A data processing method, apparatus, device, medium, and program product

Abstract

This application provides a data processing method, apparatus, device, medium, and program product to improve the efficiency and accuracy of content classification. It can be applied to fields such as artificial intelligence and content moderation. The method includes: acquiring data to be classified; performing a first classification process on the data to be classified to obtain a first classification result; when the first classification result does not meet preset conditions, acquiring profile data and behavioral data of the interactive object corresponding to the data to be classified, the interactive object including the object that uploaded the data to be classified and the object to which the data to be classified belongs; performing feature encoding processing on the profile data, the behavioral data, and the data to be classified respectively to obtain feature data to be classified; performing a second classification process on the feature data to be classified to obtain a second classification result for the data to be classified; and outputting the second classification result.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more particularly to a data processing method, apparatus, device, medium, and program product. Background Technology

[0002] Currently, internet content is subject to strict control, requiring content creators to connect to a content review system. Content must undergo machine or human review before being published, ensuring its safety and legality. Therefore, content review is crucial for platform development. For example, on social media platforms, the reporting function is a key entry point, vital for detecting and handling potential malicious behavior, and plays a decisive role in maintaining the healthy ecosystem of the social media platform. Accurately categorizing reported content based on evidence provided by social media users is a core task of security strategies.

[0003] Therefore, there is an urgent need for a solution that can improve content categorization. Summary of the Invention

[0004] This application provides a data processing method, apparatus, device, medium, and program product to improve the efficiency and accuracy of content classification.

[0005] In view of this, this application provides a data processing method, comprising: acquiring data to be classified; performing a first classification process on the data to be classified to obtain a first classification result; when the first classification result does not meet preset conditions, acquiring profile data and behavior data of the interactive object corresponding to the data to be classified, wherein the interactive object includes the object that uploaded the data to be classified and the object to which the data to be classified belongs; performing feature encoding processing on the profile data, the behavior data and the data to be classified respectively to obtain feature data to be classified; performing a second classification process on the feature data to be classified to obtain a second classification result of the data to be classified; and outputting the second classification result.

[0006] Another aspect of this application provides a data classification apparatus, including: an acquisition module for acquiring data to be classified;

[0007] The processing module is used to perform a first classification process on the data to be classified in order to obtain the first classification result;

[0008] The acquisition module is used to acquire the profile data and behavior data of the interactive object corresponding to the data to be classified when the first classification result does not meet the preset conditions. The interactive object includes the object that uploaded the data to be classified and the object to which the data to be classified belongs.

[0009] The processing module is used to perform feature encoding processing on the portrait data, the behavior data, and the data to be classified, respectively, to obtain the feature data to be classified; and to perform a second classification processing on the feature data to be classified, to obtain the second classification result of the data to be classified.

[0010] The output module is used to output the second classification result.

[0011] In one possible design, in another implementation of another aspect of the embodiments of this application, the processing module is used to call the first encoder to perform feature encoding on the portrait data to obtain a first feature vector;

[0012] The second encoder is invoked to encode the behavior data to obtain a second feature vector;

[0013] The third encoder is invoked to encode the features of the data to be classified, so as to obtain the third feature vector;

[0014] The first feature vector, the second feature vector, and the third feature vector are used as the feature data to be classified.

[0015] In one possible design, in another implementation of another aspect of the embodiments of this application, the processing module is used to call the first classification model to classify the feature data to be classified in order to obtain the second classification result.

[0016] In one possible design, in another implementation of another aspect of the embodiments of this application, the acquisition module is used to acquire historical classification data and newly added classification data;

[0017] This processing module is used to call the first classification model to classify the historical classification data and the newly added classification data to obtain an evaluation classification result; and to evaluate the first classification model based on the evaluation classification result.

[0018] In one possible design, in another implementation of another aspect of the embodiments of this application, the acquisition module is used to acquire a pre-trained large language model, training samples and prompt words. The training samples include historical profile data, historical behavior data, historical classification data and label data. The input data of the prompt words is the training samples, and the output data of the prompt words is the label data.

[0019] The processing module is used to train the pre-trained large language model based on the training samples and the prompt words to obtain the first classification model.

[0020] In one possible design, in another implementation of another aspect of the embodiments of this application, the acquisition module is used to acquire second reported data, which is historical classification data and historical classification results of the historical classification data, and the historical classification results are secondary label results; and the historical classification data and the historical classification results are used as training samples.

[0021] In one possible design, in another implementation of another aspect of the embodiments of this application, the processing module is used to call a second classification model to classify the data to be classified in order to obtain the first classification result. The second classification model is a text classification model, an image classification model, or a multimodal classification model.

[0022] In one possible design, in another implementation of another aspect of the embodiments of this application, the acquisition module is used to acquire third reported data, which includes key indicator data of the second classification result;

[0023] This processing module is used to send out alarm information when it is determined that there is a problem with the second classification result based on the key indicator data.

[0024] In one possible design, in another implementation of another aspect of the embodiments of this application, the output structure is used to output the first classification result when the first classification result meets the preset conditions.

[0025] This application also provides a computer device, including: a memory, a processor, and a bus system;

[0026] The memory is used to store programs;

[0027] The processor is used to execute programs in memory, and the processor is used to execute the methods mentioned above according to the instructions in the program code;

[0028] Bus systems are used to connect memory and processor to enable communication between them.

[0029] Another aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the methods described above.

[0030] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the above aspects.

[0031] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: In the process of content classification, the content is reviewed by combining profile data and behavioral data, which provides more classification reference data for the content to be classified. This can improve the accuracy and coverage of classification, thereby enabling more sensitive capture of subtle abnormal patterns and providing some interpretability for the classification results. Attached Figure Description

[0032] Figure 1 A schematic diagram of the review process for the security review of the content of this application;

[0033] Figure 2 This is a schematic diagram of an audit process involving both machine review and human review, as illustrated in this application.

[0034] Figure 3 This is a schematic diagram of a process architecture for machine review in an embodiment of this application;

[0035] Figure 4 This is a schematic diagram illustrating an application scenario of the data processing method in the embodiments of this application;

[0036] Figure 5 This is a schematic diagram of one embodiment of the data processing method in this application;

[0037] Figure 6 This is a schematic diagram of a data classification process of an LLM model in an embodiment of this application;

[0038] Figure 7 This is a schematic diagram of one embodiment of the data classification device in this application;

[0039] Figure 8 This is a schematic diagram of one embodiment of the server in this application;

[0040] Figure 9 This is a schematic diagram of one embodiment of the terminal in this application. Detailed Implementation

[0041] This application provides a data processing method, apparatus, device, medium, and program product to improve the efficiency and accuracy of content classification.

