markov state transition.

[0171] It should be noted that the data processing method in one or more embodiments of the present disclosure considers the defects of the scheme of using CoT, PoT, ToT, etc. for logical reasoning, and proposes a corresponding solution.

[0172] Wherein, for low utilization of training data: the large model training data in the above scheme need to have a complete reasoning link for each question, which is relatively difficult to construct, and the current reasoning link can only satisfy the reasoning of this question, so the reasoning link is unique.

[0173] However, the data processing method in one or more embodiments of the present disclosure converts a complex problem into a series of difficult to simple sub-problems one-step reasoning problems, called atomic steps, after converting the model into the form of MCoT, which breaks the uniqueness of the original reasoning link and ensures that when solving new difficult problems, suitable steps can be found in the previous atomic steps to combine and solve.

[0174] Wherein, for low decoding efficiency: the data processing method in one or more embodiments of the present disclosure guides the reasoning efficiency / time of the large model to decrease significantly in the form of o(n^2) / exponential as the reasoning length increases, and after converting into the form of MCoT, the reasoning process of each sub-problem will be very short, which will significantly reduce the attention window of the large model and significantly reduce the demand for video memory of the large model decoding, improving the decoding efficiency.

[0175] Based on this, the data processing method in one or more embodiments of the present disclosure uses the Markov thought chain (MCoT) reasoning method to divide the mathematical logic reasoning into one-step reasoning and problem reduction two processes, and utilizes the independence of each sub-problem reasoning to realize the effective improvement of problem reduction on the data set and the decoding efficiency of the large model. The technical effect achieved by the method is:

[0176] First, through the one-step reasoning and problem reduction two processes, the original complex problem is simplified and converted into a simpler and more intuitive problem, and multiple complex problems can be converted into the same simple and intuitive problem (problem reduction process), thereby reducing the learning difficulty of the large model reasoning path on the original complex problem (the large model converts from learning to solve the complex problem to simplifying the complex problem and solving the simple problem. The combination of simple problems / one-step reasoning problems is more common in large model training corpus and is easier to learn), achieving more efficient use of pre-training models and task training data, and improving the utilization efficiency of data;

[0177] Secondly, by converting the original complex long sequence reasoning process into multiple short sequence reasoning processes, the inference cost of the large model is significantly reduced, and the inference efficiency of the large model is improved (because in the large model inference process, each step of inference needs to interact with all previous steps, making the decoding calculation and memory occupation very high, if the inference is performed in the form of a fragmented area, the decoding calculation and memory occupation can be significantly optimized), the current solution can make the training large model not need longcontext (i.e., long context), which is a longcontext replacement solution in inference generation; in combination with rag, it can even solve all the needs of longcontext.

[0178] Finally, the MCTR method has the advantages of: the process is more in line with human reasoning logic, the reasoning process is efficient and accurate, and the data set does not need to be artificially constructed.

[0179] Referring to FIG. 4, FIG. 4 shows a flowchart of a problem processing method according to one embodiment of the present disclosure, which specifically includes the following steps.

[0180] Step 402: Determine the initial problem text, and use a data processing model to perform problem reasoning on the initial problem text to obtain an initial problem reasoning result.

[0181] Step 404: In the case where it is determined according to the initial problem reasoning result that the initial problem text satisfies a problem reasoning condition, use the data processing model to perform problem updating on the initial problem text according to the initial problem reasoning result to obtain an updated problem text, wherein the data quantity of the updated problem text is less than the data quantity of the initial problem text, and the data processing model is a machine learning model.

[0182] Step 406: Use the data processing model to perform problem reasoning on the updated problem text to obtain an updated problem reasoning result.

[0183] Step 408: In the case where it is determined according to the updated problem reasoning result that the updated problem text does not satisfy the problem reasoning condition, determine the updated problem reasoning result as a target problem processing result.

[0184] In one or more embodiments of the present disclosure, the determination of the initial problem text includes:

[0185] Receiving the initial problem text sent by a client, wherein the initial problem text is generated by the client according to a problem processing operation performed by a user on a problem processing page;

[0186] The method further includes, in a case where it is determined, according to the update question reasoning result, that the update question text does not satisfy the question reasoning condition, determining the update question reasoning result as a target question processing result.

