Question processing method and device

Abstract

The embodiment of the invention provides a question processing method and device.The method comprises the steps that question text data corresponding to question image data are determined, and the question text data are text data not matched with question answer data; the question text data is input into a question analysis model, question integrity information corresponding to the question text data output by the question analysis model is obtained, and the question analysis model is a model which is based on a pre-training model and is subjected to fine adjustment through training data marked according to question solvability rules; and determining a question processing strategy corresponding to the question image data according to the question integrity information, and processing the question image data based on the question processing strategy. Accurate recognition of deep semantic incompleteness of the topic text is realized, and the accuracy and recall rate of judgment are remarkably improved. And the occupation of subsequent storage and computing power resources by invalid data is fundamentally avoided.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of image processing technology, and in particular to a problem processing method and apparatus. Background Technology

[0002] Against the backdrop of rapid development in digital education, mobile-based photo-based question search technology has become a widely used learning aid. However, the industry faces a significant challenge in automating the identification of incomplete question text information caused by incomplete user photos, poor image quality, or OCR recognition errors. If these incomplete questions are allowed to proceed directly to processing or storage without filtering, they not only waste valuable computing and storage resources but may also be mistakenly pushed to other users in the future, thereby compromising the seriousness of the learning experience and the reliability of the product. Therefore, developing a precise, efficient, and universal question integrity filtering mechanism is a pressing issue that needs to be addressed. Summary of the Invention

[0003] In view of the above, embodiments of this specification provide a problem processing method. One or more embodiments of this specification also relate to a problem processing apparatus, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0004] According to a first aspect of the embodiments of this specification, a problem processing method is provided, including: Determine the question text data corresponding to the question image data, wherein the question text data is text data for which no question answer data was matched; The question text data is input into the question analysis model to obtain the question completeness information corresponding to the question text data output by the question analysis model. The question analysis model is a model based on a pre-trained model, which is fine-tuned using training data labeled according to the question solvability rules. Based on the question integrity information, a question processing strategy corresponding to the question image data is determined, and the question image data is processed based on the question processing strategy.

[0005] According to a second aspect of the embodiments of this specification, a question processing apparatus is provided, comprising: The determination module is configured to determine the question text data corresponding to the question image data, wherein the question text data is text data for which no question answer data was matched; The input / output module is configured to input the question text data into the question analysis model and obtain the question completeness information corresponding to the question text data output by the question analysis model. The question analysis model is a model based on a pre-trained model and fine-tuned using training data labeled according to the question solvability rules. The processing module is configured to determine the question processing strategy corresponding to the question image data based on the question integrity information, and to process the question image data based on the question processing strategy.

[0006] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned problem-solving method.

[0007] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of the above-described problem-solving method.

[0008] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described problem processing method.

[0009] One embodiment of this specification implements a question analysis model to determine the completeness of the question text data when no matching answer data is found in the question image data. This achieves accurate identification of deep semantic deficiencies in the question text, significantly improving the accuracy and recall rate. Based on the question completeness information output by the question analysis model, a question processing strategy corresponding to the question image data is determined. By applying differentiated question processing strategies to the question image data, automated decision-making and resource optimization are achieved. This fundamentally avoids invalid data consuming subsequent storage and computing resources, eliminates the risk of displaying incomplete questions to users, and ensures user experience and product data quality. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a problem-solving method provided in one embodiment of this specification; Figure 2 This is a flowchart illustrating the process of a problem-solving method provided in one embodiment of this specification. Figure 3 This is a schematic diagram of the structure of a problem processing device provided in one embodiment of this specification; Figure 4 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0011] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0012] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0013] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0014] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0015] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0016] Large Language Model (LLM): A Large Language Model is a natural language processing model built using deep learning techniques. After pre-training, LLM possesses powerful general language understanding and generation capabilities. It can not only understand the literal meaning of text but also perform a wide range of tasks such as contextual reasoning, summarization, translation, and question answering. In the embodiments described in this specification, its deep semantic understanding and logical reasoning capabilities are primarily utilized to determine the completeness and solvability of the question text.

[0017] Supervised Fine-Tuning of Large Models (SFT) is a machine learning strategy primarily used to handle large-scale pre-trained models. These models are typically trained to understand or learn from large amounts of unlabeled data, such as web pages, books, and other types of text. However, despite being trained, they may not directly solve specific tasks, such as text classification or sentiment analysis. Supervised fine-tuning is then necessary. This process mainly involves further training using labeled datasets. The parameters of some (or all) layers of the model are updated based on this new task; they are fine-tuned to better perform this specific task. In the embodiments described in this specification, a dedicated question analysis model is created through supervised fine-tuning of data labeled according to specific rules.

[0018] Incomplete questions: In the specific application scenario of photo-based question search, incomplete questions refer to questions whose text obtained after Optical Character Recognition (OCR) processing is incomplete and cannot form a valid question that can be solved independently.

[0019] In the field of digital education, photo-based question searching has become a core means for students to obtain instant learning assistance. Its general technical process typically includes: the user takes a picture of the question and uploads an image; the server preprocesses the image and converts it into machine-readable text using OCR technology; then, a similarity matching algorithm retrieves the corresponding answer and explanation from an existing question bank. In this process, for questions that cannot be matched with an answer, the system usually needs to include them in an "entry quality control" stage to expand the question bank resources. A key task in this stage is to determine whether the question is complete, that is, to identify and filter out "incomplete questions" due to incomplete photos, blurry images, or OCR recognition errors. Failure to effectively intercept such incomplete questions will result in invalid data consuming storage and computing resources, and may be incorrectly displayed to users later, seriously affecting product experience and credibility.

[0020] To automate the process of determining the completeness of questions (or "incompleteness"), the industry has made numerous attempts, mainly resulting in the following two representative technical solutions: 1. Rule-based approach: This approach relies on a pre-compiled rule base to determine the completeness of a question. Its typical implementation involves first formatting the OCR output text, then comparing it against a series of pre-defined, rigid rules. These rules typically include, but are not limited to: text length thresholds (e.g., questions with fewer than a certain number of characters are considered incomplete), structure checks (e.g., whether they contain question markers or punctuation), and subject-specific keyword matching (e.g., math problems should contain numbers or operators). Once the text triggers any of these rules, it is deemed "incomplete" and discarded.

