Text processing method, system and device based on multi-round decision of large language model

Abstract

The embodiment of the specification provides a text processing method, system and device based on multi-round decision of a large language model, which comprises the following steps: obtaining a target text, classifying the target text by using a text classification rule to determine a first classification result corresponding to the target text; inputting the target text into a large language model to obtain a second classification result corresponding to the target text; in the case that the first classification result and the second classification result are inconsistent, determining associated text and sample example data corresponding to the target text; inputting the target text, the associated text and the sample example data into the large language model to obtain a text classification result output by the large language model, and determining the text classification result as a target classification result of the target text; and classifying by further dynamically combining the associated text of the target text and the sample example data of the determined classification result, so that the output of the large language model becomes more reliable, thereby guaranteeing the accuracy of the target classification result.

Description

Technical Field

[0001] This specification relates to the field of text classification technology, and in particular to a text processing method based on multi-round decision-making using a large language model. One or more embodiments of this specification also relate to a text processing system based on multi-round decision-making using a large language model, a text processing device based on multi-round decision-making using a large language model, a computing device, a computer-readable storage medium, and a computer program. Background Technology

[0002] Text classification and grading is a natural language processing task that aims to assign text data to predefined categories or levels. This process typically involves identifying specific features or patterns in the text and categorizing it based on these features. The classification results can help readers make a general judgment about the text content and can also enhance data security.

[0003] In existing methods, predefined classification rules or text processing models are typically used to classify and grade texts to obtain corresponding classification results. However, when a classification result is obtained after making a single decision, the obtained classification result is not reliable enough. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a text processing method based on multi-turn decision-making using a large language model. One or more embodiments of this specification also relate to a text processing system based on multi-turn decision-making using a large language model, a text processing apparatus based on multi-turn decision-making using a large language model, a computing device, a computer-readable storage medium, and a computer program, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a text processing method based on multi-round decision-making using a large language model is provided, comprising:

[0006] Obtain the target text, classify the target text using text classification rules, and determine the first classification result corresponding to the target text;

[0007] The target text is input into a large language model to obtain the second classification result corresponding to the target text.

[0008] If the first classification result is inconsistent with the second classification result, determine the associated text and sample example data corresponding to the target text;

[0009] The target text, the associated text, and the sample example data are input into the large language model to obtain the text classification result output by the large language model, and the text classification result is determined as the target classification result of the target text.

[0010] According to a second aspect of the embodiments of this specification, a text processing system based on multi-turn decision-making using a large language model is provided, including a client and a server, wherein...

[0011] The client is used to send a text classification request to the server;

[0012] The server is configured to respond to the text classification request by: acquiring the target text; classifying the target text using text classification rules to determine a first classification result corresponding to the target text; inputting the target text into a large language model to obtain a second classification result corresponding to the target text; if the first classification result and the second classification result are inconsistent, determining the associated text and sample example data corresponding to the target text; inputting the target text, the associated text, and the sample example data into the large language model to obtain the text classification result output by the large language model, and determining the text classification result as the target classification result of the target text; and returning the target classification result to the client.

[0013] According to a third aspect of the embodiments of this specification, a text processing apparatus based on multi-turn decision-making using a large language model is provided, comprising:

[0014] The first determining module is configured to acquire target text, classify the target text using text classification rules, and determine the first classification result corresponding to the target text;

[0015] The second determining module is configured to input the target text into a large language model to obtain a second classification result corresponding to the target text.

[0016] The text enhancement module is configured to determine the associated text and sample example data corresponding to the target text when the first classification result and the second classification result are inconsistent.

[0017] The target determination module is configured to input the target text, the associated text, and the sample example data into the large language model, obtain the text classification result output by the large language model, and determine the text classification result as the target classification result of the target text.

[0018] According to a fourth aspect of the embodiments of this specification, a computing device is provided, comprising:

[0019] Memory and processor;

[0020] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the above-described text processing method based on multi-round decision-making of a large language model.

[0021] According to a fifth 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 text processing method based on a large language model for multi-turn decision-making.

[0022] According to a sixth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the above-described text processing method based on a large language model for multi-round decision-making.

[0023] This specification provides an embodiment of a text processing method based on a large language model with multi-round decision-making, comprising: acquiring target text; classifying the target text using text classification rules to determine a first classification result corresponding to the target text; inputting the target text into a large language model to obtain a second classification result corresponding to the target text; ensuring the accuracy of the results by comparison and verification when the first and second classification results are obtained using different processing methods, specifically determining the associated text and sample example data corresponding to the target text when the first and second classification results are inconsistent; inputting the target text, the associated text, and the sample example data into the large language model to obtain the text classification result output by the large language model, and determining the text classification result as the target classification result of the target text; that is, by further dynamically combining the associated text of the target text and the sample example data of the determined classification result for classification, the output of the large language model becomes more reliable, thereby ensuring the accuracy of the target classification result. Attached Figure Description

[0024] Figure 1 This is a schematic diagram illustrating a scenario of a text processing method based on a large language model and multi-round decision-making, provided in one embodiment of this specification.

[0025] Figure 2 This is a flowchart illustrating a text processing method based on a large language model and multi-round decision-making, as provided in one embodiment of this specification.

[0026] Figure 3 This is a flowchart illustrating the processing procedure of a text processing method based on a large language model and multi-round decision-making, provided in one embodiment of this specification.

[0027] Figure 4This is a schematic diagram of the structure of a text processing device based on a large language model and multi-round decision-making, provided in one embodiment of this specification.

[0028] Figure 5 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0029] 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.

[0030] 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.

[0031] 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."

[0032] 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.