[0187] The target question processing result is sent to the client, so that the client displays the target question processing result to the user based on the question processing page.

[0188] In the question processing method provided in one or more embodiments of the present disclosure, in the process of question processing by using a question processing model, first, the question processing model is used to perform first-stage question reasoning on initial question text, to obtain an initial question reasoning result; and in a case where it is determined, according to the initial question reasoning result, that the initial question text satisfies a question reasoning condition, the initial question text is updated according to the initial question reasoning result, to obtain update question text with a smaller data quantity than the initial question text; then, the question processing model is used to perform second-stage question reasoning on the update question text with a smaller data quantity, to quickly obtain an update question reasoning result. Since the question processing model can quickly reason on the update question text with a smaller data quantity, although multiple stages of reasoning are required, the overall reasoning efficiency is higher than the question reasoning efficiency of the question processing model on question text with a larger data quantity, thereby improving the question reasoning efficiency of the question processing model and avoiding the problem of low reasoning efficiency of a neural network model in processing complex data with a large data quantity. In a case where it is determined, according to the update question reasoning result, that the update question text does not satisfy the question reasoning condition, it is determined to end the question reasoning, and the update question reasoning result is determined as a target question processing result, thereby meeting the requirement of using a neural network model to process various types of questions in actual applications.

[0189] The above is a schematic scheme of the question processing method of the present embodiment. It should be noted that the technical scheme of the question processing method belongs to the same concept as the technical scheme of the above data processing method, and details of the technical scheme of the question processing method that are not described in detail can be referred to the description of the technical scheme of the data processing method.

[0190] Corresponding to the above method embodiments, the present disclosure further provides data processing device embodiments. FIG. 5 shows a structural schematic diagram of a data processing device according to an embodiment of the present disclosure. As shown in FIG. 5, the device includes:

[0191] The first result determination module 502 is configured to determine initial data, and perform data reasoning on the initial data by using a data processing model, to obtain an initial reasoning result;

[0192] The data updating module 504 is configured to, in a case where it is determined according to the initial inference result that the initial data satisfies the data inference condition, perform data updating on the initial data according to the initial inference result by using the data processing model, to obtain updated data, wherein a data quantity of the updated data is less than a data quantity of the initial data, and the data processing model is a machine learning model.

[0193] The second result determining module 506 is configured to perform data inference on the updated data by using the data processing model, to obtain an updated inference result.

[0194] The third result determining module 508 is configured to, in a case where it is determined according to the updated inference result that the updated data does not satisfy the data inference condition, determine the updated inference result as a target data processing result.

[0195] Optionally, the data updating module 504 is further configured to:

[0196] determine a plurality of initial sub-data contained in the initial data by using the data processing model, and determine, from the plurality of initial sub-data, to-be-updated sub-data associated with the initial inference result;

[0197] perform data updating on the to-be-updated sub-data in the initial data based on the initial inference result by using the data processing model, to obtain the updated data.

[0198] Optionally, the data updating module 504 is further configured to:

[0199] perform replacement on the to-be-updated sub-data in the initial data based on the initial inference result by using the data processing model, to obtain the updated data; or

[0200] perform semantic rewriting on the to-be-updated sub-data in the initial data based on the initial inference result by using the data processing model, to obtain the updated data.

[0201] Optionally, the data updating module 504 is further configured to:

[0202] input the initial inference result and the initial data into the data processing model, perform data extraction on the initial data by using the data processing model, to determine the plurality of initial sub-data contained in the initial data;

[0203] perform semantic analysis on each initial sub-data and the initial inference result, to obtain the to-be-updated sub-data associated with the initial inference result.

[0204] Optionally, the first result determination module 502 is further configured to:

[0205] The initial data is input into the data processing model, and the data processing model is used to extract data from the initial data to determine multiple initial sub-data contained in the initial data;

[0206] The sub-data to be inferred is determined from the plurality of initial sub-data, and semantic reasoning is performed on the sub-data to be inferred to obtain the initial reasoning result.