[0042] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0043] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0044] Currently, the government employs strict controls over internet content, requiring content creators to connect to a content review system. Content must undergo machine or human review before being published, ensuring its safety and legality. Therefore, content review is crucial for platform development. For example, on social media platforms, the reporting function is a key entry point, vital for detecting and handling potential malicious behavior, and plays a decisive role in maintaining the healthy ecosystem of the platform. Accurately categorizing reported content (such as pornography or fraudulent activity) based on evidence provided by social media users is a core task of security strategies. Currently, different classification schemes are typically used for different types of content.

[0045] Specifically, it can be as follows:

[0046] 1. Text Classification: Natural Language Processing (NLP) techniques are typically used, which can accurately understand and classify text content. For example, word embeddings, such as Word2Vec and GloVe, can provide rich semantic representations of text data. Deep learning models such as Bidirectional Encoder Representations from Transformers (BERT) and Long Short-Term Memory (LSTM) networks can effectively process high-dimensional data generated by word embeddings. Through sentiment analysis and topic modeling, these models can identify malicious language, potential threats, and other illegal information in the text.

[0047] 2. Image Classification Techniques: Image classification often employs computer vision techniques such as Convolutional Neural Networks (CNNs) and image segmentation to accurately identify inappropriate content in images. These methods are effective in detecting obvious violations, such as inappropriate signs or pornographic content. With technological advancements, more advanced models, such as Vision Transformers (ViT) models, utilize self-attention mechanisms to handle more complex visual patterns, enabling them to recognize hidden symbols and subtle images. ViT models, due to their excellent performance in handling long-range image dependencies, are particularly suitable for scenarios requiring deep visual understanding. The application of these techniques expands the capabilities of image classification, improving the accuracy and efficiency of content detection.

[0048] 3. Multimodal Classification Processing Technology: Multimodal models comprehensively analyze video content by integrating image, text, and audio analysis techniques. Compared to text and image classification, multimodal model processing is more complex and costly. Therefore, a common compromise is to extract video frames and apply Automatic Speech Recognition (ASR) technology to convert the content into images and text, simplifying the processing and reducing costs.

[0049] 4. Large Language Model (LLM) Multimodal Approach: LLM leverages its superior computational power to simultaneously process inputs from both image and text modalities, and utilizes generative models for content classification. The specific process involves directly inputting data from multiple modalities into a single model, achieving comprehensive data processing. This method enables the system to more comprehensively understand and analyze mixed data, significantly improving classification accuracy and processing efficiency.

[0050] However, the above four solutions usually use a common scenario to train the model. Therefore, for content security review scenarios, it is not possible to train a suitable model based on common public datasets. Therefore, there is an urgent need for a solution that can improve content classification.

[0051] To address this technical problem, this application provides the following technical solution: acquiring data to be classified; performing a first classification process on the data to be classified to obtain a first classification result; when the first classification result does not meet preset conditions, acquiring the profile data and behavioral data of the interactive object corresponding to the data to be classified, the interactive object including the object that uploaded the data to be classified and the object to which the data to be classified belongs; performing feature encoding processing on the profile data, the behavioral data, and the data to be classified respectively to obtain feature data to be classified; performing a second classification process on the feature data to be classified to obtain a second classification result for the data to be classified; and outputting the second classification result. In this way, by combining profile data and behavioral data to review the content during the content classification process, more classification reference data is provided for the content to be classified, which can improve the accuracy and coverage of classification, thereby enabling more sensitive detection of subtle abnormal patterns and providing some interpretability to the classification results.

[0052] The data processing methods in the various optional embodiments of this application can be implemented based on artificial intelligence (AI) technology. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, pre-trained model technology, operating / interactive systems, and mechatronics. Among these, pre-trained models, also known as large models or foundational models, can be widely applied to downstream tasks across various AI fields after fine-tuning. AI software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0053] Pre-trained models, also known as foundational models or large models, refer to deep neural networks (DNNs) with a large number of parameters. These DNNs are trained on massive amounts of unlabeled data. Leveraging the function approximation capabilities of large-parameter DNNs, Proximity-Based Transformers (PTMs) extract common features from the data. Through fine-tuning, efficient parameter fine-tuning (PEFT), and prompt-tuning techniques, they are suitable for downstream tasks. Therefore, pre-trained models can achieve ideal results in small-shot or zero-shot scenarios. PTMs can be categorized according to the data modality they process, such as language models (ELMO, BERT, GPT), visual models (Swin-transformer, ViT, V-MOE), speech models (VALL-E), and multimodal models (ViBERT, CLIP, Flamingo, Gato). Multimodal models refer to models that establish feature representations for two or more data modalities. Pre-trained models are important tools for outputting AI-generated content (AIGC) and can also serve as a general interface connecting multiple task-specific models.

[0054] Computer vision (CV) is a science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes in tasks such as target recognition, tracking, and measurement, and further performs image processing to create images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multidimensional data. Large model technology has brought significant changes to the development of computer vision technology. Pre-trained models in the vision field, such as Swin-transformer, ViT, V-MOE, and MAE, can be quickly and widely applied to downstream specific tasks after fine-tuning. Computer vision technology typically includes image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping (SLAM), and other technologies, as well as common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0055] This application also relates to cloud technology. Cloud technology refers to a managed technology that unifies hardware, software, network, and other system resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing.

[0056] Cloud technology is a general term encompassing network technology, information technology, integration technology, management platform technology, and application technology applied to the cloud computing business model. It can form resource pools, providing flexible and convenient on-demand access. Cloud computing technology will become a crucial support. Backend services of technical network systems require substantial computing and storage resources, such as video websites, image websites, and many portal websites. With the rapid development and application of internet behavior, every item may possess its own identification mark in the future, requiring transmission to backend systems for logical processing. Data at different levels will be processed separately, and various industry data will require robust system support, which can only be achieved through cloud computing. The cloud technology involved in this application mainly refers to the transmission of images to be identified between terminal devices or servers via the "cloud," etc.