[0021] 2. Lightweight Pre-trained Model-Based Approach: This approach utilizes pre-trained language models with relatively small parameter sizes, such as BERT, and fine-tunes them through supervised learning to train a binary classification model ("complete" or "incomplete"). The typical implementation process is as follows: First, collect a batch of question texts already manually labeled with completeness tags as training data; then, select a lightweight model for fine-tuning on this dataset; finally, input the OCR text to be judged into the fine-tuned model and make a judgment based on its output probability value.

[0022] However, the above-mentioned technical solutions all have significant limitations in practical applications and cannot meet the requirements for high accuracy, strong generalization, and low cost in disability assessment. The specific problems are as follows: (1) The accuracy rate of disability assessment is low and the number of missed assessments is serious. Rule-based solutions can only capture surface-level text features and are completely powerless against "implicit incompleteness" at the semantic level. For example, a question may be grammatically correct, of the required length, and contain a question mark, but is actually unsolvable due to the lack of a crucial condition; such questions would be misclassified as complete by rule-based solutions. While lightweight model-based solutions possess some semantic understanding capabilities, their model capacity and inference depth are limited, making them insufficient for understanding and judging long texts, complex logical relationships, or questions containing complex formulas or diagrams. Furthermore, their generalization performance heavily relies on the coverage of the training data, making them prone to missed detections for incomplete or novel question types not present in the training set. Both types of solutions result in a large number of incomplete questions failing to be effectively recalled and flowing into subsequent stages.

[0023] (2) The solution has poor adaptability and is difficult to cover multiple scenarios. The rule base of the rule-based approach relies heavily on manual summarization and maintenance, making it difficult to encompass all subjects (such as Chinese, physics, and chemistry, each with its own unique structure), all question types, and the ever-increasing number of questions in new textbooks, resulting in incomplete scenario coverage. The performance of the small-model approach, on the other hand, is strongly correlated with the subject and question type distribution of the training data. A model trained well in one domain (such as math multiple-choice questions) may show a significant performance drop when handling questions in other domains (such as Chinese reading comprehension), exhibiting poor cross-scenario and cross-subject versatility.

[0024] (3) High dependence on manual labor and huge maintenance costs Rule-based solutions require technical personnel to continuously and manually add and adjust the rule base as new question types and problems emerge, resulting in high maintenance costs and delays. Small-model solutions rely on large amounts of high-quality labeled data for training and iterating the model, leading to high manpower and time costs for data preparation. More importantly, due to the insufficient accuracy of automatic disability assessment in existing solutions, significant manpower is often still required for secondary manual review in actual business operations, keeping overall costs high. Furthermore, inconsistent manual judgment standards make it difficult to maintain stable results.

[0025] (4) Leading to resource waste and experience risks Due to the aforementioned issues, existing technical solutions cannot effectively intercept incomplete questions at the source. This invalid data will enter subsequent processes such as question labeling, storage, retrieval, and even manual question completion, wasting a significant amount of storage space and computing resources. Most seriously, some incomplete questions may ultimately be displayed to users searching for answers, providing solutions based on incomplete information, directly harming user experience and the core value of the product.

[0026] In summary, existing technologies struggle to achieve a highly accurate, highly recall, strongly generalizable, and low-maintenance automated question integrity filtering method for the specific scenario of image-based question search. Overcoming these shortcomings and constructing a technical solution capable of accurately understanding question semantics, intelligently determining its solvability, and thus efficiently filtering incomplete questions is a pressing technical problem in this field.

[0027] Based on this, this specification provides a problem processing method, and also relates to a problem processing apparatus, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0028] See Figure 1 , Figure 1 A flowchart of a problem processing method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0029] Step 102: Determine the question text data corresponding to the question image data, wherein the question text data is text data for which no question answer data has been matched.

[0030] The question image data can be understood as a digital image file containing the question content uploaded by the end user through a photo-based question search application. The corresponding question text data is the text data identified from the question image using image recognition models or OCR recognition technology.

[0031] In practice, the system uses the text corresponding to each question image as the query input and performs similarity matching in a pre-built question bank retrieval system. The retrieval system typically employs techniques such as semantic vector matching to identify question records in the question bank that have the same or highly similar intent to the query question and already include correct answers and explanations, thus providing users with question search services based on the retrieved content. However, some questions may fail to reach the preset entry threshold in terms of similarity score after the above retrieval. This indicates that the question has not been included in the existing knowledge base or has not been successfully retrieved, thus entering the new question processing flow. The question answer data is the answer text that should be retrieved from the question image data; however, these unretrieved question texts will be treated as question text data, meaning the question text data consists of text data that did not match the question answer data.

[0032] In practical applications, during the image-based question-solving process, the system continuously receives massive amounts of user-uploaded question images. After OCR recognition, this generates a large amount of text data. If all recognized text is subjected to subsequent complex and computationally expensive intelligent integrity analysis without differentiation, it will lead to severe resource misallocation and waste. This is because many questions already have answers in the question bank, requiring no integrity judgment; or the image quality is extremely poor, rendering the recognized text itself worthless. Therefore, the core purpose of this step is to establish a precise pre-screening mechanism. It is responsible for identifying and separating from all OCR results those "candidate objects" that truly have potential for inclusion in the question bank and must undergo rigorous quality control—that is, question texts for which no existing answers have been found in the current question bank. This ensures that subsequent valuable AI computing power and processing can be focused on the subset of data most in need of evaluation.

[0033] In one specific embodiment of this specification, assume a user takes a photo of a high school math problem and uploads it. The system recognizes the problem image data and obtains the corresponding recognized text data: "Given the function f(x) = x³ - 3x, find its maximum and minimum values ​​on the interval [-2, 2]". The system converts this text into a vector and searches in the question bank. Since the question bank already contains the same problem with a detailed solution, the search is successful, and the answer data is matched. At this point, the system will directly return the answer to the user, and this text will not be marked as unmatched, therefore it will not enter the subsequent integrity filtering process.

[0034] In a specific embodiment of this specification, if the user photographs a relatively obscure question from a newly published educational supplement, the OCR result is: "If the complex number z satisfies |z-2+3i|=1, then the range of values ​​for |z| is?". After searching the question bank, no sufficiently similar questions were found, and the match failed. In this case, the OCR text is determined as "text data for which no matching answer was found," and as the question text data corresponding to the question image data, further in-depth integrity analysis is required.