[0207] Optionally, the data processing device further includes a data inference module, configured to:

[0208] If, based on the update inference result, it is determined that the updated data satisfies the data inference condition, the updated data is updated using the data processing model according to the update inference result to obtain second updated data, wherein the number of data in the second updated data is less than the number of data in the updated data.

[0209] The second updated data is determined as the updated data, and the step of using the data processing model to perform data reasoning on the updated data to obtain the updated reasoning result is continued until it is determined, based on the updated reasoning result, that the updated data does not meet the data reasoning condition.

[0210] Optionally, the first result determination module 502 is further configured to:

[0211] The client receives the initial data sent by the client, wherein the initial data is generated by the client based on the data processing operations performed by the user on the data processing page;

[0212] The data processing device further includes a result sending module, configured as follows:

[0213] The target data processing result is sent to the client so that the client can display the target data processing result based on the data processing page.

[0214] Optionally, the initial data is the initial question text, the initial reasoning result is the initial question reasoning result, and the data processing model is the question processing model;

[0215] The data update module 504 is further configured to:

[0216] If, based on the initial question reasoning result, it is determined that the initial question text satisfies the question reasoning condition, the initial question text is updated using the question processing model based on the initial question reasoning result to obtain the updated question text.

[0217] Optionally, the data updating module 504 is further configured to:

[0218] determine initial sub-data contained in the initial data by using the data processing model;

[0219] perform semantic analysis on the initial sub-data and the initial inference result, and determine that the initial data meets the data inference condition in a case where the initial sub-data is multiple and there is to-be-updated sub-data associated with the initial inference result in the multiple initial sub-data;

[0220] The third result determining module 508 is configured to:

[0221] determine update sub-data contained in the update data by using the data processing model;

[0222] perform semantic analysis on the update sub-data and the update inference result, and determine that the update data does not meet the data inference condition in a case where the update sub-data is one and the update sub-data is associated with the update inference result.

[0223] In the data processing apparatus provided in one or more embodiments of the present disclosure, in the process of data processing by using the data processing model, first, the data processing model is used to perform first-stage data inference on initial data, to obtain an initial inference result; and in a case where it is determined according to the initial inference result that the initial data meets a data inference condition, the initial data is updated according to the initial inference result, to obtain update data with a smaller data quantity than the initial data; then, the data processing model is used to perform second-stage data inference on the update data with a smaller data quantity, to quickly obtain an update inference result. Since the data processing model can quickly infer the update data with a smaller data quantity, although multiple stages of inference are required, the overall data inference efficiency is higher than that of the data processing model for data with a larger data quantity, thereby improving the data inference efficiency of the data processing model and avoiding the problem of low inference efficiency of the neural network model in processing data that is relatively complex and has a large data quantity. In a case where it is determined according to the update inference result that the update data does not meet the data inference condition, it is determined to end the data inference, and the update inference result is determined as a target data processing result, thereby meeting the requirement of data processing of various types of data by using the neural network model in actual applications.

[0224] The above is a schematic scheme of the data processing apparatus of the embodiment. It should be noted that the technical scheme of the data processing apparatus and the technical scheme of the data processing method described above belong to the same concept, and the details of the technical scheme of the data processing apparatus that are not described in detail can be referred to the description of the technical scheme of the data processing method.

[0225] Corresponding to the method embodiments described above, the disclosure also provides problem processing apparatus embodiments. FIG. 6 shows a structural schematic diagram of a problem processing apparatus according to an embodiment of the disclosure. As shown in FIG. 6, the apparatus includes:

[0226] A first result determination module 602 configured to determine an initial question text, and perform question reasoning on the initial question text by using a data processing model to obtain an initial question reasoning result;

[0227] A question updating module 604 configured to, in a case where it is determined according to the initial question reasoning result that the initial question text satisfies a question reasoning condition, perform question updating on the initial question text by using the data processing model according to the initial question reasoning result to obtain an updated question text, wherein a data quantity of the updated question text is less than a data quantity of the initial question text, and the data processing model is a machine learning model;

[0228] A second result determination module 606 configured to perform question reasoning on the updated question text by using the data processing model to obtain an updated question reasoning result;

[0229] A third result determination module 608 configured to, in a case where it is determined according to the updated question reasoning result that the updated question text does not satisfy the question reasoning condition, determine the updated question reasoning result as a target question processing result.