[0057] The technical solutions of this application and their effects are described below through several exemplary embodiments. It should be noted that the following embodiments can be referenced, borrowed from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0058] This application provides a data processing method, apparatus, device, and storage medium to improve the efficiency and accuracy of content classification. The following describes exemplary applications of the electronic devices provided in this application. These electronic devices can be implemented as various types of user terminals or as servers.

[0059] Electronic devices, by running the data processing method provided in the embodiments of this application, improve the efficiency and accuracy of content classification. In other words, they enhance the efficiency and accuracy of content classification within the electronic device.

[0060] The above solution can be applied to many content moderation fields, such as content moderation for posts published on social media platforms and content moderation for reported content. When using the data processing method provided in this application to assist users in content moderation, the method can be implemented as a standalone online application, installed on the user's computer device or backend server, making it convenient for users to use the program for content moderation.

[0061] In this scenario, the social media platform's backend server receives user-uploaded data to be categorized. This data is then input into a content moderation model or a finely tuned LLM model to classify the data and obtain a secondary classification result. The following explanation uses a specific scenario of report moderation as an example; the process can be described as follows: Figure 1 As shown, it is mainly divided into three parts: the user side, the review process, and supporting services.

[0062] The user-side process includes initiating a complaint, selecting the complaint type, checking the boxes for evidence (which can be in the form of text, images, or videos), uploading additional evidence (which can also be in the form of text, images, or videos), and submitting the complaint. It should be understood that the above complaint process is merely an example and is not limited to any specific implementation, as long as the submission of evidence can be completed.

[0063] The review process and supporting services can be as follows: First, a screening logic determines whether the reported evidence should enter the machine review or human review process. If it enters human review, the human review result is directly used as the review result. If it enters machine review, the reported evidence is classified by the model to obtain the machine review result; then the machine review result can be automatically processed and analyzed; if automatically processed, it can be directly output as the review result. It should be understood that, to better ensure the review result, the review result can be subject to secondary sampling. Simultaneously, the review result can be uploaded, and actions can be taken against the user indicated by the reported evidence based on the review result, and the review effectiveness can be evaluated to obtain the offline evaluation result of the machine review model. On the other hand, the review result can also be used for training sample adaptation to update and train the machine review model; then the model is evaluated offline. When the offline evaluation result meets the requirements, the model is deployed to provide machine review services.

[0064] In the review process, the combination of machine review and human review is particularly important. For example... Figure 2 As shown, the results of machine review serve as preliminary judgments, providing a "reference answer" for manual review, thereby accelerating the identification and processing of suspected violations by reviewers. Simultaneously, based on the machine review results, samples judged as malicious by the machine can be prioritized, thus increasing the detection rate of malicious content in manual review and effectively avoiding wasting manpower on large amounts of normal data. For example, data sent for manual review can be sorted according to the machine review results, allowing malicious samples to be reviewed more quickly.

[0065] Furthermore, the machine review model is constantly iterating and optimizing, along with standardized data annotation processes, to ensure data quality and consistency, thereby significantly improving the effectiveness and efficiency of data processing.

[0066] As described above, automated review is a crucial part of the entire content review process. Therefore, this application can optimize this automated review process when it is necessary to improve the efficiency and accuracy of content review. The following is an example diagram of an automated review architecture to illustrate the data processing method of this application. Figure 3 As shown.

[0067] In the machine-based review architecture, it can be divided into left and right branches. The right branch is used to classify the data to be classified first using a common classification model to obtain the corresponding classification result. The left branch combines user profile data and behavioral data to perform a second classification using an LLM model. If the classification result meets the preset confidence probability requirements, the classification result of the common model is directly used as the machine review result. If the classification of the data to be classified is complex and the classification result does not meet the preset confidence probability requirements, then the user's profile data and behavioral data are combined and input into the LLM model. This design aims to save costs: for simple data to be classified, a basic model is used; for more complex cases, more profile data and behavioral data are introduced, and an LLM model with a larger number of parameters is used to improve the accuracy and recall of the classification.

[0068] Of course, in addition to the above-mentioned scenarios, the method provided in this application embodiment can also be applied to other scenarios that require content review. This application embodiment does not limit the specific application scenarios.

[0069] Based on the aforementioned technical principles or related theoretical foundations, the data processing method proposed in this application can be applied to any computer device capable of data classification and calculation, and this computer device can be various types of terminals or servers. When the computer device in the embodiment is a server, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud service content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0070] It should be further noted that the terminals involved in the embodiments of this application include, but are not limited to, smartphones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, and aircraft. The embodiments of this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0071] In some possible implementations, embodiments of the present invention provide a computer program for a data processing method that can be deployed and executed on a single computer device, or on multiple computer devices located in one location; or, on multiple computer devices distributed across multiple locations and interconnected via a communication network, wherein the multiple computer devices distributed across multiple locations and interconnected via a communication network can form a blockchain system.

[0072] Since multiple computer devices can form a blockchain system, the data processing method provided in this application can be executed and completed by a node in the blockchain; and the node used to execute the data processing method can be any mobile terminal that can provide a front-end page, such as a smartphone, tablet computer, or personal computer (PC).

[0073] The embodiments of this application can be applied to scenarios such as artificial intelligence, content moderation, and data security. The following examples illustrate this. Figure 4 An exemplary architecture is illustrated below, comprising a terminal 100, a server 200, a database 300, and a network 400. The terminal 100, the server 200, and the database 300 are connected via the network 400. The database 300 is used to store historical classification data, historical classification results, and various models. Figure 4 The number of terminals 100, servers 200 and databases 300 in the system shown is only an example. For example, there may be multiple terminals 100, servers 200 and databases 300. This application does not limit the number of terminals 100, servers 200 and databases 300.

[0074] In this system, terminal 100 communicates with server 200 via a network. Database 300 can be integrated onto server 200 or hosted on the cloud or other servers. During data processing, interaction can occur between terminal 100 and server 200. For example, a user uploads data to be classified, profile data, and behavioral data through terminal 100. Terminal 100 then sends these data to server 200, which classifies the data based on the data to be classified, profile data, and behavioral data to obtain a second classification result.