[0033] In one or more embodiments of this specification, a large model refers to a deep learning model with a large number of model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even tens of trillions of model parameters. A large model can also be called a foundation model. It is pre-trained using large-scale unlabeled corpora to produce a pre-trained model with hundreds of millions of parameters. Such models can adapt to a wide range of downstream tasks and have good generalization ability. Examples include Large Language Models (LLMs) and multi-modal pre-training models.

[0034] In practical applications, large models only require a small number of samples to fine-tune the pre-trained model before they can be applied to different tasks. Large models can be widely used in fields such as Natural Language Processing (NLP) and Computer Vision. Specifically, they can be applied to computer vision tasks such as Visual Question Answering (VQA), Image Captioning (IC), and Image Generation, as well as natural language processing tasks such as text-based sentiment classification, text summarization, and machine translation. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design.

[0035] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0036] Large language models (MLMs) are typically based on deep neural networks, especially the Transformer architecture, and are trained on massive amounts of data to learn complex language patterns and semantic relationships. Through self-attention mechanisms, MLMs can capture complex relationships between words and extract semantic information at multiple levels, thereby understanding the deeper meaning of sentences. MLMs often have a very large number of parameters, but it is precisely this large number that gives them context-awareness and deep semantic expression capabilities, which play a crucial role in text classification.

[0037] The core function of text classification and categorization lies in organizing and structuring large amounts of text information, making it easier to disseminate, circulate, and process in a targeted manner. It also supports various application scenarios such as decision-making, personalized recommendations, and content filtering. Text classification and categorization play a significant role in security, business, and economic aspects. Furthermore, regardless of the specific field, whether it's finance, healthcare, or general enterprise office and media communication scenarios, text classification and categorization are indispensable.

[0038] However, existing methods for text classification and grading are usually based on text classification models. For example, when processing based on text classification models, the model structure is optimized by knowledge base enhancement and attention network. That is, by introducing an external knowledge base to enhance the semantic representation of the text and by using an improved attention mechanism to improve the accuracy of multi-label classification, the core is static feature modeling and label dependency mining.

[0039] However, this approach is not only complex and costly to train, but also lacks deep semantic understanding capabilities. Furthermore, the classification results are not reliable when a single decision is made to obtain the classification result.

[0040] This specification provides a text processing method based on multi-turn decision-making using a large language model. It also relates to a text processing system based on multi-turn decision-making using a large language model, a text processing apparatus based on multi-turn decision-making using a large language model, a computing device, and a computer-readable storage medium, which will be described in detail in the following embodiments.

[0041] See Figure 1 , Figure 1 This illustration shows a scenario diagram of a text processing method based on a large language model and multi-round decision-making, provided in one embodiment of this specification.

[0042] Specifically, the text processing method based on multi-round decision-making of a large language model is applied to a text processing system based on multi-round decision-making of a large language model. The text processing system includes an end device 102 and a server 104. The end device 102 is used to send text classification requests to the server 104.

[0043] A large language model is deployed in server 104. Server 104 is used to respond to the text classification request, obtain the target text, classify the target text using text classification rules, and determine the first classification result corresponding to the target text; input the target text into the large language model to obtain the second classification result corresponding to the target text; if the first classification result and the second classification result are inconsistent, determine the associated text and sample example data corresponding to the target text; input the target text, the associated text, and the sample example data into the large language model to obtain the text classification result output by the large language model, and determine the text classification result as the target classification result of the target text; return the target classification result to the end device 102.

[0044] The edge device 102 may include a browser, an app (application), or a web application such as an H5 (Hypertext Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. The edge device can be developed based on a software development kit (SDK) provided by the server, such as a real-time communication (RTC) SDK. The edge device can be deployed in an electronic device and depends on the device's operation or certain apps within the device to run. The electronic device may have a display screen and support information browsing, such as a personal mobile terminal like a mobile phone, tablet, or personal computer. Various other types of applications can also be configured in the electronic device, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platform software.

[0045] Server 104 can be understood as a server providing various services, including physical servers and cloud servers. Examples include servers providing communication services to multiple clients, servers supporting backend training of models used on clients, and servers processing data sent by clients. It's important to note that Server 104 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. Server 104 can also be a server in a distributed system, or a server integrated with blockchain. Server 104 can also be a cloud server providing 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 networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0046] It is worth noting that the text processing method provided in the embodiments of this specification can be executed by the server 104. In other embodiments of this specification, the large language model can be deployed on the terminal device 102, so that the terminal device 102 can also have similar functions to the server 104, thereby executing the text processing method provided in the embodiments of this specification. In other embodiments, the text processing method provided in the embodiments of this specification can also be jointly executed by the terminal device 102 and the server 104.

[0047] This specification provides an embodiment of a text processing method based on a large language model with multi-round decision-making, comprising: acquiring target text; classifying the target text using text classification rules to determine a first classification result corresponding to the target text; inputting the target text into a large language model to obtain a second classification result corresponding to the target text; ensuring the accuracy of the results by comparison and verification when the first and second classification results are obtained using different processing methods, specifically determining the associated text and sample example data corresponding to the target text when the first and second classification results are inconsistent; inputting the target text, the associated text, and the sample example data into the large language model to obtain the text classification result output by the large language model, and determining the text classification result as the target classification result of the target text; that is, by further dynamically combining the associated information of the target text and the sample information of the determined classification result for classification, the output of the large language model becomes more reliable, thereby ensuring the accuracy of the target classification result.

[0048] See Figure 2 , Figure 2 A flowchart is shown of a text processing method based on a large language model and multi-round decision-making according to an embodiment of this specification, which specifically includes the following steps.