[0035] Based on this, by introducing "unsuccessful matching" as a key gate to trigger subsequent intelligent analysis, intelligent process diversion is achieved. This avoids indiscriminate and costly AI analysis of all questions (including solved questions and severely ambiguous invalid questions), and concentrates computing resources on truly unknown questions with the potential to be included in the database, significantly reducing the overall operational computing cost.

[0036] Furthermore, since the raw data submitted by the user is in image format, it cannot be directly processed by retrieval algorithms or analysis models. Therefore, it is necessary to identify the question text from the question image. Specifically, determining the question text data corresponding to the question image data includes: identifying the question image data and performing character recognition on the question image data; and determining the question text data corresponding to the question image data based on the character recognition results.

[0037] The question image data consists of raw question image files received from the user's end or read from storage. Character recognition of the question image data involves using image recognition models or OCR technology to identify characters in the image. The question text data is then determined based on the character recognition results. This can be achieved by generating structured question text data from the text identified in the image through steps such as noise reduction and sorting.

[0038] In practical applications, the original data submitted by users is in image format and cannot be directly processed by retrieval algorithms or analysis models. If fuzzy matching or analysis is directly performed based on images, the accuracy and efficiency will be extremely low. Therefore, the purpose of the implementation methods in this part is to clarify and refine the specific technical path of the process of "determining the question text data", that is, how to reliably convert unstructured image inputs into structured text data to provide an accurate data basis for subsequent matching and judgment.

[0039] During specific implementation, before recognition, the image is usually preprocessed to improve OCR accuracy, including but not limited to grayscale conversion, binarization, denoising, skew correction, layout analysis, etc., aiming to optimize the image quality and highlight the text area. The preprocessed image is used for text detection and recognition by an OCR engine. This engine locates the text lines or word regions in the image through a deep learning model and converts them into corresponding character encodings (such as UTF-8 text). The original text recognized by OCR is cleaned and normalized, which may include correcting common OCR errors (such as the confusion between "0" and "O"), removing irrelevant characters, merging text lines that are wrongly segmented, etc. The post-processed text is encapsulated into a structured data object (such as a JSON string), and this object is the question text data that uniquely corresponds to the original question image and can be consumed by downstream processes. This data object usually contains at least the recognized text content and may also include meta-information such as recognition confidence and text location.

[0040] In a specific embodiment of this specification, assume that a student uses a photo search question application to take a photo of a physics experiment question and submit it. The following operations are sequentially performed by the system backend: The server receives the uploaded photo file of the physics question. The system first automatically preprocesses the photo, for example, rotates the picture to restore the text lines to horizontal. Subsequently, the processed image is sent to the OCR service. The OCR engine successfully recognizes the text content in the image, and its initial output may contain some text with messy formats. The system post-processes the initial OCR text, for example, automatically corrects the word recognized as "density" to "density" according to the context and reasonably integrates the sentences that are wrongly line-wrapped. Finally, the system generates a clean and complete text description, which is: "In the experiment of measuring the density of a liquid, a balance, a graduated cylinder, a beaker, water, and the liquid to be measured are provided. Please write out the experimental steps and derive the density calculation formula." This text together with its source information is jointly determined as the question text data corresponding to this physics question image and then enters the question bank matching link for retrieval.

[0041] Based on this, through systematic image preprocessing and intelligent text post-processing, the quality and consistency of the original input data are significantly improved, providing a reliable and clean input source for subsequent precise matching and in-depth semantic analysis, and reducing the introduction of errors at the data entry point.

[0042] Step 104: Input the question text data into the question analysis model to obtain the question completeness information corresponding to the question text data output by the question analysis model. The question analysis model is a model based on a pre-trained model, which is fine-tuned using training data labeled according to the question solvability rules.

[0043] The question analysis model is a machine learning model with a specific parameter scale, specifically optimized for the task of "determining the solvability of a question." This model is not trained from scratch but starts with a pre-trained model, such as a large language model. To enable the general-purpose pre-trained model to excel in the specific task of "incompleteness detection," the embodiments in this specification have undergone supervised fine-tuning using training data.

[0044] In practice, the training data is labeled according to a set of explicit "question solvability rules." These rules focus on whether a question is "solvable," covering multiple dimensions such as information completeness, sufficiency of conditions, and structural compliance (e.g., whether there are missing options, insufficient conditions, or incomplete materials). The fine-tuning process involves supervised learning of the pre-trained model using the training data. This process guides the model to adjust its internal parameters, focusing its powerful general language capabilities on the specific goal of "evaluating question completeness based on solvability rules," thus making it an expert in the field. After obtaining the fine-tuned model, it can be used as a question analysis model. Inputting question text data into the model yields the question completeness information output by the model. Question completeness information can be understood as structured judgment information. It contains at least an overall judgment about whether the question is incomplete (e.g., "complete" or "incomplete"). In practice, this information often includes finer-grained cause classifications (e.g., "incomplete, cause: missing key physical quantity—mass").

[0045] In practical applications, traditional methods based on surface rules or lightweight models lack deep semantic understanding capabilities, making them unable to effectively identify deeper issues such as logical incompleteness and missing conditions, resulting in low accuracy. The purpose of this step is to introduce a dedicated question analysis model with deep understanding and reasoning capabilities to perform in-depth analysis of the question text, thereby outputting accurate completeness judgments.

[0046] In a specific embodiment of this specification, the system has identified an OCR text as an unmatched question: "An object starts from rest and undergoes uniformly accelerated linear motion. Its displacement in the first second is 2 meters. Find the acceleration of the object." This text is input into a finely tuned question analysis model. After analysis and reasoning, the model generates and outputs question completeness information, such as "Judgment result: Complete, Confidence level: 0.95".

[0047] In another specific embodiment of this specification, if the input text is: "An object undergoes uniformly accelerated linear motion. It displaces 2 meters in the first second. Find its velocity at the end of the fifth second." The model analysis and reasoning find that it requires solving for the final velocity Vt, but the known conditions only include a displacement (S1) and the corresponding time, lacking crucial information—acceleration (a) or other intermediate displacements. The existing conditions alone cannot uniquely determine the answer. The model outputs: "Judgment result: Incomplete, Reason category: Insufficient solution conditions, Explanation: A unique acceleration cannot be derived from the given conditions, therefore the velocity at the end of the fifth second cannot be solved."