[0075] Terminal 100 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, smart voice interaction device, smart home appliance, or in-vehicle terminal, but is not limited to these. Server 200 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN), big data, and artificial intelligence platforms.

[0076] In short, a database can be viewed as an electronic filing cabinet—a place to store electronic files, where users can perform operations such as adding, querying, updating, and deleting data. A "database" is a collection of data stored together in a certain way, shared by multiple users, with minimal redundancy, and independent of application programs. A Database Management System (DBMS) is a computer software system designed to manage databases, generally possessing basic functions such as storage, retrieval, security, and backup. DBMSs can be classified according to the database model they support, such as relational or Extensible Markup Language (XML); or according to the type of computer they support, such as server clusters or mobile phones; or according to the query language used, such as Structured Query Language (SQL) or XQuery; or according to performance priorities, such as maximum scale or maximum operating speed; or other classification methods. Regardless of the classification method used, some DBMSs can cross categories, for example, supporting multiple query languages ​​simultaneously.

[0077] It is understood that in the specific implementation of this application, data such as model data, profile data, behavioral data and data to be classified are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0078] Given that this application involves some technical terms, these terms will be introduced below.

[0079] Classification: In machine learning and deep learning, the classification problem refers to predicting the category to which input data belongs based on its features. This typically involves training an algorithm or model to identify and assign new input instances from a set of known categories. Classification is a type of supervised learning and is widely used in various scenarios such as image recognition, speech recognition, and text classification.

[0080] LLM: Large Language Model, refers to a machine learning model trained with a large amount of data and possessing extensive language processing capabilities.

[0081] Modality: In machine learning and multimedia processing, a modality refers to different forms of data or communication channels. Common data modalities include text, images, and speech. These modalities can be processed individually or in combination in multimodal learning scenarios to extract richer information and improve the accuracy of decision-making.

[0082] Fine-tuning: In machine learning, this specifically refers to the process of retraining a pre-trained model to better adapt it to a specific task or dataset.

[0083] Prompt: Guiding text input to the model when using LLM for a task, used to instruct the model to generate specific outputs.

[0084] Self-attention mechanism: Self-attention is a widely used mechanism in deep learning, especially in the field of natural language processing (NLP). It is a method that can compute the interaction relationships between different parts of a sequence, allowing the model to weight information from different parts of the sequence.

[0085] Fully connected layer (Multilayer Perceptron, MLP): A fully connected layer is typically a layer in a multilayer perceptron that contains a set of neurons, where each neuron is connected to all neurons in the previous layer. The main function of a fully connected layer is to map nonlinear combinations of input data, thereby allowing the network to learn more complex data patterns and relationships.

[0086] Based on the above introduction, the data processing method in this application will be described below with the server as the execution entity. Please refer to [link / reference]. Figure 5 One embodiment of the data processing method in this application includes:

[0087] 501. Obtain the data to be classified.

[0088] In this embodiment, the data to be classified can be content uploaded and published by users on an internet platform, or it can be content reported on an internet platform. That is, the data to be classified can be data actively collected by the internet platform or data sent by a third party; the specific form is not limited here. Furthermore, based on the data carrying format on the internet platform, the data to be classified can be in the form of text data, image data, video data, and audio data; the specific form is not limited here.

[0089] The following is a specific application scenario to illustrate the data to be categorized: Suppose that in the XX software, user A feels that an article published by a public account is detrimental to them. User A can initiate the complaint process by selecting the "Complaint" button in the XX software's function keys. Then, for the infringement of the public account's article content, such as defamation, plagiarism, or paraphrasing, the appropriate complaint type can be selected to report it. For example, if the public account's article defames user A's company, user A needs to upload relevant evidence from their computer and file a complaint against the public account's article content. In this case, the public account's article and user A's complaint reasons can be used as the data to be reviewed. If user A's article is plagiarized or paraphrased, user A can click "Complaint" on their mobile phone - "Unauthorized Content / Abuse of Originality" - "Plagiarism / Paraphrasing" - fill in the materials and submit the report. In this case, the materials uploaded by user A will be used as the data to be reviewed.

[0090] 502. Perform the first classification process on the data to be classified to obtain the first classification result.

[0091] In this embodiment, the server inputs the data to be classified into the machine review process according to the screening logic, and determines the data type of the data to be classified, that is, whether the data to be classified is text data, image data, video data or voice data; then the server calls the corresponding classification model to classify the data to be classified in order to obtain the corresponding first classification result.

[0092] In this embodiment, the text data can call a text classification model, the image data can call an image classification model, and the video data or the audio data can call a multimodal model.

[0093] It should be understood that although the text classification model, image classification model, and multimodal model have broad application potential, their training objectives and the distribution of pre-training data differ significantly from those of the target model. Therefore, this embodiment requires corresponding fine-tuning training of the text classification model, image classification model, and multimodal model. The specific implementation process is as follows: Obtain training samples and an initial classification model, wherein the model architecture of the initial classification model is the same as that of the target classification model. For example, if a text classification model needs to be trained, the model architecture of the initial classification model is the same as that of the text classification model. Finally, train the initial classification model based on the training samples to obtain the target classification model.

[0094] To better align with the current training objectives, the training sample can be constructed as follows: user identifier, upload timestamp of the categorized data, categorized data, and categorized data labels. For example, when constructing a training sample from a report by user A, its structure could be as follows: (user A's identifier, 2024-10-11, article A from the public account and user A's complaint content, plagiarism).

[0095] The specific approach for data annotation can be as follows:

[0096] 1. Align the data with the classification rules and the model's classification categories to improve the accuracy of training sample data labeling.

[0097] 2. Conduct a second round of random checks on the training samples to ensure their quality.

[0098] In the training process of the aforementioned text classification model, image classification model, and multimodal classification model, the following scheme can be adopted to further ensure the quality of model training:

[0099] 1. Increase the number of training samples. In practical applications, the model's performance gradually improves as the amount of training data increases (from tens of thousands to hundreds of thousands, and then to millions). However, when the amount of data reaches millions, further increasing the amount of data no longer significantly improves online performance, while the time cost of training and iteration increases significantly. Therefore, the number of training samples should be increased based on the actual results.

[0100] 2. Adjust the ratio of positive to negative samples in the training samples. When faced with imbalanced samples in the scenario (for example, in fraud classification, the ratio with other categories may be as high as 1:1000), in practical applications, controlling the ratio of samples of each category between 1:10 and 1:15 can achieve the best results.