[0049] Step 202: Obtain the target text, classify the target text using text classification rules, and determine the first classification result corresponding to the target text.

[0050] The target text can be understood as the raw text data to be classified, which can be a single piece of text or multiple pieces of text (such as user comments, news headlines, customer service conversations, etc.). Text classification rules can be understood as a predefined set of judgment criteria and conditions, including but not limited to keyword matching rules and regular expressions. The first classification result can be understood as the category label output by matching the target text according to the text classification rules.

[0051] Specifically, the target text to be classified is obtained, and features are extracted from the target text. These features include, but are not limited to, keyword extraction and entity recognition. Taking keyword matching as an example, the target text is classified using the keyword matching rule to obtain the first classification result corresponding to the target text.

[0052] For example, in a product review scenario, the target text could be "The logistics speed is too slow; it was supposed to arrive the next day, but it took three days." The extracted keywords include "logistics speed" and "too slow." Therefore, according to the keyword matching rule that "if the keyword contains 'slow,' the text will be classified as negative feedback," the target text is classified as negative, meaning the first classification result is negative.

[0053] In one or more embodiments of this specification, text classification rules can be extracted based on sample data in a sample database. Specific implementation methods are described below:

[0054] Before classifying the target text using text classification rules and determining the first classification result corresponding to the target text, the method further includes:

[0055] A sample database is determined, and rules are extracted from the sample data in the sample database to obtain the text classification rules.

[0056] The sample database can be understood as a collection of data used to provide sample data. In the embodiments of this specification, the sample data in the sample database includes sample text and the corresponding classification results. Rule extraction can be understood as summarizing reusable classification judgment logic from the sample data; text classification rules can be understood as the final decision-making basis system.

[0057] Specifically, cluster analysis can be used to label typical text patterns to form a basic rule set, and the final text classification rules can be obtained through manual verification. Of course, to improve data processing efficiency, data processing models can also be used to analyze sample data in the sample database to obtain text classification rules, which is not limited here.

[0058] For example, in the context of legal document processing, the text classification rule generated by analyzing 50,000 contract samples is that texts with "the breaching party shall" in the first line and a percentage number at the end can be classified as "breach of contract liability clauses".

[0059] The text processing method provided in the embodiments of this specification can accurately obtain text classification rules by extracting rules from sample data with known classification results. Furthermore, when the sample database is scalable, it can also promote the updating of text classification rules, thereby improving the accuracy of the first classification result.

[0060] In one or more embodiments of this specification, by performing semantic segmentation and normalization on the obtained initial text, at least one segmented text with different semantics is obtained, and the target text is determined from the at least one segmented text. Specific implementation methods are as follows:

[0061] The acquisition of the target text includes:

[0062] Determine an initial text, perform semantic segmentation and normalization on the initial text, and obtain at least one segmented text;

[0063] The target text is determined from the at least one segmented text, wherein the target text is any one of the at least one segmented text.

[0064] Here, "initial text" can be understood as unprocessed raw text data, such as news articles, technical documents, or web page content; it is typically unstructured data. Semantic segmentation can be understood as a segmentation operation based on the semantic boundaries of the text; the text obtained through semantic segmentation contains complete semantic information. Normalization processing can be understood as the transformation process of converting text into a standard form, including but not limited to transformations in form, content, and language style.

[0065] Segmented text can be understood as independent text units obtained after semantic segmentation and normalization. Segmented text has semantic integrity and format consistency.

[0066] Specifically, taking a news article as the initial text, the news article is semantically segmented into multiple semantic units. These segments are then standardized (by performing operations such as error correction, case unification, and symbol unification) to obtain a standardized segmented text.

[0067] In practice, the initial text can be semantically segmented using a large language model to improve the efficiency of semantic segmentation; however, no specific limitations are imposed here.

[0068] Any one of the segmented texts can be used as the target text for subsequent classification processing to obtain the corresponding target classification result.

[0069] The text processing method provided in the embodiments of this specification, when the target text is obtained through semantic segmentation and normalization, contains complete semantic information, thereby ensuring that a unified classification result is obtained based on consistent semantic information. Furthermore, the normalized target text facilitates the subsequent structured processing of large language models, improving processing efficiency.

[0070] Step 204: Input the target text into the large language model to obtain the second classification result corresponding to the target text.

[0071] The second classification result can be understood as an effective classification output obtained through a large language model. When the second classification result is obtained through a large language model, it has semantic understanding depth and can be compared with the first classification result obtained by the above text classification rules as an effective classification result.

[0072] Specifically, the target text is input into a large language model, which extracts features from the target text and makes a classification decision based on the extracted text features to obtain the output second classification result.

[0073] In one or more embodiments of this specification, an initial classification result corresponding to the target text is obtained through a large language model, and a confidence score corresponding to the initial classification result is determined. If the initial classification result is determined to be valid based on the confidence score, the initial classification result is determined as a second classification result; in fact, the initial classification result is determined to be valid if the confidence score reaches a preset threshold. If the initial classification result is determined to be invalid, context enhancement processing is performed on the target text, thereby enabling the large language model to classify the context-enhanced target text and obtain a second classification result. Specific implementation methods are described below:

[0074] The step of inputting the target text into a large language model to obtain the second classification result corresponding to the target text includes:

[0075] The target text is input into the large language model to obtain the initial classification result output by the large language model, and the confidence score corresponding to the initial classification result is determined.

[0076] Determine whether the confidence score reaches a preset threshold; if so, determine the initial classification result as the second classification result.

[0077] If not, then the target text is subjected to context enhancement processing, and the second classification result is obtained based on the context-enhanced target text and the large language model.