[0230] Optionally, the first result determination module 602 is further configured to:

[0231] receive the initial question text sent by a client, wherein the initial question text is generated by the client according to a question processing operation performed by a user on a question processing page;

[0232] The problem processing apparatus further includes a result sending module configured to:

[0233] send the target question processing result to the client, so that the client displays the target question processing result to the user based on the question processing page.

[0234] The problem processing device provided in one or more embodiments of the present disclosure, in the process of problem processing by using a problem processing model, first, the initial problem text is subjected to first-stage problem reasoning by using the problem processing model to obtain an initial problem reasoning result; and in the case where it is determined according to the initial problem reasoning result that the initial problem text satisfies the problem reasoning condition, the initial problem text is updated according to the initial problem reasoning result, thereby obtaining an updated problem text with a smaller data quantity than the initial problem text; then, the problem processing model is used to perform second-stage problem reasoning on the updated problem text with a smaller data quantity, thereby quickly obtaining an updated problem reasoning result. Since the problem processing model can quickly reason on the updated problem text with a smaller data quantity, although multiple stages of reasoning are required, the overall reasoning efficiency will be higher than the problem reasoning efficiency of the problem processing model for problem texts with a larger data quantity, thereby improving the problem reasoning efficiency of the problem processing model and avoiding the problem of low reasoning efficiency of the neural network model in processing complex data with a large data quantity. In the case where it is determined according to the updated problem reasoning result that the updated problem text does not satisfy the problem reasoning condition, the problem reasoning is determined to be ended, and the updated problem reasoning result is determined as a target problem processing result, thereby meeting the demand in actual application for processing various types of problems by using a neural network model.

[0235] The above is a schematic scheme of the problem processing device of the present embodiment. It should be noted that the technical scheme of the problem processing device belongs to the same concept as the technical scheme of the problem processing method described above, and the details of the technical scheme of the problem processing device that are not described in detail can be referred to the description of the technical scheme of the problem processing method.

[0236] FIG. 7 shows a structural block diagram of a computing device 700 according to an embodiment of the present disclosure. The components of the computing device 700 include but are not limited to a memory 710 and a processor 720. The processor 720 is connected with the memory 710 through a bus 730, and a database 750 is used to save data.

[0237] The computing device 700 also includes an access device 740 that enables the computing device 700 to communicate via one or more networks 760. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or combinations of such networks, such as the Internet. The access device 740 can include one or more of any type of network interface (for example, a network interface card (NIC)), such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC).

[0238] In one embodiment of the present disclosure, the above-mentioned components of the computing device 700 and other components not shown in FIG. 7 can also be connected to each other, for example, through a bus. It should be understood that the computing device structure block diagram shown in FIG. 7 is only for the purpose of example, and is not a limitation on the scope of the present disclosure. Those skilled in the art can add or replace other components as needed.

[0239] The computing device 700 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (for example, a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, and the like), a mobile phone (for example, a smartphone), a wearable computing device (for example, a smartwatch, smart glasses, and the like), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 700 can also be a mobile or stationary server.

[0240] The processor 720 is configured to execute computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned data processing method or problem processing method.

[0241] Each of the embodiments in the present disclosure is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, for the computing device embodiment, since it is basically similar to the data processing method or problem processing method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the data processing method or problem processing method embodiment.

[0242] An embodiment of the present disclosure further provides a computer readable storage medium storing computer programs / instructions, which are executed by a processor to implement the steps of the above data processing method or problem processing method.

[0243] Each of the embodiments in the present disclosure is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, for the computer readable storage medium embodiment, since it is basically similar to the data processing method or problem processing method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the data processing method or problem processing method embodiment.