[0101] 3. Adjust the method of selecting negative samples. For negative samples that are not of interest to the classification, random sampling can achieve the best results in practical applications.

[0102] 4. Adjust the weight values ​​of training samples during the training process. For example, if a training sample is determined to be of high quality after two rounds of sampling, its weight coefficient can be increased in the loss function calculation.

[0103] The following describes a specific model fine-tuning training process. Assume the text classification model is based on Roberta's model architecture, with a fully connected layer trained independently as the classification head. During training, training samples can be input into the text classification model. The fully connected layer then generates predicted classification results for the training samples. These predicted classification results are then compared with the classification labels to calculate a loss value. This loss value is then used for back-training to obtain the final text classification model.

[0104] The reason for choosing the Roberta model is that its pre-training phase employs a whole word masking (WWM) strategy, which is particularly effective for Chinese semantic understanding. Simply put, this word-piece-based segmentation method typically divides a complete word into multiple sub-words. When generating training samples, if these sub-words are randomly selected for masking, other sub-words belonging to the same word will also be masked, thus achieving whole-word masking. An example scheme is shown in Table 1.

[0105] Table 1

[0106]

[0107] After the above operations, the text classification model can enhance its ability to capture semantic integrity, which is particularly helpful for processing Chinese text.

[0108] Similarly, the training process for image classification models and multimodal models can also follow the above procedure, which will not be elaborated here.

[0109] 503. Determine whether the first classification result meets the preset conditions. If it does, proceed to step 508; otherwise, proceed to step 504.

[0110] In this embodiment, after obtaining the first classification result, the server determines whether the confidence probability of the first classification result meets a preset condition, and then determines subsequent operations based on the confidence probability. The preset condition can be set as follows: if the confidence probability of the first classification result is greater than a first threshold, the first classification result is directly output; if the confidence probability of the first classification result is less than or equal to the first threshold, the data to be classified is reclassified by combining the profile data and behavioral data.

[0111] 504. Obtain the profile data and behavior data of the interactive object corresponding to the data to be classified. The interactive object includes the object that uploaded the data to be classified and the object to which the data to be classified belongs.

[0112] When obtaining the categorized data, the server can also obtain the upload object and the profile and behavioral data of the object to which the data to be categorized belongs.

[0113] This profile data is generated by collecting and analyzing multi-dimensional data on individuals, such as social attributes, consumption habits, and preferences. This data is then used to characterize the individual or product's attributes, analyze and statistically analyze these characteristics, and uncover potential value information, thereby abstracting a complete picture of the individual or product. For example, for interactive objects on social platforms, profile data can be constructed by collecting data on the interactive object's gender, age, education, occupation, residence, interests, and browsing habits. For instance, an interactive object's profile might look like this: a male around 25 years old, with a bachelor's degree, an internet engineer, residing in a first- or second-tier city, frequently interested in digital products, and enjoys playing games.

[0114] Behavioral data refers to data that records and analyzes the behavioral patterns and characteristics of individuals or groups in a specific environment. This data typically originates from interactions between the interacting object and products, services, or the environment, and can be used to understand the interacting object's needs, preferences, and decision-making processes. The collection and analysis of behavioral data is crucial for personalized recommendations, optimizing the user experience, and predicting market trends. For example, behavioral data for online interactions can be clickstream data, including pages visited, links clicked, dwell time, and scrolling behavior. For instance, an e-commerce website might record the duration of a user's browsing of product pages, items added to their cart, items ultimately purchased, and keywords entered in the search box.

[0115] 505. Perform feature encoding processing on the portrait data, the behavior data, and the data to be classified, respectively, to obtain the feature data to be classified.

[0116] In this embodiment, the server uses different encoders to encode and map the portrait data, the behavior data, and the data to be classified, respectively, to obtain the input data for the second classification model. Specifically, the server calls the first encoder to encode the portrait data to obtain the first feature vector; calls the second encoder to encode the behavior data to obtain the second feature vector; and calls the third encoder to encode the data to be classified to obtain the third feature vector. The first, second, and third feature vectors are then used as the input data for the second classification model, i.e., the feature data to be classified.

[0117] 506. Perform a second classification process on the feature data to be classified to obtain the second classification result of the data to be classified.

[0118] In this embodiment, the server can call the second classification model to perform a second classification process on the feature data to be classified, so as to obtain the second classification result of the data to be classified.

[0119] In this embodiment, the second classification model can be a model fine-tuned based on an LLM model. The fine-tuning methods for the LLM model can include LoRa and P-Tuning, among others. The core idea of ​​LoRa is to achieve fine-tuning by updating the LLM model in low rank, rather than completely retraining all parameters. This method not only significantly reduces the necessary computational resources and time but also maintains the model's stability and generalization ability. A simplified formula is as follows:

[0120] h = W0x + ΔWx = W0x + BAx

[0121]

[0122] r << min(d,k)

[0123] Here, W0 indicates the model's weight parameters, ΔWx indicates the model's change, and B and A are matrices. When fine-tuning the model using the LoRa method described above, it primarily involves introducing additional small-scale matrices (A and B in the formula) into the model's self-attention layer to adjust existing weights rather than replace them. This allows for adaptation to a specific task without sacrificing the language understanding capabilities accumulated during pre-training. This makes LoRa highly suitable for LLM models with a large number of parameters, effectively balancing tuning accuracy and model efficiency.

[0124] It should be understood that adjusting the sample format and standardizing the Prompt directive are key steps in fine-tuning a Large Language Model (LLM), which helps to standardize the output of the generative model. In an exemplary scheme, the Prompt format can be configured as follows: Prompt word + content to be categorized + tag. For example, a specific example of this Prompt could be: You are a content security reviewer. Please determine the specific category tag in the content to be categorized based on the following: <|sep|>…<|sep|>. The specific category tag is: fraudulent order placement.