[0078] The initial classification result can be understood as the category determination output by the large language model after preliminary analysis of the target text; the confidence score can be understood as the quantitative assessment of the certainty of the initial classification result by the large language model, usually represented by a probability value (0-1), and the higher the value, the greater the model's confidence in the classification result.

[0079] The preset threshold can be understood as a manually set confidence threshold, used to distinguish between high-confidence and low-confidence classification results, ensuring output reliability. Context enhancement processing can be understood as supplementing semantic information to improve classification accuracy by using contextual information or external knowledge associated with the target text.

[0080] Specifically, taking the news text "Apple performed exceptionally well" as the target text as an example, the target text is input into a large language model. Because "apple" has the ambiguity of "fruit" and "brand," the large language model initially classifies it as "agricultural product," thus determining the initial classification result as "agricultural product." However, the confidence score of this initial classification result is 0.62. With a confidence score corresponding to a preset threshold of 0.7, the target text is enhanced with context. For example, the search finds that the previous text mentions "iPhone sales growth." Therefore, the context-enhanced target text "Apple (brand) performed exceptionally well" is re-input into the large language model. The enhanced target text eliminates ambiguity, and the classification result "consumer electronics market" is obtained from the output of the large language model. The confidence score of this classification result is increased to 0.75, which meets the preset threshold requirement. Therefore, "consumer electronics market" is determined as the second classification result.

[0081] The text processing method provided in the embodiments of this specification obtains the initial classification result corresponding to the target text through a large language model. When determining the confidence score corresponding to the initial classification result, it determines whether the obtained initial classification result can be identified as a second classification result based on whether the confidence score reaches a preset threshold. If the initial classification result cannot be used as a second classification result, the semantic information is enhanced through context enhancement to improve the accuracy of the model's output classification result, thereby obtaining an effective second classification result.

[0082] In one or more embodiments of this specification, context enhancement processing of the target text is achieved by determining the associated text corresponding to the target text and sample example data. The enhanced classification result obtained after the enhancement processing is then verified to ensure that if the enhanced classification result is valid, it is determined as the second classification result. Specific implementation methods are described below:

[0083] The step of performing context enhancement processing on the target text, and obtaining the second classification result based on the context-enhanced target text and the large language model, includes:

[0084] Determine the associated text corresponding to the target text and the sample example data, input the target text, the associated text, and the sample example data into the large language model, and obtain the enhanced classification result output by the large language model;

[0085] The enhanced classification result is determined as the initial classification result, and the steps of determining the confidence score corresponding to the initial classification result and judging whether the confidence score reaches a preset threshold are continued until the confidence score reaches the preset threshold or the number of iterations reaches a preset number, and the initial classification result is determined as the second classification result.

[0086] Among them, related text can be understood as contextual content that has a direct or indirect semantic, thematic, or logical connection with the target text, used to supplement the complete semantic information of the target text. Sample example data can be understood as typical data extracted from a pre-set sample database that is related to the category corresponding to the target text, serving as a classification reference standard.

[0087] Enhanced classification results can be understood as optimized classification results output by a large language model after combining the target text, related text, and sample example data. Their confidence level is typically higher than that of a single target text input. The iteration rounds can be understood as the number of times context enhancement and classification validation are executed in a loop, used to control the termination conditions of the process.

[0088] Specifically, if the confidence score of the initial classification result does not reach the preset threshold, the associated context of the target text and sample example data of the same type are extracted. These three are input into the large language model to generate an enhanced classification result, and the confidence score of the enhanced classification result is calculated. If the confidence score still does not reach the preset threshold, the above enhancement process is repeated until the threshold requirement is met or the maximum number of iterations is reached. Finally, the second classification result after multiple rounds of optimization is output.

[0089] In practical applications, taking the target text "interface returned error status code 44" as an example, the initial classification result "network communication" has a confidence score of 0.65. By obtaining the preceding context "interface request failed when user login" and matching it with sample example data in the sample database, and inputting the target text, preceding context, and sample data into the large language model, the large language model combines the context "interface request" with the sample example data "4xx status code is a client error" to output the enhanced classification result "client request abnormal," and determines that the confidence score of this enhanced classification result is 0.85. This confidence score meets the preset threshold, so the process terminates, and this enhanced classification result is determined as the second classification result.

[0090] In another embodiment of this specification, the target text "a party breaching the contract shall pay liquidated damages" undergoes three context enhancement processes (associating with contract types, sample breach situations, and case law databases), and the confidence score of the enhanced classification result always hovers between 0.5 and 0.6; after the number of iterations reaches a preset number (e.g., 5 times), the output enhanced classification result is obtained, and this enhanced classification result is determined as the second classification result.

[0091] The text processing method provided in the embodiments of this specification achieves dynamic knowledge fusion by associating text with sample example data when the confidence score of the initial classification result does not reach a preset threshold. That is, it supplements the target text with both immediate context and introduces domain prior knowledge to ensure that the credibility of the enhanced classification result is improved. If the confidence score of the enhanced classification result does not reach the preset threshold, it can be continuously optimized through multiple iterations. The preset number of iterations can prevent infinite loops and ensure system response efficiency.

[0092] In one or more embodiments of this specification, when a first classification result based on text classification rules and a second classification result output by a large language model are obtained, the consistency between the first classification result and the second classification result is determined. If they are consistent, the first classification result and / or the second classification result is determined as the target classification result of the target text. Specific implementation methods are as follows:

[0093] After inputting the target text into the large language model to obtain the second classification result corresponding to the target text, the method further includes:

[0094] If the first classification result is consistent with the second classification result, the first classification result and / or the second classification result shall be determined as the target classification result of the target text.