[0048] Based on this, a dedicated analysis model based on a pre-trained large model and specifically fine-tuned was deployed to achieve deep-level, semantic-level accurate judgment. This greatly improved the accuracy and recall of disability assessment, fundamentally solving the problem of serious missed cases in traditional methods.

[0049] Furthermore, inputting the question text data into the question analysis model includes: determining the question solvability rules, and constructing prompt text data based on the question solvability rules and the question text data; inputting the prompt text data into the question analysis model; obtaining the question completeness information corresponding to the question text data output by the question analysis model includes: performing question completeness analysis on the question text data based on the prompt text data through the question analysis model, and obtaining the question completeness information corresponding to the question text data output by the question analysis model.

[0050] Among them, the question solvability rules can be understood as a set of standard criteria abstracted from business logic to determine whether a question is solvable. These rules are predefined, and include, for example, "checking whether it contains politically sensitive content," "checking whether the reading material is complete," "checking whether it lacks the key information or conditions necessary to solve the question," and "checking whether multiple-choice questions lack options." These rules constitute the specific dimensions and basis for the model's analysis. Constructing prompt text data is a process of formatting and integrating task instructions, analysis rules, and the data to be analyzed. Prompt text data is a carefully designed string that acts as a structured interface between the user (system) and the large language model. Its construction is usually achieved by filling in a predefined text template. A typical prompt text data includes the following layers: Role definition, setting the model's role, such as "You are a strict question integrity review expert." Task and rule description, clearly listing the task objectives and describing the question solvability rules one by one and unambiguously, which is the core of guiding the model's reasoning. Input and output format specifications, specifying the location of the question to be analyzed (usually by replacing the actual question text data with placeholders), and strictly requiring the data format of the model output (e.g., it must be a JSON object containing specific fields).

[0051] In practice, when inputting the question text data into the question analysis model, the question text data can be combined with guiding question solvability rules to generate prompt text data. The question analysis model receives and parses the entire prompt text data. It understands its assigned role, the rules it must follow, the specific content to be analyzed, and the output format it must adhere to. The model will examine, reason, and verify the question text one by one according to each rule explicitly listed in the prompt. Finally, the model uses its rule-based analysis conclusions to generate structured question integrity information according to the mandatory format and outputs it.

[0052] In practical applications, a key challenge when using large language models to determine question completeness is ensuring that the model strictly adheres to specific completeness standards defined by the business side, rather than relying on potentially inaccurate or inconsistent generalized knowledge learned during pre-training. Simply inputting the question text into the model and asking it to make a judgment may result in unstable, uncontrollable outputs that are difficult to correlate with explicit business rules. Therefore, this specification provides a controllable, interpretable model interaction paradigm strongly correlated with business rules. By constructing prompt text data that integrates specific rules, the model's analysis logic is explicitly guided, ensuring that its judgment process and output highly conform to predefined question solvability rules. This precisely constrains the model's powerful capabilities to solving specific business problems.

[0053] In a specific embodiment of this specification, assuming the text data of the question to be judged is: "The lengths of two sides of a triangle are 3 and 4, find the length of the third side.", the system invokes a core solvability rule: "Determine whether the conditions provided in the question are sufficient to solve for all unknowns." Based on the template, the following prompt text data is generated: "

Role

[0054] [Task and Rules] Please strictly follow the following rules to determine whether a problem is complete and solvable: Rule: Check if the problem provides all the known conditions and key information necessary to solve it. The conditions must be sufficient and explicit.

[0055]

Problem

[0056] [Output Requirements] Please output in JSON format, which must include two fields: verdict (value is complete or incomplete) and reason (brief explanation of the reason). After receiving this prompt, the model analyzes the problem according to the rule of "whether the conditions are sufficient". The model outputs structured problem completeness information: "verdict: incomplete, reason: the solution conditions are insufficient. Only two side lengths are given, and the unique length of the third side cannot be determined. Other necessary conditions such as the included angle between the two sides are missing."

[0057] Based on this, by using explicit prompt text to construct the rules, the model's analysis is strictly limited to the preset business rule framework. Its output directly corresponds to specific rule clauses, effectively eliminating the uncertainty and randomness caused by the model's free play, and ensuring a high degree of stability and consistency in the judgment results of similar questions at different times.

[0058] Furthermore, the question integrity information includes integrity analysis information and / or question incompleteness information.

[0059] The question completeness information is a structured data set output by the question analysis model regarding the question's completeness status. As a whole, it encompasses the core conclusions and relevant details of the analysis. Completeness analysis information can be understood as the final, generalized judgment reached by the model after analyzing the question. It is a clear state identifier, usually binary or with a few discrete values, such as "complete" or "incomplete." This information is the direct, unconditional basis for downstream systems to decide how to handle the question. Question incompleteness information can be understood as the specific details provided by the model to explain and describe the incomplete state when the completeness analysis information determines it to be incomplete. This information is directly related to the specific clauses in the "question solvability rules" that have been violated, typically including the category of reasons for incompleteness, such as missing key information, insufficient solution conditions, and / or more detailed descriptive explanations, such as not providing the included angle between the two known sides. This information is mainly used for problem diagnosis, process optimization, and statistical analysis.

[0060] In practical applications, after obtaining the judgment result through the question analysis model, it is necessary to clarify how this result should be organized and output. A simple "yes" or "no" conclusion, while triggering basic processing actions, cannot meet deeper business needs, such as: statistically classifying the judgment result, performing root cause analysis on specific incompleteness types, or providing more explicit feedback to users. Therefore, the purpose of this step is to define the specific composition of the question completeness information, designing it as an information structure containing different logical levels. This ensures that the output not only provides a clear final judgment but also selectively supports specific details supporting that judgment based on the judgment result, thereby achieving operability, analyzability, and interpretability of the judgment result.

[0061] In a specific embodiment of this specification, the input problem text is: "The lengths of two sides of a triangle are 3 and 4. Find the length of the third side." The model outputs problem completeness information including completeness analysis information: "Incomplete"; problem incompleteness information: "The solution conditions are insufficient. Only the lengths of two sides are provided, which cannot uniquely determine the length of the third side. Additional constraints such as the included angle between the two sides and the type of triangle must be provided." In this case, the output contains two complete parts: the final "incomplete" conclusion and the specific reasons for this conclusion.