[0125] In the classification process of the LLM model that combines image data and behavioral data, the Prompt format also needs to be adjusted accordingly, and it can be configured as follows: Prompt + content to be classified + image data + behavioral data + label. For example, a specific example of this Prompt can be as follows: You are a content security auditor. Please based on the following content: <|sep|>Content to be classified: …… <|sep|>Behavioral data: …… <|sep|>Image data: …… <|sep|>Judge the specific classification label in the content to be classified. The specific classification label is: brush orders. In an exemplary solution, the content to be classified can be "Report that the content published by user A is brush orders, and the evidence includes text content 1, text content 2, and screenshot image 1". The image data can be that "User A (the reported user): A male in his mid-20s, with a bachelor's degree, an Internet engineer, who has long settled in first- and second-tier cities, often pays attention to digital products, likes to play games, has 70 friends on the current social platform, belongs to 10 groups on the current social platform, and shops on this social platform 3 times a day... User B (the reporter): A male in his late 20s, with a junior college degree, a salesperson, who settled in City A, often pays attention to financial data, has 80 friends on the current social platform, belongs to 30 groups on the current social platform, and shops on this social platform 5 times a day...". The behavioral data can be that "User A (the reported user): Start the social software - scan the QR code - enter a new group - send a link - send a red envelope - share the group... User B (the reporter): Start the social software - scan the QR code - enter a new group - speak - take a screenshot...".

[0126] In order to ensure that the behavioral data, image data, and data to be classified can all be adapted to this LLM model, when this LLM model processes the above data, it can also select different encoders to encode different data to obtain input data that conforms to this LLM model. In an exemplary solution, as Figure 6 shown in the architecture of the LLM model, where the first encoder is called to perform feature encoding on the image data to obtain the first feature vector; the second encoder is called to perform feature encoding on the behavioral data to obtain the second feature vector; the third encoder is called to perform feature encoding on the data to be classified to obtain the third feature vector; then the first feature vector, the second feature vector, and the third feature vector are used as the input data and the Prompt is input into this LLM model to obtain the final second classification result.

[0127] In an exemplary scheme, during online applications and offline model training, the behavioral data, profile data, and data to be classified can be extracted using different encoders depending on their format. For example, if the behavioral data is text data, then when using an encoder for feature extraction, the text data can first be preprocessed and cleaned to obtain valid data; then, the valid data can be serialized to obtain a behavioral sequence. Sequence processing of text data typically involves converting the data structure or object state into a storable or transmittable format. This process allows data to be saved to files, memory buffers, or transmitted over a network to other computing environments. Serialization forms include binary serialization and text serialization. Binary serialization converts data into a compact binary representation, often used in performance-sensitive systems or low-bandwidth network communications. Text serialization converts data into text formats such as XML, JSON, and YAML, offering good readability, ease of debugging, and suitability for Web APIs and configuration files. Then, feature extraction is performed on the behavioral sequence to obtain the corresponding feature vector. Similarly, when the profile data and data to be classified are text data, the above scheme can also be used for feature extraction. When the data to be classified is image data, the encoder can extract features from the image data to obtain feature vectors. In practical applications, feature vectors in the semantic space of the LLM model usually have some special specifications and characteristics. Therefore, feature mapping is required for the feature vectors extracted by the various encoders to obtain feature vectors that the LLM model can process.

[0128] It should be understood that the first encoder, the second encoder, and the third encoder can be trained individually or jointly with the LLM during training. Specific details are not limited here.

[0129] The LLM model trained using the above method shows a significant improvement in performance in practical applications, as shown in Table 2.

[0130] Table 2

[0131]

[0132] 507. Output the result of the second classification.

[0133] The server directly outputs the second classification result, and then labels the data to be classified based on the second classification result.

[0134] 508. Output the first classification result.

[0135] The server directly outputs the first classification result, and then labels the data to be classified based on the first classification result.

[0136] In this embodiment, the server can also report various data during the entire machine review process, thereby realizing data monitoring of the machine review process.

[0137] In one exemplary scheme, the server reports the classification result of the data to be classified and uses the data to be classified and its classification result as training samples for subsequent model updates.

[0138] In another exemplary scheme, the server reports the classification result of the data to be classified, and then performs statistical analysis on the data reported periodically and the classification results to obtain the key indicator data of the machine review model. Then, when it is determined that there is a problem with the second classification result based on the key indicator data, the server sends out alarm information so that the operation and maintenance personnel can optimize the machine review model based on the alarm information.

[0139] In another exemplary scheme, the server reports the classification result of the data to be classified. Then, before the machine review model is updated, the server runs the data to be classified and the newly added data on the machine review model to obtain the evaluation classification result. Based on the evaluation classification result, the machine review model is evaluated to obtain the evaluation result. Based on the evaluation result, it is determined whether the machine review model should be updated and deployed.

[0140] The data classification device in this application is described in detail below. Please refer to [link / reference]. Figure 7 , Figure 7 This is a schematic diagram of one embodiment of the data classification device in this application. The data classification device 20 includes:

[0141] Module 201 is used to acquire data to be classified.

[0142] The processing module 202 is used to perform a first classification process on the data to be classified in order to obtain a first classification result;

[0143] The acquisition module 201 is used to acquire the profile data and behavior data of the interactive object corresponding to the data to be classified when the first classification result does not meet the preset conditions. The interactive object includes the object that uploaded the data to be classified and the object to which the data to be classified belongs.

[0144] The processing module 202 is used to perform feature encoding processing on the portrait data, the behavior data, and the data to be classified, respectively, to obtain the feature data to be classified; and to perform a second classification processing on the feature data to be classified, to obtain the second classification result of the data to be classified.

[0145] Output module 203 is used to output the second classification result.

[0146] This application provides a data classification device. Using this device, during the content classification process, both profile data and behavioral data are combined to review the content, providing more classification reference data for the content to be classified. This improves the accuracy and coverage of the classification, enabling more sensitive detection of subtle abnormal patterns and providing interpretability to the classification results.

[0147] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application,

[0148] The processing module 202 is used to call the first encoder to perform feature encoding on the portrait data to obtain a first feature vector;

[0149] The second encoder is invoked to encode the behavior data to obtain a second feature vector;

[0150] The third encoder is invoked to encode the features of the data to be classified, so as to obtain the third feature vector;

[0151] The first feature vector, the second feature vector, and the third feature vector are used as the feature data to be classified.

[0152] This application provides a data classification device. Using this device, different encoders are invoked to encode features of different data, thereby obtaining corresponding feature vectors. This maps the original variable-length different sequence modes to fixed-length feature vectors, thus improving the model's processing efficiency.