[0095] The target classification result can be understood as the final classification result corresponding to the target text.

[0096] Specifically, the first classification result is obtained by classifying the target text using text classification rules, and the second classification result is obtained by classifying the target text using a large language model. When the consistency of the first and second classification results obtained by multiple classification methods is judged, if the multiple classification results are consistent, the first classification result and / or the second classification result is determined as the final classification result of the target text.

[0097] The text processing method provided in the embodiments of this specification improves the reliability of the final classification result by judging the consistency of classification results obtained using different classification methods. If consistency is determined, the obtained classification result is determined as the final classification result.

[0098] Step 206: If the first classification result is inconsistent with the second classification result, determine the associated text and sample example data corresponding to the target text.

[0099] Specifically, if the first classification result is inconsistent with the second classification result, it means that the currently obtained first classification result and second classification result are unreliable. Therefore, context enhancement processing is performed on the target text, that is, the associated text and sample example data corresponding to the target text are determined.

[0100] In fact, because large language models have excellent semantic understanding capabilities, when the classification of the target text is determined to be unreliable, context enhancement processing of the target text can be performed, combined with the semantic understanding capabilities of the large language model, to improve the reliability of the classification results of the target text.

[0101] In one or more embodiments of this specification, the associated text corresponding to the target text contains contextual information about the target text. Specifically, when the target text is determined from at least one segmented text, and the at least one segmented text is obtained based on the initial text segmentation, there is a contextual relationship between the at least one segmented text. Therefore, the associated text corresponding to the target text can be determined from the at least one segmented text, thereby obtaining the contextual information of the target text. When determining sample example data from the sample database, typical data related to the obtained classification results can be filtered from the sample database. Specific implementation methods are described below:

[0102] The process of determining the associated text and sample example data corresponding to the target text includes:

[0103] Determine the associated text corresponding to the target text from the at least one segmented text, and determine the sample example data corresponding to the target text from the sample database based on the first classification result.

[0104] Among them, associated text can be understood as contextual fragments (such as preceding / following paragraphs, dialogue history, etc.) located from the original text data that are related to the target text; sample example data can be understood as typical data related to the first classification result selected from the sample database. The sample example data includes text samples and the classification results corresponding to the text samples.

[0105] Specifically, within at least one segmented text obtained through semantic segmentation, related texts of the target text can be determined through methods such as positional adjacency (e.g., text within three consecutive paragraphs) and semantic coherence (consistent theme). When determining the sample example data corresponding to the target text, typical data related to that category label can be matched from the sample database based on the category label of the first classification result.

[0106] In practical applications, since related texts can be determined through different acquisition methods, the weight ratio of related elements can be automatically adjusted according to the characteristics of the target text. For example, when processing legal texts, emphasis can be placed on the correlation of clause positions, while when analyzing academic papers, the correlation of literature citations can be strengthened. Furthermore, when determining sample example data, some negative samples can be identified from the sample database for comparative learning of the large language model, improving the clarity of classification boundaries. For example, in medical scenarios, both confirmed and suspected case samples can be provided for the model to distinguish, thereby improving classification accuracy.

[0107] The text processing method provided in the embodiments of this specification improves the accuracy of target text classification by combining the context information of the target text with sample example data of the determined classification results, where at least one segmented text is obtained based on the segmentation of the initial text, and both the target text and the associated text are determined from at least one segmented text.

[0108] In one or more embodiments of this specification, sample example data corresponding to the target text is determined from a sample database based on a first classification result using a similarity matching method. Specific implementation methods are described below:

[0109] The step of determining the sample example data corresponding to the target text from the sample database based on the first classification result includes:

[0110] The first classification result is matched with the sample database for similarity, and the sample example data corresponding to the target text is determined from the sample database based on the similarity matching result.

[0111] Similarity matching can be understood as evaluating the degree of association between the first classification result and the sample database entries through quantitative calculation. In fact, similarity matching can include two dimensions: surface feature similarity and deep semantic relevance. That is, in the field of natural language processing, similarity matching not only considers word form overlap, but also focuses on the consistency of contextualized semantic expression.

[0112] Specifically, when the first classification result is inconsistent with the second classification result, the target text continues to undergo context enhancement processing. Therefore, it is necessary to determine the sample example data corresponding to the target text from the sample database.

[0113] Once the first classification result is determined, keyword matching and / or semantic matching are performed in the sample database to determine the similarity score. The sample data are then sorted in descending order according to the similarity score, and the top K (set according to the actual situation) sample data are selected as sample example data for the target text. In fact, by using dynamic threshold (K) control, it can be ensured that different categories can obtain a suitable number of samples.

[0114] The text processing method provided in the embodiments of this specification determines sample example data related to the target text from the sample database through similarity matching, thereby ensuring the reliability of the target text classification results when using sample example data as reference data for classifying target text.

[0115] Step 208: Input the target text, the associated text, and the sample example data into the large language model to obtain the text classification result output by the large language model, and determine the text classification result as the target classification result of the target text.

[0116] Specifically, since large language models have excellent semantic understanding capabilities, by combining the related text of the target text with sample example data of the already determined classification results, and having the large language model perform a classification again, it can be determined that the text classification results output by the large language model are reliable enough. Therefore, the text classification results are determined as the target classification results.

[0117] In one or more embodiments of this specification, after obtaining the target classification result of the target text, the target text and the target classification result can be stored in a sample database to expand the sample database and facilitate subsequent updates to the text classification rules. Specific implementation methods are as follows:

[0118] After determining the target classification result of the target text, the method further includes:

[0119] The target text and the target classification result are identified as sample data and stored in the sample database.