[0062] In one specific embodiment of this specification, the input problem text is: "The lengths of the two legs of a right triangle are 3 and 4. Find the length of the hypotenuse." The model outputs problem completeness information including completeness analysis information: "Complete". Question Incomplete Information: None or "Not Applicable". In this case, the output mainly contains a "complete" conclusion. Since the question is complete, the "Question Incomplete Information" field can be empty, set to zero, or filled with a value indicating no anomalies.

[0063] Based on this, by clearly dividing the output information structure into "completeness analysis information" that indicates the conclusion and "incomplete information" that provides details, the downstream system can unconditionally rely on the "completeness analysis information," a clear conclusion, to immediately trigger core business actions (such as storing complete information or discarding incomplete information), thus ensuring the simplicity and efficiency of the main process.

[0064] Furthermore, the training process of the question analysis model includes: determining a sample question text set, which includes sample question text data obtained by character recognition of sample question image data; labeling the sample question text data in the sample question text set based on the question solvability rules to obtain the label text data corresponding to the sample question text data; and fine-tuning the pre-trained model using the sample question text data and the label text data as training data to obtain the question analysis model.

[0065] The sample question text set can be understood as the dataset used for model training. Each sample question text simulates a real-world application scenario and originates from the text obtained after character recognition of sample question image data. This ensures that the training data and the actual data that the model will process in the future are consistent in distribution and features, avoiding performance degradation due to differences in data sources. Annotating the sample question text data based on question solvability rules can be understood as reviewing and judging each sample question text according to pre-defined, detailed question solvability rules (such as: "checking for missing key information," "checking for sufficient solution conditions," "checking for complete multiple-choice options," etc.). In practice, this step can be implemented manually or by a large model with annotation capabilities. After annotation, labeled text data corresponding to the sample question text data is obtained, with each labeled text corresponding to a specific sample question text. This labeled data contains at least one integrity status label, such as complete or incomplete. When judged as "incomplete," the label also includes a specific category of incompleteness, such as missing key information or no angle provided, thus providing the model with more refined learning signals. Subsequently, the sample question text data and label text data can be used as training data to fine-tune the model, thereby obtaining the question analysis model.

[0066] In practical applications, a core issue in obtaining a dedicated model capable of accurately judging question completeness is how to effectively instill the abstract solvability rules defined by the business side into a general pre-trained model. Directly using an unadjusted general model or training with randomly labeled data will not enable the model to master this specific judgment logic. Therefore, the purpose of this implementation is to elaborate on the construction method of the question analysis model, that is, how to use a systematic, rule-guided data preparation process to specifically fine-tune a powerful base model, thereby enabling it to acquire superior capabilities in performing the specific task of question completeness analysis, ensuring the accuracy and reliability of the final filtering effect from the source.

[0067] In a specific embodiment of this specification, an OCR is performed on an image containing a math problem to obtain sample problem text data, such as "The lengths of two sides of a triangle are 5 and 7, find the third side." The annotator judges the problem according to the problem's solvability rules, applying the rule "check if the solution conditions are sufficient" for analysis. Since only two sides are known, the third side cannot be uniquely determined, and the conditions are insufficient. The generated label text data is "Completeness Status: Incomplete, Reason Category: Insufficient Solution Conditions, Explanation: Only two side lengths are provided; necessary conditions are missing." During training, the model will see a large number of similar data pairs. By repeatedly learning such correspondences, the model gradually internalizes the rules and learns to make similar judgments for new problems. For example, when it encounters "Given that the two sides of a triangle are 3 and 4, find the third side," it can automatically output the conclusion of "incomplete" and the corresponding reason.

[0068] Based on this, manually summarized and clearly defined "solvability rules" are solidified into high-quality training data through an annotation process, and then fine-tuned to be encoded into the model's parameters. This ensures that the judgment logic of the final question analysis model is highly consistent with business requirements, solving the core problem of uncontrollable and unpredictable model behavior. Because the training data originates from real-world text recognition and covers various types of incompleteness, the model can learn complex patterns and boundary conditions relevant to real-world applications. Compared to training with small models or non-specific data, this method significantly improves the model's judgment accuracy in real-world scenarios and its generalization ability to various new types of questions.

[0069] Step 106: Determine the question processing strategy corresponding to the question image data based on the question integrity information, and process the question image data based on the question processing strategy.

[0070] In this context, question processing strategies can be understood as different predefined automated processing paths based on the conclusions of "completeness analysis information." Different question processing strategies can be applied to question image data in different ways, such as adding or removing data.

[0071] In practical applications, after obtaining accurate completeness judgments through the question analysis model, the system faces a critical decision-making and execution problem: how to translate this judgment into specific, automated business operations to realize the ultimate value of filtering. Without this step, the intelligent judgment will be ineffective, the entire process will be interrupted, and incomplete questions may remain in the system or complete questions may not be effectively included. Therefore, the purpose of this step is to establish a closed-loop link from intelligent judgment to business execution. It is responsible for automatically deciding and triggering corresponding question processing strategies based on structured question completeness information, thereby completing the final routing and processing of question data, ensuring that technical judgments effectively save resources and improve product quality.

[0072] Furthermore, determining the question processing strategy corresponding to the question image data based on the question integrity information includes: if the question text data is determined to be complete based on the question integrity information, determining the question processing strategy corresponding to the question image data as a question entry strategy; if the question text data is determined to be incomplete based on the question integrity information, determining the question processing strategy corresponding to the question image data as a question removal strategy.

[0073] The question inclusion strategy can be understood as the processing strategy determined when the integrity analysis results in a complete question. Specifically, this strategy may involve authorizing and storing all data related to the question (including original images, OCR text, metadata, etc.) into a designated new question-to-process database, marking its status as "pending supplementation," to initiate subsequent answer production processes (such as manual solution, AI generation, etc.). The question rejection strategy can be understood as the processing strategy determined when the integrity analysis results in an incomplete question. Specifically, this strategy may involve rejecting the question data from any valuable data storage or subsequent processes, typically removing it from the processing pipeline and only retaining a log record (recording the reason for incompleteness). Rejection means discarding, aiming to completely release the temporary resources it occupies and prevent any misuse.