[0244] An embodiment of the present disclosure further provides a computer program product, which includes computer programs / instructions, which are executed by a processor to implement the steps of the above data processing method or problem processing method.

[0245] The above is a schematic scheme of the computer program product of the embodiment. It should be noted that the technical scheme of the computer program product and the technical scheme of the above data processing method or problem processing method belong to the same concept, and the details of the technical scheme of the computer program product which are not described in detail can be referred to the description of the technical scheme of the above data processing method or problem processing method.

[0246] The above describes a specific embodiment of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps recited in the claims can be performed in a different order than the order in which they are recited and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0247] The computer readable medium can include any entity or apparatus capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, software distribution medium, etc. It should be noted that the computer readable medium can include appropriate additions or subtractions according to the requirements of patent practice. For example, according to the patent practice in some regions, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0248] It should be noted that, for the foregoing method embodiments, in order to facilitate description, they are all described as a combination of a series of actions, but those skilled in the art should know that the disclosed embodiments are not limited to the order of the actions described, because according to the disclosed embodiments, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the disclosure are all preferred embodiments, and the actions and modules involved are not necessarily essential to the disclosed embodiments.

[0249] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0250] The preferred embodiments of the disclosure disclosed above are only used to help explain the disclosure. The alternative embodiments do not describe all the details and do not limit the disclosure to the specific embodiments described. Obviously, according to the content of the disclosed embodiments, many modifications and changes can be made. The disclosure selects and describes these embodiments in order to better explain the principles and practical applications of the disclosed embodiments, so that those skilled in the art can well understand and utilize the disclosure. The disclosure is limited only by the claims and their full scope and equivalents.

Claims

1. A data processing method comprising: determining initial data and performing data inference on the initial data using a data processing model to obtain an initial inference result; in a case where it is determined according to the initial inference result that the initial data satisfies a data inference condition, performing data update on the initial data using the data processing model according to the initial inference result to obtain updated data, wherein a data quantity of the updated data is less than a data quantity of the initial data, and the data processing model is a machine learning model; performing data inference on the updated data using the data processing model to obtain an updated inference result; in a case where it is determined according to the updated inference result that the updated data does not satisfy the data inference condition, determining the updated inference result as a target data processing result.

2. The data processing method of claim 1, wherein the performing data update on the initial data using the data processing model according to the initial inference result to obtain updated data comprises: determining a plurality of initial sub-data contained in the initial data using the data processing model, and determining to-be-updated sub-data associated with the initial inference result from the plurality of initial sub-data; performing data update on the to-be-updated sub-data in the initial data based on the initial inference result using the data processing model to obtain the updated data.

3. The data processing method of claim 2, wherein the performing data update on the to-be-updated sub-data in the initial data based on the initial inference result using the data processing model to obtain the updated data comprises: replacing the to-be-updated sub-data in the initial data based on the initial inference result using the data processing model to obtain the updated data; or performing semantic rewriting on the to-be-updated sub-data in the initial data based on the initial inference result using the data processing model to obtain the updated data.

4. The data processing method of claim 2 or 3, wherein the determining a plurality of initial sub-data contained in the initial data using the data processing model, and determining to-be-updated sub-data associated with the initial inference result from the plurality of initial sub-data comprises: inputting the initial inference result and the initial data into the data processing model, and performing data extraction on the initial data using the data processing model to determine the plurality of initial sub-data contained in the initial data; performing semantic analysis on each initial sub-data and the initial inference result to obtain the to-be-updated sub-data associated with the initial inference result.