[0120] Specifically, the sample database can be expanded by identifying the target text and target classification results as sample data and storing them in the sample database.

[0121] When text classification rules are extracted from a sample database, the rules can be dynamically updated as the sample database expands, thus covering more fields and becoming more comprehensive.

[0122] The text processing method provided in the embodiments of this specification ensures the accuracy of the final classification result by comparing and verifying the output result of the large language model (the second classification result) with the first classification result. That is, the reliability of the final classification result is further improved by using classification methods from multiple sources. Specifically, when the first classification result and the second classification result are inconsistent, the contextual information of the target text and the sample information of the determined result are dynamically combined for classification, making the text classification result output by the large language model more reliable. This process is fully automated and does not require manual intervention, saving human resources.

[0123] See Figure 3 , Figure 3 The diagram illustrates a process flow of a text processing method based on a large language model and multi-round decision-making, according to one embodiment of this specification.

[0124] Obtain the original text (i.e., the initial text in the above embodiment), use a large language model to perform semantic segmentation on the original text, divide the text into multiple semantic units (chunks), and perform normalization processing (such as standardization, error correction, etc.) on the segmented chunks to obtain a normalized set of chunks (i.e., the segmented text in the above embodiment).

[0125] Select a chunk (i.e., the target text in the above embodiment) from the normalized chunk set, apply predefined rules (i.e., the text classification rules in the above embodiment, which are extracted from the sample library) to the normalized single chunk, and output a preliminary classification result (result1, i.e., the first classification result in the above embodiment).

[0126] The first classification result is obtained using the above method, and the following steps are performed on the path where the other classification result was obtained:

[0127] Confidence verification: Use a large language model to perform inference classification on the target text, output the initial classification result and confidence score, and determine whether the confidence score is greater than 0.7 or whether the maximum number of iterations has been reached. If so, determine the initial classification result as the second classification result (result2) and proceed to the result consistency verification step. If not, proceed to the context enhancement processing step.

[0128] Consistency verification of results: Compare whether result11 and result2 are consistent. If they are consistent, output result1 and / or result2 as the final result (i.e., the target classification result). If not, proceed to the context enhancement processing step.

[0129] Context enhancement processing: Collect the context chunks of the current chunk (i.e., the associated text in the above embodiments), then sample some sample example data from the sample library (i.e., the sample database in the above embodiments) for the large language model to refer to, and finally form a new input to be re-inputted into the large language model for inference. If the context enhancement processing step is entered from the result consistency verification step, the final classification result is output; otherwise, the confidence verification step is required.

[0130] The text processing methods described above enable text classification and grading, allowing the system to more intelligently understand textual information and respond to user needs, thereby improving user experience. Therefore, text classification and grading are widely used across various industries, especially in the business world. Examples include: Email classification: automatically identifying spam and legitimate emails, improving email management efficiency; Sentiment analysis: determining consumer sentiment towards products or services by analyzing social media comments; News classification: helping news aggregation websites categorize articles by topic, improving user experience; Legal document processing: automatically identifying and classifying different types of legal documents, assisting lawyers in quickly finding relevant information; Medical record management: classifying medical records for doctors' reference; Financial risk assessment: assessing potential financial risks by classifying transaction records and customer information; Educational material classification: providing automatic classification of teaching materials and resources for online learning platforms, facilitating use by students and teachers; E-commerce product classification: ensuring products are correctly listed under their appropriate categories, enhancing the shopping experience.

[0131] Even within the context of financial data privacy sensitivity classification (stemming from the high sensitivity and complexity of financial data; on the one hand, financial data encompasses core privacy information such as personal identity, assets, and transactions, and its leakage could lead to risks such as identity theft and financial loss; on the other hand, global privacy regulations impose stringent requirements on the processing of financial data, mandating that companies implement differentiated protection based on data sensitivity; simultaneously, the rapid development of fintech, such as mobile payments and blockchain, has made data diverse, dynamic, and highly interconnected, exacerbating the conflict between privacy protection and business innovation. Through privacy sensitivity classification, financial institutions can clarify data protection priorities and dynamically adapt to business scenario needs, thereby maximizing data value under the premise of compliance), this study researches a data classification model to address the differences in privacy sensitivity of diverse and integrated data. This model categorizes data into four levels—basic data, core data, value-added data, and special data—based on privacy sensitivity.

[0132] Specifically, given a document, the document is semantically segmented, and each semantic block is classified according to its data privacy sensitivity into four categories. Based on the data classification results, a specific data circulation and transmission method is adopted to balance security and commercial value.

[0133] By classifying data according to privacy sensitivity, corresponding measures are taken to avoid legal risks arising from the misuse of highly sensitive data. Simultaneously, implementing specific strategies can meet the needs of financial services (payment services, credit risk control, etc.). Furthermore, low-sensitivity data (basic data) can be shared with partners to develop innovative services (such as market analysis tools), while highly sensitive data (value-added / special data) is securely utilized through privacy-preserving computation technology. This allows value-added data to be used for user profiling, targeted marketing, and other scenarios while protecting privacy, thus balancing security and commercial value.

[0134] In other words, when text classification and grading can be achieved using the methods described above, the structured management of text through classification tags facilitates rapid location and retrieval, thereby improving retrieval efficiency; text is graded according to sensitivity, and differentiated protection strategies are matched to control security and compliance risks; the classification results support the automated execution of business logic, enabling business scenario adaptation and decision support; grading reduces redundant processing, improves resource utilization, and achieves resource optimization and cost savings; and the classified text can be used as structured data input to analysis models to support business insights, realize knowledge mining, and unlock commercial value.