[0074] In practical applications, after obtaining the completeness information of a question, the system needs to make a clear and unambiguous decision based on this judgment. If the decision logic is vague or undefined, it will lead to uncertainty in the processing flow. For example, a question judged as incomplete may be improperly retained, or a complete question may be wrongly discarded. Therefore, the purpose of this implementation is to establish a deterministic, rule-mapping-based decision-making mechanism that accurately maps the abstract judgment conclusion of completeness information to specific, executable business operation instructions, thereby ensuring that each analyzed question receives standardized processing that perfectly matches its state.

[0075] In a specific embodiment of this specification, the integrity information output by the model, "Integrity Analysis Information: Complete," determines the question entry strategy. The system will automatically tag the data packet containing the question with "Pending Entry" and route it to the new question storage area.

[0076] In a specific embodiment of this specification, the integrity information output by the model is "Integrity Analysis Information: Incomplete," which determines the processing strategy as a question removal strategy. The system will automatically tag the data packet of this question as "Removed" and remove it from the main processing pipeline, transferring it to the discard or log area.

[0077] Based on this, the system eliminates the need for any complex secondary judgments, issuing final operation instructions solely through simple condition matching. This ensures extremely fast processing speeds and absolute consistency in decision-making results, achieving complete automation of decision-making. Encoding the core business principle of "preserve the complete, discard the incomplete" into immutable system rules allows high-quality judgments to be executed unconditionally and without loss, fundamentally guaranteeing the realization of the technical solution's commercial value.

[0078] Furthermore, processing the question image data based on the question processing strategy includes: when the question processing strategy is a question entry strategy, constructing question data to be entered into the database based on the question image data and the question text data, and storing the question data to be entered into the database in the question database, wherein the question database is used for question answers; when the question processing strategy is a question removal strategy, removing questions from the question image data.

[0079] The "question data to be added" can be understood as the data package that the system needs to integrate from scattered data components into a complete package that can be uniformly processed in subsequent processes when the strategy is to add the questions. The question data to be added can include question image data, question text data, or related metadata, such as source identifiers, processing time, and integrity judgment tags. The question database can be understood as a persistent storage system dedicated to storing and managing questions, specifically serving the question-answering stage. The "removal" process is the operation performed on the question data when the strategy is to remove it. It mainly targets question image data and all its derived data (such as OCR text and intermediate processing results). The system will remove this data from the current processing queue or temporary storage. In practice, before performing removal, the system will record the key removal reasons (from incomplete question information) and question identifiers in the log for subsequent auditing and analysis, but the question data itself will no longer be retained.

[0080] In practical applications, after determining a clear processing strategy (adding to the database or removing data), the system faces the problem of how to specifically execute these strategies. If the strategies are not translated into concrete operations on data entities, then the previous intelligent judgments and decisions will be meaningless. The purpose of this implementation is to clarify the specific execution actions and data processing methods under two different strategies, ensuring that the system can perform the correct physical operations on the question data based on the decision conclusions—either organizing and persistently storing it to create value, or thoroughly cleaning it to release resources, thereby completing the final stage of the entire automated filtering process.

[0081] In a specific embodiment of this specification, the processing strategy has been determined to be a question-input strategy. The system associates the original question image data uploaded by the user with the identified question text data, and may also add metadata such as integrity status, processing stage, and subject category. After constructing the question data to be input into the database, the system stores the data in the question database. Upon successful writing, the question enters a pending-answer state, waiting for human or AI systems to generate answers and explanations.

[0082] In a specific embodiment of this specification, the processing strategy has been determined to be a question elimination strategy. The system locates all temporary data corresponding to the incomplete question, including the uploaded original image, intermediate text generated by OCR, etc. The system permanently deletes this data from the server's temporary storage space or processing queue. At the same time, a message is recorded in the system log: "Question ID [XYZ] has been eliminated, reason: insufficient solution conditions".

[0083] Based on this, for complete questions, standardized construction and persistent storage ensure that they are properly preserved as digital assets and enter the value creation process; for incomplete questions, timely and thorough removal avoids the accumulation of "data garbage" and resource leakage, reflecting refined data governance.

[0084] Furthermore, before determining the question text data corresponding to the question image data, the method further includes: receiving the question image data uploaded by the user, and performing answer matching based on the question image data in the question database; if a question answer data is matched in the question database, displaying the question answer data to the user; if no question answer data is matched in the question database, performing the step of determining the question text data corresponding to the question image data.

[0085] The process of receiving user-uploaded question image data involves the system front-end or interface service acquiring the question image file submitted by the user through the photo-based question search application. To provide the question search service to the user, the system first needs to identify the solution text data from the question image data, and then perform subsequent answer matching steps in the question database based on the solution text data.

[0086] In practice, during the answer matching process, the system does not directly use the original image for comparison. Instead, it typically performs rapid OCR and / or feature vector extraction on the image first, converting the image content into a query request (such as text vectors or image feature vectors) that can be used for retrieval. The system then uses this query request to perform a similarity search in the question database. This database stores a massive number of already included questions and their corresponding standard answers and explanations. The matching algorithm aims to find existing questions that are highly similar to the user-uploaded question in semantics or content. The question answer data refers to the solution content contained in the entries successfully retrieved from the question database that match the user's question. This typically includes the final answer, detailed explanation steps, and possibly explanations of key knowledge points. If the similarity score exceeds a preset threshold, it is determined that "question answer data has been matched." The system then interrupts potentially high-consumption subsequent processes and directly displays the answer data to the user. If no matching item exceeding the threshold is found, it is determined that "question answer data has not been matched." At this point, the system determines the question as a "potential new question," formally triggering a series of subsequent processes aimed at filtering and including new questions, including determining the question text data and integrity checks.

[0087] In practical applications, before initiating the complete question processing flow (including OCR conversion and completeness judgment), the system first needs to address a direct user need: Is there an existing answer to this question? If an answer already exists in the question bank, returning it directly is the most economical and provides the best user experience; otherwise, a more complex processing flow aimed at expanding the question bank resources is initiated. Without this judgment, all question images will undergo indiscriminate deep processing, leading to huge and unnecessary resource consumption (computing power, storage) and delaying the response speed of questions with existing answers. Therefore, the purpose of this implementation is to establish an efficient pre-screening and routing mechanism. It prioritizes meeting the user's immediate needs, and only triggers the completeness judgment and processing flow aimed at "entering the question bank" when these needs cannot be met, thereby achieving optimal allocation of system resources and immediate satisfaction of the user experience.