[0153] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application, the processing module 202 is used to call the first classification model to classify the feature data to be classified in order to obtain the second classification result.

[0154] This application provides a data classification device. Using this device, the feature data to be classified is processed using a model, enabling automatic machine classification and thus avoiding manual review, thereby improving the accuracy and efficiency of content review.

[0155] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application,

[0156] The acquisition module 201 is used to acquire historical classification data and newly added classification data;

[0157] The processing module 202 is used to call the first classification model to classify the historical classification data and the newly added classification data to obtain the evaluation classification result; and to evaluate the first classification model based on the evaluation classification result.

[0158] This application provides a data classification device. Using this device, before deploying a first classification model online, the first classification model is run on historical data and newly collected data to achieve a comprehensive evaluation of the first classification model. This allows for prediction of the model's actual performance and ensures that the model's deployment will bring the expected improvement.

[0159] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application,

[0160] The acquisition module 201 is used to acquire a pre-trained large language model, training samples, and prompt words. The training samples include historical profile data, historical behavior data, historical classification data, and label data. The input data of the prompt words is the training samples, and the output data of the prompt words is the label data.

[0161] The processing module 202 is used to train the pre-trained large language model based on the training samples and the prompt words to obtain the first classification model.

[0162] This application provides a data classification device. Using this device, historical profile data, historical behavior data, historical classification data, and tag data are used as training samples for the first classification model. By combining profile data and behavior data to review content, more classification reference data is provided for the content to be classified, thus improving the accuracy and coverage of classification. This allows for more sensitive detection of subtle abnormal patterns and provides interpretability to the classification results. Simultaneously, the prompt words for the large language model are also adjusted accordingly based on the training samples, further optimizing the model.

[0163] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application, the acquisition module 201 is used to acquire second reported data, which is historical classification data and historical classification results of the historical classification data, and the historical classification results are secondary label results; and the historical classification data and the historical classification results are used as training samples.

[0164] This application provides a data classification device. Using this device, the second classification result is automatically reported, thereby improving the speed of data flow and significantly reducing the risk of human error. Simultaneously, the classification labels are defined more precisely, reducing confusion during model training and improving the reliability and consistency of the classification results.

[0165] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application, the processing module 202 is used to call a second classification model to classify the data to be classified in order to obtain the first classification result. The second classification model is a text classification model, an image classification model, or a multimodal classification model.

[0166] This application provides a data classification device. Using this device, a model is employed to classify the feature data to be classified, enabling automatic machine classification and thus avoiding manual review, thereby improving the accuracy and efficiency of content review. Furthermore, different models are used to classify different data to further enhance the accuracy of content review.

[0167] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application,

[0168] The acquisition module 201 is used to acquire third reported data, which includes key indicator data of the second classification results;

[0169] The processing module 202 is used to provide alarm information when it is determined that there is a problem with the second classification result based on the key indicator data.

[0170] This application provides a data classification device. Using this device, key indicator data of model operation are collected and analyzed, and data mining and trend analysis are performed based on the key indicator data to identify the model's strengths and weaknesses, thereby enabling targeted optimization and ultimately improving the model's performance.

[0171] Optionally, in the above Figure 7 Based on the corresponding embodiments, in another embodiment of the data classification device 20 provided in this application,

[0172] The output module 203 is used to output the first classification result when the first classification result meets the preset conditions.

[0173] This application provides a data classification device. Using this device, basic classification processing can be applied to simple review content, thus reducing the amount of data computation.

[0174] The data classification device provided in this application can be used on a server; please refer to [link / reference]. Figure 8 , Figure 8 This is a schematic diagram of a server structure provided in an embodiment of this application. The server 300 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 322 (e.g., one or more processors) and memory 332, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 342 or data 344. The memory 332 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the server. Furthermore, the CPU 322 may be configured to communicate with the storage media 330 and execute the series of instruction operations stored in the storage media 330 on the server 300.

[0175] Server 300 may also include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input / output interfaces 358, and / or one or more operating systems 341, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0176] The steps performed by the server in the above embodiments can be based on this Figure 8 The server structure shown.

[0177] The data classification device provided in this application can be used in terminal devices. Please refer to [link / reference]. Figure 9 For ease of explanation, only the parts relevant to the embodiments of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of this application. In the embodiments of this application, a smartphone is used as an example for illustration:

[0178] Figure 9 This is a block diagram illustrating a portion of the structure of a smartphone related to the terminal device provided in the embodiments of this application. (Reference) Figure 9The smartphone includes components such as a radio frequency (RF) circuit 410, a memory 420, an input unit 430, a display unit 440, a sensor 450, an audio circuit 460, a wireless fidelity (WiFi) module 470, a processor 480, and a power supply 490. Those skilled in the art will understand that... Figure 9 The smartphone structure shown does not constitute a limitation on smartphones and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0179] The following is combined with Figure 9 A detailed introduction to the various components of a smartphone:

[0180] RF circuit 410 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and processes it with processor 480; additionally, it transmits uplink data to the base station. Typically, RF circuit 410 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), a duplexer, etc. Furthermore, RF circuit 410 can also communicate wirelessly with networks and other devices. The aforementioned wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile Communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Message Service (SMS), etc.

[0181] The memory 420 can be used to store software programs and modules. The processor 480 executes various functions and data processing of the smartphone by running the software programs and modules stored in the memory 420. The memory 420 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the smartphone (such as audio data, phonebook, etc.). In addition, the memory 420 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0182] The input unit 430 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the smartphone. Specifically, the input unit 430 may include a touch panel 431 and other input devices 432. The touch panel 431, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 431), and drive the corresponding connected devices according to a pre-set program. Optionally, the touch panel 431 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 480, and can also receive and execute commands sent by the processor 480. In addition, the touch panel 431 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 431, the input unit 430 may also include other input devices 432. Specifically, other input devices 432 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0183] Display unit 440 can be used to display information input by the user or information provided to the user, as well as various menus of the smartphone. Display unit 440 may include display panel 441, optionally configured as a liquid crystal display (LCD), organic light-emitting diode (OLED), or similar form. Further, touch panel 431 may cover display panel 441. When touch panel 431 detects a touch operation on or near it, it transmits the information to processor 480 to determine the type of touch event. Subsequently, processor 480 provides corresponding visual output on display panel 441 based on the type of touch event. Although in Figure 9 In this embodiment, the touch panel 431 and the display panel 441 are two separate components to realize the input and output functions of the smartphone. However, in some embodiments, the touch panel 431 and the display panel 441 can be integrated to realize the input and output functions of the smartphone.