[0135] The text processing method provided in the embodiments of this specification is based on the large language model, which has excellent semantic understanding capabilities and universality. For data in different scenarios, it does not require retraining the model, making it more automated. Furthermore, through a multi-round decision-making principle, the output results of the large language model are compared and verified to ensure the accuracy of the results. Specifically, if the classification result is deemed unreliable by the large language model, the large language model is reclassified by combining the contextual information of the target text and the sample information of the already determined results, until the large language model considers the output classification result to be sufficiently reliable. This multi-round decision-making method has good interpretability and conforms to real-world application scenarios. In other words, this method has higher recognition accuracy, greater reliability, and better interpretability.

[0136] Corresponding to the above method embodiments, this specification also provides an embodiment of a text processing device based on multi-round decision-making using a large language model. Figure 4 This specification illustrates a schematic diagram of a text processing apparatus based on a large language model and multi-turn decision-making, according to one embodiment of this specification. Figure 4 As shown, the device includes:

[0137] The first determining module 402 is configured to acquire target text, classify the target text using text classification rules, and determine the first classification result corresponding to the target text;

[0138] The second determining module 404 is configured to input the target text into a large language model to obtain a second classification result corresponding to the target text.

[0139] The text enhancement module 406 is configured to determine the associated text and sample example data corresponding to the target text when the first classification result and the second classification result are inconsistent.

[0140] The target determination module 408 is configured to input the target text, the associated text, and the sample example data into the large language model, obtain the text classification result output by the large language model, and determine the text classification result as the target classification result of the target text.

[0141] Optionally, the first determining module 402 is further configured to:

[0142] Determine an initial text, perform semantic segmentation and normalization on the initial text, and obtain at least one segmented text;

[0143] The target text is determined from the at least one segmented text, wherein the target text is any one of the at least one segmented text.

[0144] Optionally, the second determining module 404 is further configured to:

[0145] The target text is input into the large language model to obtain the initial classification result output by the large language model, and the confidence score corresponding to the initial classification result is determined.

[0146] Determine whether the confidence score reaches a preset threshold; if so, determine the initial classification result as the second classification result.

[0147] If not, then the target text is subjected to context enhancement processing, and the second classification result is obtained based on the context-enhanced target text and the large language model.

[0148] Optionally, the second determining module 404 is further configured to:

[0149] Determine the associated text corresponding to the target text and the sample example data, input the target text, the associated text, and the sample example data into the large language model, and obtain the enhanced classification result output by the large language model;

[0150] The enhanced classification result is determined as the initial classification result, and the steps of determining the confidence score corresponding to the initial classification result and judging whether the confidence score reaches a preset threshold are continued until the confidence score reaches the preset threshold or the number of iterations reaches a preset number, and the initial classification result is determined as the second classification result.

[0151] Optionally, the text enhancement module 406 is further configured to:

[0152] Determine the associated text corresponding to the target text from the at least one segmented text, and determine the sample example data corresponding to the target text from the sample database based on the first classification result.

[0153] Optionally, the text enhancement module 406 is further configured to:

[0154] The first classification result is matched with the sample database for similarity, and the sample example data corresponding to the target text is determined from the sample database based on the similarity matching result.

[0155] Optionally, the target determination module 408 is further configured to:

[0156] If the first classification result is consistent with the second classification result, the first classification result and / or the second classification result shall be determined as the target classification result of the target text.

[0157] The device further includes:

[0158] The rule extraction module is configured to determine a sample database and extract rules from the sample data in the sample database to obtain the text classification rules.

[0159] The device further includes:

[0160] The sample storage module is configured to identify the target text and the target classification result as sample data and store them in the sample database.

[0161] The text processing device based on a large language model and multi-round decision-making provided in this specification acquires target text, classifies the target text using text classification rules, and determines a first classification result corresponding to the target text; inputs the target text into the large language model to obtain a second classification result corresponding to the target text; when the first and second classification results are obtained using different processing methods, the accuracy of the results is ensured through comparison and verification. Specifically, when the first and second classification results are inconsistent, the associated text and sample example data corresponding to the target text are determined; the target text, the associated text, and the sample example data are input into the large language model to obtain the text classification result output by the large language model, and the text classification result is determined as the target classification result of the target text; that is, by further dynamically combining the associated information of the target text and the sample information of the determined classification result for classification, the output of the large language model becomes more reliable, thereby ensuring the accuracy of the target classification result.

[0162] The above is an illustrative scheme of a text processing device according to this embodiment. It should be noted that the technical solution of this text processing device and the technical solution of the above-described text processing method belong to the same concept. For details not described in detail in the technical solution of the text processing device, please refer to the description of the technical solution of the above-described text processing method.

[0163] Figure 5 A structural block diagram of a computing device 500 according to one embodiment of this specification is shown. The components of the computing device 500 include, but are not limited to, a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and a database 550 is used to store data.

[0164] The computing device 500 also includes an access device 540, which enables the computing device 500 to communicate via one or more networks 560. 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 540 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.

[0165] In one embodiment of this specification, the above-described components of the computing device 500 and Figure 5 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 5 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.

[0166] Computing device 500 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). Computing device 500 can also be a mobile or stationary server.

[0167] The processor 520 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-described text processing method.

[0168] 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 text 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 text processing method.

[0169] 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 text processing method.

[0170] 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 text processing method belong to the same concept, and all details not described in detail in the technical solution of the storage medium can be found in the description of the technical solution of the above-described text processing method.