[0088] In one specific embodiment of this specification, the user uploads an image containing the question, "The two legs of a right triangle are 3 and 4. Find the length of the hypotenuse." The system performs OCR to extract the key text. This text is then used to search the question database. The search results show that an identical question and its answer (hypotenuse length 5) exist in the database. The system does not perform subsequent integrity checks or other steps; instead, it directly displays the calculation process and answer to the user.

[0089] In a specific embodiment of this specification, a user uploads an image containing the question, "Two sides of a triangle have lengths of 3 and 4. Find the length of the third side." The system performs OCR to extract the text. This text is then used to search the question database. The search results show no identical or highly similar questions found (because the question's conditions are incomplete, it is usually not included). Since no matching answer data was found, the system determines this to be an unknown question requiring further processing. Subsequently, the previously described steps of "determining the question text data corresponding to the question image data" and subsequent completeness analysis are formally initiated and executed to determine whether to add it as a new question to the database.

[0090] Based on this, most requests for common questions with existing answers are intercepted in a low-cost, fast retrieval process, avoiding the need to initiate high-cost integrity analysis models and subsequent data entry procedures. This significantly saves overall computing resources and processing time.

[0091] This specification provides a question processing method, comprising: determining the question text data corresponding to question image data, wherein the question text data is text data for which no answer data has been matched; inputting the question text data into a question analysis model to obtain question completeness information corresponding to the question text data output by the question analysis model, wherein the question analysis model is a model based on a pre-trained model, fine-tuned using training data labeled according to question solvability rules; determining a question processing strategy corresponding to the question image data based on the question completeness information, and processing the question image data based on the question processing strategy. This method enables the use of a question analysis model to judge the question completeness of question text data when no answer data has been matched in the question image data, achieving accurate identification of deep semantic deficiencies in the question text and significantly improving the accuracy and recall rate of the judgment. Based on the question integrity information output by the question analysis model, the question processing strategy corresponding to the question image data is determined. By implementing differentiated question processing strategies on the question image data, automated decision-making and resource optimization are achieved. This fundamentally avoids the occupation of subsequent storage and computing resources by invalid data, and also eliminates the risk of displaying incomplete questions to users, thus ensuring user experience and product data quality.

[0092] The following is in conjunction with the appendix Figure 2 Taking the question processing method provided in this manual as an example in the application of photo-based question search, the question processing method will be further explained. Among other things, Figure 2 The present specification shows a flowchart of a problem-solving method according to an embodiment, which includes the following steps.

[0093] Step 202: Receive the question image data uploaded by the user, and perform answer matching in the question database based on the question image data.

[0094] In one embodiment, a user takes a picture of a question and uploads it via an app. The question is: "Given that the perimeter of a square is 20 centimeters, find its side length." After receiving the image, the system immediately uses fast OCR to extract text features and searches for similar questions in a massive question bank.

[0095] Step 204: If no question answer data is found in the question database, determine the question image data, perform character recognition on the question image data, and determine the question text data corresponding to the question image data based on the character recognition results.

[0096] In one embodiment, the question is not in the question bank, and the matching fails. The system performs a formal OCR to obtain accurate question text data: "Given that the perimeter of a square is centimeters, find its side length."

[0097] Step 206: Determine the solvability rules of the question, and construct the prompt text data based on the solvability rules and the question text data. Input the prompt text data into the question analysis model.

[0098] In one embodiment, the system invokes preset rules, such as checking whether the conditions are sufficient, and combines the rules with the question text data obtained in the previous step according to a fixed format to form prompt text data, which is then submitted to the question analysis model. One of the rules is "to determine whether the conditions provided by the question are sufficient to solve for all unknowns." The prompt text data constructed by the system may look like the following: "You are a problem completeness review expert. Please determine whether a problem is complete and solvable according to the rules. Rule: Check whether the problem provides all the known conditions necessary to solve the problem. Problem: Given that the perimeter of a square is centimeters, find its side length. Please output the result in JSON format."

[0099] Step 208: Perform question integrity analysis on the question text data based on the prompt text data using the question analysis model to obtain the question integrity information corresponding to the question text data output by the question analysis model.

[0100] In one embodiment, the question analysis model reads and understands the rules and questions in the prompt text, and then performs logical analysis.

[0101] The problem completeness information includes completeness analysis information and incomplete problem information. The model analysis determined that the specific perimeter value was missing from the conditions, thus classifying it as incomplete. The output problem completeness information is: "Completeness Analysis Information: Incomplete, Reason: Missing key information, no specific perimeter value provided."

[0102] Step 210: If the question text data is determined to be incomplete based on the question integrity information, then the question processing strategy corresponding to the question image data is determined to be the question elimination strategy.

[0103] In one embodiment, the system parses the integrity information output by the model and reads the integrity analysis information field. The value of this field is incomplete. Based on predefined rules, the system automatically determines the processing strategy as a question elimination strategy.

[0104] Step 212: If the question processing strategy is a question elimination strategy, perform elimination processing on the question image data.

[0105] In one embodiment, the system executes a rejection strategy, permanently deleting all temporary data, including the user's initially uploaded question image and the text recognized by OCR. Simultaneously, it logs "Question automatically filtered due to missing key information." Thus, an invalid or incomplete question is successfully blocked from entering the question bank, avoiding resource waste and potential future misleading of users.

[0106] Corresponding to the above method embodiments, this specification also provides embodiments of a question processing device. Figure 3 A schematic diagram of a problem-solving apparatus according to one embodiment of this specification is shown. Figure 3 As shown, the device includes: The determining module 302 is configured to determine the question text data corresponding to the question image data, wherein the question text data is text data for which no question answer data has been matched.

[0107] The input / output module 304 is configured to input the question text data into the question analysis model to obtain the question completeness information corresponding to the question text data output by the question analysis model. The question analysis model is a model based on a pre-trained model, which is fine-tuned using training data labeled according to the question solvability rules.

[0108] The processing module 306 is configured to determine the question processing strategy corresponding to the question image data based on the question integrity information, and to process the question image data based on the question processing strategy.

[0109] Optionally, the determining module 302 is further configured to determine the question image data, perform character recognition on the question image data, and determine the question text data corresponding to the question image data based on the character recognition result.

