A method and system for chat topic identification based on text tag prediction

Abstract

This invention proposes a chat topic recognition method and system based on text tag prediction. The method includes the following steps: acquiring and preprocessing the chat data information to be processed; processing the preprocessed chat data information using a trained deep learning model, including sentence encoding and sequence labeling; and identifying specific content from the chat data information using the deep learning model and further processing it. By introducing Sentence Bert to extract semantic features from chat statements, and using sentences as semantic units input into the model, the length limit of the feature labeling network is greatly expanded. The feature labeling method is used to label behaviors commonly seen in online fraud, ensuring that even if the wording is updated, the truly crucial inducement behaviors remain largely unchanged. By modifying the convolutional neural network model, the model can perform sequence labeling tasks while retaining the high inference performance of the convolutional neural network.

Description

Technical Field

[0001] This invention belongs to the field of electronic data forensics technology, specifically relating to a method and system for identifying chat topics based on text tag prediction. Background Technology

[0002] Currently, instant messaging software, primarily represented by QQ and WeChat, is widely used. While providing convenient information sharing among users, it has also become a tool for some criminals to commit crimes. In recent years, telecommunications fraud cases have been on the rise. Criminals use online chat and other means to fabricate false information and set up scams to defraud victims. In order to quickly sift through massive amounts of instant messaging data to find key content and important clues, it is necessary to use NLP text classification technology to perform semantic analysis on online chat content and identify chat content with telecommunications fraud tendencies.

[0003] Currently, the development of text classification technology includes the following aspects: Machine learning methods: Naive Bayes, Support Vector Machines, Decision Trees, etc. These methods have been widely used in the field of text classification, but their accuracy is limited; Deep learning methods: Convolutional Neural Networks, Recurrent Neural Networks, etc. Compared with machine learning methods, they have significantly improved in accuracy; Pre-trained models: These mainly refer to models trained on large-scale corpora, such as BERT and GPT, which are fine-tuned to adapt to text classification tasks and have further improved in accuracy.

[0004] Unlike traditional crimes, online and telecommunications fraud is highly deceptive. Criminals tailor their messages to different groups, employing a wide variety of themes that are rapidly updated. Developing text classification models is costly and time-consuming, making it difficult to keep pace with real-world needs.

[0005] Furthermore, online scam chat sessions often last a long time, and the length of the chat text is unpredictable. Text classification models using deep learning methods or pre-trained models must impose certain limits on the length of the input content, which may cause the model to miss a large amount of crucial information.

[0006] Therefore, it is very meaningful to propose a chat topic recognition method and system based on text label prediction. Summary of the Invention

[0007] To address the existing problems, this invention provides a chat topic recognition method and system based on text tag prediction, thereby resolving the aforementioned technical deficiencies.

[0008] In a first aspect, the present invention proposes a chat topic identification method based on text tag prediction, which includes the following steps:

[0009] In response to acquiring and preprocessing the chat data information that needs to be processed;

[0010] The pre-processed chat data information is processed using a trained deep learning model, including sentence encoding and sequence labeling;

[0011] The deep learning model identifies the specific content needed from the chat data and performs further processing.

[0012] Preferably, the sentence encoding process includes: using the BERT-based network structure Sentence BERT, which consists of three parts: a BERT pre-trained model, a max-pooling layer, and a sentence vector output layer (Sentencerepresentation). Each speech content is converted into a sentence vector of a specified dimension, and then the sentence vector is used as the input for the sequence labeling part.

[0013] More preferably, the sentence encoding process specifically includes:

[0014] The pre-trained BERT model uses a multi-layer transformer encoder. Each encoder consists of a multi-head self-attention mechanism and a feedforward network. The output of each encoder is used as the input of the next encoder. The final output matrix has dimensions [B,L,H], where B represents the number of sentences imported into the model in a batch, L represents the maximum length of the input sentences, and H represents the dimension of the encoder.

[0015] Max-pooling samples the output of the BERT pre-trained layer, retains the most significant semantic features, reduces the number of parameters, prevents model overfitting, and transforms the feature matrix of each sentence from two dimensions to one dimension, which can be represented as [B,L].

[0016] The sentence representation output layer transforms the output vector of the max pooling layer into a vector of a specified dimension as the final sentence vector. The final output matrix has dimensions [B, E], where E represents the dimension of the sentence vector.

[0017] Preferably, sequence labeling processing includes:

[0018] The traditional convolutional neural network model is modified to enable it to perform entity annotation tasks.

[0019] Semantic features are extracted using convolutional kernels, and the convolutional pooling results at the same position are concatenated. The dimension of each feature matrix is ​​[B, S, C].

[0020] After multiple convolutional pooling operations, the feature matrix dimension is adjusted to [B,L,N] through a fully connected layer, where N is equal to the number of features required for the specific content, thus obtaining the most likely label for each sentence.

[0021] Further preferably, the dimension of each layer of feature matrix is ​​[B, S, C], specifically including:

[0022] B represents the amount of data imported into the model in batches. If the online chat between two users is defined as one dialogue scenario, then if B equals 128, it means that 128 dialogue scenarios are imported in each batch for parallel analysis.