[0171] An embodiment of this specification also provides a computer program, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the above-described text processing method.

[0172] The above is an illustrative example of a computer program according to this embodiment. It should be noted that the technical solution of this computer program and the technical solution of the text processing method described above belong to the same concept. Details not described in detail in the technical solution of the computer program can be found in the description of the technical solution of the text processing method described above.

[0173] 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.

[0174] 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.

[0175] 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.

[0176] 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.

[0177] The preferred embodiments disclosed above are merely illustrative of this specification. The 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 herein. 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. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A text processing method based on large language model multi-round decision, characterized in that, include: Obtain the target text, classify the target text using text classification rules, and determine the first classification result corresponding to the target text; The target text is input into a large language model to obtain a second classification result corresponding to the target text. The second classification result is obtained based on the context-enhanced target text and the large language model when the confidence score corresponding to the initial classification result does not reach a preset threshold. The initial classification result is obtained by inputting the target text into the large language model. In the case where the first classification result and the second classification result are inconsistent, the associated text and sample example data corresponding to the target text are determined, wherein the associated text and the target text originate from the same initial text, the associated text is a context fragment that is related to the target text, and the sample example data is determined from the sample database; The target text, the associated text, and the sample example data are input into the large language model to obtain the text classification result output by the large language model, and the text classification result is determined as the target classification result of the target text.

2. The method of claim 1, wherein, The acquisition of the target text includes: Determine an initial text, perform semantic segmentation and normalization on the initial text, and obtain at least one segmented text; The target text is determined from the at least one segmented text, wherein the target text is any one of the at least one segmented text.

3. The method according to claim 1 or 2, characterized in that, The step of inputting the target text into a large language model to obtain the second classification result corresponding to the target text further includes: The target text is input into the large language model to obtain the initial classification result output by the large language model, and the confidence score corresponding to the initial classification result is determined. If the confidence score reaches the preset threshold, the initial classification result is determined as the second classification result.

4. The method of claim 3, wherein, The step of performing context enhancement processing on the target text, and obtaining the second classification result based on the context-enhanced target text and the large language model, includes: Determine the associated text corresponding to the target text and the sample example data, input the target text, the associated text, and the sample example data into the large language model, and obtain the enhanced classification result output by the large language model; The enhanced classification result is determined as the initial classification result, and the steps of determining the confidence score corresponding to the initial classification result and judging whether the confidence score reaches a preset threshold are continued until the confidence score reaches the preset threshold or the number of iterations reaches a preset number, and the initial classification result is determined as the second classification result.

5. The method of claim 2, wherein, The process of determining the associated text and sample example data corresponding to the target text includes: Determine the associated text corresponding to the target text from the at least one segmented text, and determine the sample example data corresponding to the target text from the sample database based on the first classification result.

6. The method of claim 5, wherein, The step of determining the sample example data corresponding to the target text from the sample database based on the first classification result includes: The first classification result is matched with the sample database for similarity, and the sample example data corresponding to the target text is determined from the sample database based on the similarity matching result.

7. The method of claim 1, wherein, After inputting the target text into the large language model to obtain the second classification result corresponding to the target text, the method further includes: If the first classification result is consistent with the second classification result, the first classification result and / or the second classification result shall be determined as the target classification result of the target text.

8. The method of claim 1 or 2, wherein, Before classifying the target text using text classification rules and determining the first classification result corresponding to the target text, the method further includes: A sample database is determined, and rules are extracted from the sample data in the sample database to obtain the text classification rules.

9. The method according to claim 1 or 7, characterized in that, After determining the target classification result of the target text, the method further includes: The target text and the target classification result are identified as sample data and stored in the sample database.

10. A text processing system based on large language model multi-round decision, characterized in that, Including both client and server sides, among which, The client is used to send a text classification request to the server; The server, in response to the text classification request, acquires the target text, classifies the target text using text classification rules, and determines a first classification result corresponding to the target text; inputs the target text into a large language model to obtain a second classification result corresponding to the target text, wherein the second classification result is obtained based on the context-enhanced target text and the large language model when the confidence score corresponding to the initial classification result does not reach a preset threshold, and the initial classification result is obtained by inputting the target text into the large language model; if the first classification result and the second classification result are inconsistent, it determines the associated text and sample example data corresponding to the target text, wherein the associated text and the target text originate from the same initial text, the associated text is a context fragment related to the target text, and the sample example data is determined from a sample database; inputs the target text, the associated text, and the sample example data into the large language model to obtain the text classification result output by the large language model, and determines the text classification result as the target classification result of the target text; and returns the target classification result to the client.

11. A text processing apparatus based on large language model multi-round decision, characterized in that, include: The first determining module is configured to acquire target text, classify the target text using text classification rules, and determine the first classification result corresponding to the target text; The second determining module is configured to input the target text into a large language model to obtain a second classification result corresponding to the target text. The second classification result is obtained based on the context-enhanced target text and the large language model when the confidence score corresponding to the initial classification result does not reach a preset threshold. The initial classification result is obtained by inputting the target text into the large language model. The text enhancement module is configured to determine the associated text and sample example data corresponding to the target text when the first classification result and the second classification result are inconsistent. The associated text and the target text originate from the same initial text. The associated text is a context fragment that is related to the target text. The sample example data is determined from a sample database. The target determination module is configured to input the target text, the associated text, and the sample example data into the large language model, obtain the text classification result output by the large language model, and determine the text classification result as the target classification result of the target text.

12. A computing device, comprising: include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 9.

13. A computer-readable storage medium, characterized in that, It stores a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 9.

14. A computer program product, characterised in that, Includes a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 9.

Images