[0110] Optionally, the input / output module 304 is further configured to determine the solvability rules of the question, and construct prompt text data based on the solvability rules and the question text data; input the prompt text data into the question analysis model; and obtain the question integrity information corresponding to the question text data output by the question analysis model, including: performing question integrity analysis on the question text data based on the prompt text data through the question analysis model to obtain the question integrity information corresponding to the question text data output by the question analysis model.

[0111] Optionally, the input / output module 304 is further configured such that the question integrity information includes integrity analysis information and / or question incompleteness information.

[0112] Optionally, the device further includes a training module configured to determine a sample question text set, the sample question text set including sample question text data obtained by character recognition of sample question image data; to label the sample question text data in the sample question text set based on the question solvability rules to obtain label text data corresponding to the sample question text data; and to fine-tune the pre-trained model using the sample question text data and the label text data as training data to obtain a question analysis model.

[0113] Optionally, the processing module 306 is further configured to, if the question text data is determined to be complete based on the question integrity information, determine the question processing strategy corresponding to the question image data as a question entry strategy; and if the question text data is determined to be incomplete based on the question integrity information, determine the question processing strategy corresponding to the question image data as a question removal strategy.

[0114] Optionally, the processing module 306 is further configured to, when the question processing strategy is a question entry strategy, construct question data to be entered into the database based on the question image data and the question text data, and store the question data to be entered into the database in the question database, wherein the question database is used for question answers; and when the question processing strategy is a question removal strategy, perform removal processing on the question image data.

[0115] Optionally, the device further includes a matching module configured to receive question image data uploaded by a user, and perform answer matching based on the question image data in the question database; if a question answer data is matched in the question database, display the question answer data to the user; if no question answer data is matched in the question database, perform the step of determining the question text data corresponding to the question image data.

[0116] The above is an illustrative scheme of a question processing device according to this embodiment. It should be noted that the technical solution of this question processing device and the technical solution of the question processing method described above belong to the same concept. For details not described in detail in the technical solution of the question processing device, please refer to the description of the technical solution of the question processing method described above.

[0117] Figure 4 A structural block diagram of a computing device 400 according to one embodiment of this specification is shown. The components of the computing device 400 include, but are not limited to, a memory 410 and a processor 420. The processor 420 is connected to the memory 410 via a bus 430, and a database 450 is used to store data.

[0118] The computing device 400 also includes an access device 440, which enables the computing device 400 to communicate via one or more networks 460. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 440 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0119] In one embodiment of this specification, the aforementioned components of the computing device 400 and Figure 4 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 4 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0120] The computing device 400 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 400 can also be a mobile or stationary server.

[0121] The processor 420 is used to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned problem-solving method.

[0122] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above-described problem processing method belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-described problem processing method.

[0123] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described problem-solving method.

[0124] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the above-described problem processing method belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-described problem processing method.

[0125] An embodiment of this specification also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described problem processing method.

[0126] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the above-described problem processing method belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the above-described problem processing method.

[0127] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0128] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0129] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0130] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0131] The preferred embodiments disclosed above are merely illustrative of this specification. Optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described in this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification.

Claims

1. A problem-solving method, characterized in that, include: Determine the question text data corresponding to the question image data, wherein the question text data is text data for which no question answer data was matched; The question text data is input into the question analysis model to obtain the question completeness information corresponding to the question text data output by the question analysis model. The question analysis model is a model based on a pre-trained model, which is fine-tuned using training data labeled according to the question solvability rules. Based on the question integrity information, a question processing strategy corresponding to the question image data is determined, and the question image data is processed based on the question processing strategy.

2. The method according to claim 1, characterized in that, Determine the text data corresponding to the image data of the question, including: Determine the question image data and perform character recognition on the question image data; The question text data corresponding to the question image data is determined based on the character recognition results.

3. The method according to claim 1, characterized in that, Inputting the question text data into the question analysis model includes: Determine the rules for the solvability of the questions, and construct prompt text data based on the rules for the solvability of the questions and the question text data; Input the prompt text data into the question analysis model; Obtain the question completeness information corresponding to the question text data output by the question analysis model, including: The question analysis model performs a question integrity analysis on the question text data based on the prompt text data to obtain the question integrity information corresponding to the question text data output by the question analysis model.

4. The method according to claim 1, characterized in that, The question integrity information includes integrity analysis information and / or question incompleteness information.

5. The method according to claim 1, characterized in that, The training process of the problem analysis model includes: Determine a sample question text set, which includes sample question text data obtained by character recognition of sample question image data; Based on the question solvability rules, the sample question text data in the sample question text set is labeled to obtain the label text data corresponding to the sample question text data; The sample question text data and the label text data are used as training data to fine-tune the pre-trained model, thereby obtaining the question analysis model.

6. The method according to claim 1, characterized in that, Determining the question processing strategy corresponding to the question image data based on the question integrity information includes: If the question text data is determined to be complete based on the question completeness information, the question processing strategy corresponding to the question image data is determined to be the question storage strategy. If the question text data is determined to be incomplete based on the question integrity information, then the question processing strategy corresponding to the question image data is determined to be the question elimination strategy.

7. The method according to claim 6, characterized in that, The question image data is processed based on the aforementioned question processing strategy, including: When the question processing strategy is a question entry strategy, question data to be entered into the database is constructed based on the question image data and the question text data, and the question data to be entered into the database is stored in the question database, wherein the question database is used for question answers; When the question processing strategy is a question elimination strategy, the question image data is eliminated.

8. The method according to claim 7, characterized in that, Before determining the question text data corresponding to the question image data, the following steps are also included: Receive question image data uploaded by users, and perform answer matching based on the question image data in the question database; If a question and answer are matched in the question database, the question and answer data will be displayed to the user. If no matching question answer data is found in the question database, the step of determining the question text data corresponding to the question image data is executed.

9. A problem processing device, characterized in that, include: The determination module is configured to determine the question text data corresponding to the question image data, wherein the question text data is text data for which no question answer data was matched; The input / output module is configured to input the question text data into the question analysis model and obtain the question completeness information corresponding to the question text data output by the question analysis model. The question analysis model is a model based on a pre-trained model and fine-tuned using training data labeled according to the question solvability rules. The processing module is configured to determine the question processing strategy corresponding to the question image data based on the question integrity information, and to process the question image data based on the question processing strategy.

10. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 8.

12. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 8.

Images