[0184] The smartphone may also include at least one sensor 450, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor can adjust the brightness of the display panel 441 according to the ambient light level, and the proximity sensor can turn off the display panel 441 and / or the backlight when the smartphone is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that recognize the smartphone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, tapping), etc. Other sensors that may be configured in the smartphone, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0185] Audio circuit 460, speaker 461, and microphone 462 provide an audio interface between the user and the smartphone. Audio circuit 460 converts received audio data into electrical signals and transmits them to speaker 461, where speaker 461 converts them into sound signals for output. On the other hand, microphone 462 converts collected sound signals into electrical signals, which are received by audio circuit 460, converted into audio data, and then processed by processor 480 before being transmitted via RF circuit 410 to, for example, another smartphone, or the audio data can be output to memory 420 for further processing.

[0186] WiFi is a short-range wireless transmission technology. Smartphones, through their WiFi modules (470), can help users send and receive emails, browse web pages, and access streaming media, providing wireless broadband internet access. Although Figure 9 WiFi module 470 is shown, but it is understood that it is not an essential component of a smartphone and can be omitted as needed without changing the nature of the invention.

[0187] The processor 480 is the control center of the smartphone, connecting various parts of the smartphone through various interfaces and lines. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 420, and by calling data stored in the memory 420, thereby providing overall monitoring of the smartphone. Optionally, the processor 480 may include one or more processing units; optionally, the processor 480 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the aforementioned modem processor may also not be integrated into the processor 480.

[0188] The smartphone also includes a power supply 490 (such as a battery) that supplies power to various components. Optionally, the power supply can be logically connected to the processor 480 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system.

[0189] Although not shown, smartphones may also include a camera, Bluetooth module, etc., which will not be described in detail here.

[0190] The steps performed by the terminal device in the above embodiments can be based on this Figure 9 The terminal device structure is shown.

[0191] This application also provides a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the methods described in the foregoing embodiments.

[0192] This application also provides a computer program product including a program, which, when run on a computer, causes the computer to perform the methods described in the foregoing embodiments.

[0193] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0194] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0195] The unit described as a separate component may or may not be physically separate. The component shown as a unit may or may not be a physical unit; that is, it may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0197] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0198] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A data processing method, characterized in that, include: Obtain the data to be classified; The data to be classified is subjected to a first classification process to obtain a first classification result; When the first classification result does not meet the preset conditions, the profile data and behavior data of the interactive object corresponding to the data to be classified are obtained. The interactive object includes the object that uploaded the data to be classified and the object to which the data to be classified belongs. The portrait data, the behavior data, and the data to be classified are respectively subjected to feature encoding processing to obtain the feature data to be classified; The feature data to be classified is subjected to a second classification process to obtain the second classification result of the data to be classified. Output the second classification result.

2. The method according to claim 1, characterized in that, The step of performing feature encoding processing on the portrait data, the behavioral data, and the data to be classified to obtain the feature data to be classified includes: The first encoder is invoked to perform feature encoding on the portrait data to obtain a first feature vector; The second encoder is invoked to perform feature encoding on the behavioral data to obtain a second feature vector; The third encoder is invoked to perform feature encoding on the data to be classified in order to obtain a third feature vector; The first feature vector, the second feature vector, and the third feature vector are used as the feature data to be classified.

3. The method according to claim 1, characterized in that, The second classification process performed on the feature data to be classified to obtain the second classification result of the data to be classified includes: The first classification model is invoked to classify the feature data to be classified, so as to obtain the second classification result.

4. The method according to claim 3, characterized in that, Before calling the second classification model to classify the feature data to be classified in order to obtain the second classification result, the method further includes: Retrieve historical category data and newly added category data; The first classification model is invoked to classify the historical classification data and the newly added classification data to obtain an evaluation classification result; The first classification model is evaluated based on the evaluation and classification results.

5. The method according to claim 3, characterized in that, The method further includes: Obtain a pre-trained large language model, training samples, and prompt words. The training samples include historical profile data, historical behavior data, historical classification data, and label data. The input data for the prompt words is the training samples, and the output data for the prompt words is the label data. The pre-trained large language model is trained based on the training samples and the prompt words to obtain the first classification model.

6. The method according to claim 5, characterized in that, The acquisition of training samples includes: Obtain second reported data, which is historical classification data and the historical classification results of the historical classification data, wherein the historical classification results are secondary label results; The historical classification data and the historical classification results are used as the training samples.

7. The method according to any one of claims 1 to 6, characterized in that, The first classification process performed on the data to be classified to obtain the first classification result includes: The second classification model is invoked to classify the data to be classified in order to obtain the first classification result. The second classification model is a text classification model, an image classification model, or a multimodal classification model.

8. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Obtain third reported data, which includes key indicator data of the second classification results; When it is determined that there is a problem with the second classification result based on the key indicator data, an alarm message is sent.

9. The method according to any one of claims 1 to 6, characterized in that, The method further includes: When the first classification result meets the preset conditions, the first classification result is output.

10. A data classification device, characterized in that, include: The acquisition module is used to acquire the data to be classified. The processing module is used to perform a first classification process on the data to be classified in order to obtain a first classification result; The acquisition module is used to acquire the profile data and behavior data of the interactive object corresponding to the data to be classified when the first classification result does not meet the preset conditions. The interactive object includes the object that uploaded the data to be classified and the object to which the data to be classified belongs. The processing module is used to perform feature encoding processing on the portrait data, the behavior data, and the data to be classified, respectively, to obtain the feature data to be classified. The feature data to be classified is subjected to a second classification process to obtain the second classification result of the data to be classified. The output module is used to output the second classification result.

11. A computer device, characterized in that, include: Memory, processor, and bus system; The memory is used to store programs; The processor is configured to execute a program in the memory, and the processor is configured to execute the method of any one of claims 1 to 9 according to instructions in the program code; The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

12. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method as claimed in any one of claims 1 to 9.

13. A computer program product, comprising a computer program, characterized in that, The computer program is executed by a processor using the method as described in any one of claims 1 to 9.

Images