5. The data processing method of any one of claims 1 to 4, wherein the performing data inference on the initial data using the data processing model to obtain an initial inference result comprises: inputting the initial data into the data processing model, and performing data extraction on the initial data using the data processing model to determine a plurality of initial sub-data contained in the initial data. ​ determining to-be-reasoned sub-data from the plurality of initial sub-data, and performing semantic reasoning on the to-be-reasoned sub-data to obtain the initial reasoning result. 6.The data processing method of any one of claims 1 to 5, after the data reasoning on the update data by using the data processing model to obtain an update reasoning result, further comprising: in a case where it is determined, according to the update reasoning result, that the update data satisfies the data reasoning condition, performing data update on the update data by using the data processing model according to the update reasoning result to obtain second update data, wherein a data quantity of the second update data is less than a data quantity of the update data; determining the second update data as update data, and continuing to perform the data reasoning on the update data by using the data processing model to obtain an update reasoning result until it is determined, according to the update reasoning result, that the update data does not satisfy the data reasoning condition. 7.The data processing method of any one of claims 1 to 6, wherein the determining initial data comprises: receiving the initial data sent by a client, wherein the initial data is generated by the client according to a data processing operation performed by a user on a data processing page; after the determining, according to the update reasoning result, that the update data does not satisfy the data reasoning condition, the update reasoning result as a target data processing result, further comprising: sending the target data processing result to the client to enable the client to display the target data processing result based on the data processing page. 8.The data processing method of any one of claims 1 to 7, wherein the initial data is initial question text, the initial reasoning result is an initial question reasoning result, and the data processing model is a question processing model; the performing, according to the initial reasoning result, data update on the initial data by using the data processing model to obtain update data in a case where it is determined, according to the initial reasoning result, that the initial data satisfies a data reasoning condition, comprises: performing, according to the initial question reasoning result, question update on the initial question text by using the question processing model to obtain update question text in a case where it is determined, according to the initial question reasoning result, that the initial question text satisfies a question reasoning condition. 9.The data processing method of any one of claims 1 to 8, wherein the determining, according to the initial reasoning result, that the initial data satisfies a data reasoning condition, comprises: determining initial sub-data contained in the initial data by using the data processing model; performing semantic analysis on the initial sub-data and the initial reasoning result, and determining that the initial data satisfies the data reasoning condition in a case where it is determined that the initial sub-data are multiple, and there is to-be-updated sub-data associated with the initial reasoning result in the multiple initial sub-data; the determining, according to the update reasoning result, that the update data does not satisfy the data reasoning condition, comprises: determining update sub-data contained in the update data by using the data processing model; performing semantic analysis on the update sub-data and the update reasoning result, and determining that the update data does not satisfy the data reasoning condition in a case where the update sub-data is one and the update sub-data is associated with the update reasoning result.

10. A question processing method, comprising: determining initial question text, and performing question reasoning on the initial question text by using a data processing model to obtain an initial question reasoning result; in a case where it is determined that the initial question text satisfies a question reasoning condition according to the initial question reasoning result, performing question updating on the initial question text by using the data processing model according to the initial question reasoning result to obtain update question text, wherein a data quantity of the update question text is less than a data quantity of the initial question text, and the data processing model is a machine learning model; performing question reasoning on the update question text by using the data processing model to obtain an update question reasoning result; in a case where it is determined that the update question text does not satisfy the question reasoning condition according to the update question reasoning result, determining the update question reasoning result as a target question processing result.

11. The question processing method of claim 10, wherein the determining initial question text comprises: receiving the initial question text sent by a client, wherein the initial question text is generated by the client according to a question processing operation performed by a user on a question processing page; after the determining the update question reasoning result as the target question processing result in the case where it is determined that the update question text does not satisfy the question reasoning condition according to the update question reasoning result, the method further comprises: sending the target question processing result to the client to enable the client to display the target question processing result to the user based on the question processing page.

12. A computing device, comprising: a memory and a processor; the memory is configured to store computer programs / instructions, and the processor is configured to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the data processing method of any one of claims 1 to 9 or the question processing method of any one of claims 10 to 11.

13. A computer-readable storage medium storing computer programs / instructions, which, when executed by a processor, implement the steps of the data processing method of any one of claims 1 to 9 or the question processing method of any one of claims 10 to 11.

14. A computer program product comprising computer programs / instructions, which, when executed by a processor, implement the steps of the data processing method of any one of claims 1 to 9 or the question processing method of any one of claims 10 to 11.

Images