[0023] S represents the input length of each scenario. If S is defined as 1024, it means that a maximum of 1024 messages can be imported for analysis in each scenario.

[0024] C represents the number of convolutional kernels in each layer of the convolutional neural network. As a hyperparameter of the model, the number of convolutional kernels in each layer can be freely set.

[0025] Preferably, identifying the specific content required and performing further processing includes:

[0026] The relevant content is arranged in chronological order to further determine the specific category of the chat content and predict the outcome.

[0027] Secondly, embodiments of the present invention also provide a chat topic recognition system based on text tag prediction, the system comprising:

[0028] The acquisition module is configured to acquire and preprocess the chat data information that needs to be processed.

[0029] A deep learning model module is configured to process the preprocessed chat data information using a trained deep learning model, including sentence encoding and sequence labeling.

[0030] The processing module is configured to identify specific content from the chat data and arrange it in chronological order, further determine the specific category of the chat content, and predict the result.

[0031] Thirdly, embodiments of the present invention provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect.

[0032] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any of the implementations of the first aspect.

[0033] Compared with the prior art, the beneficial results of the present invention are as follows:

[0034] (1) By introducing Sentence Bert to extract semantic features from chat messages and using sentences as semantic units to input into the model, the length limit of the feature labeling network is greatly expanded; by using the feature labeling method, the behaviors that often appear in online fraud are labeled respectively. Even if the wording is updated, the truly key inducement behaviors will not change much; by modifying the convolutional neural network model, the model can perform sequence labeling tasks and retain the high inference performance of the convolutional neural network.

[0035] (2) The model of the present invention can adapt well to the analysis needs of massive chat data and quickly identify chat content with characteristics of online fraud. The model can not only adapt to the situation of emerging online fraud topics, but also accurately determine whether the fraudulent action has caused substantial harm to the parties involved. Attached Figure Description

[0036] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many anticipated advantages of the embodiments will be readily recognized as they become better understood through reference to the following detailed description. Elements in the drawings are not necessarily to scale. The same reference numerals refer to corresponding similar parts.

[0037] Figure 1 This is a flowchart illustrating the chat topic recognition method based on text tag prediction, as an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the SentenceBert model structure in the chat topic recognition method based on text tag prediction according to an embodiment of the present invention.

[0039] Figure 3 This is a schematic diagram of the feature annotation network structure in the chat topic recognition method based on text tag prediction according to an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the architecture of a chat topic recognition system based on text tag prediction, according to an embodiment of the present invention.

[0041] Figure 5 This is a schematic diagram of the structure of a computer device suitable for implementing electronic devices according to embodiments of the present invention. Detailed Implementation

[0042] In the following detailed description, reference is made to the accompanying drawings, which form part of the detailed description and are illustrated by specific illustrative embodiments in which the invention may be practiced. In this regard, directional terms such as “top,” “bottom,” “left,” “right,” “up,” “down,” etc., are used with reference to the orientation of the described figures. Because components of the embodiments can be positioned in several different orientations, directional terms are used for illustrative purposes and are by no means limiting. It should be understood that other embodiments may be utilized or logical changes may be made without departing from the scope of the invention. Therefore, the following detailed description should not be taken in a limiting sense, and the scope of the invention is defined by the appended claims.

[0043] Unlike traditional crimes, online and telecommunications fraud is highly deceptive. Criminals tailor their messages to different groups, employing a wide variety of themes that are rapidly updated. Developing text classification models is costly and time-consuming, making it difficult to keep pace with real-world needs.

[0044] Furthermore, online scam chat sessions often last a long time, and the length of the chat text is unpredictable. Text classification models using deep learning methods or pre-trained models must impose certain limits on the length of the input content, which may cause the model to miss a large amount of crucial information.

[0045] To address the two major challenges mentioned above, this invention discloses a new model structure that provides a more granular classification of the characteristics of online fraud. It eliminates the need to set corresponding models or labels for different fraud schemes. Furthermore, it utilizes the Sentence Bert model to extract the semantic encoding of each statement and inputs it into the deep learning model, greatly expanding the text input length.

[0046] In a first aspect, embodiments of the present invention disclose a chat topic identification method based on text tag prediction, such as... Figure 1 As shown, the method includes the following steps:

[0047] S1. Responds to obtaining the chat data information that needs to be processed and preprocessing it;

[0048] S2. The pre-processed chat data information is processed using a trained deep learning model, including sentence encoding and sequence labeling;

[0049] S3. Identify the specific content required from the chat data information using the deep learning model and perform further processing.

[0050] Specifically, common characteristics of online scams include: flirtatious chat, invitations to place orders, video chats, providing bank card information, transactions and transfers, clicking on web links, etc. A set of feature tags for online scams should be designed based on these key characteristics. Scammers typically need to guide potential victims to complete one or more specific actions, each guided action being conducted through multiple rounds of dialogue. To conceal their true intentions, the content of the conversations during these guided actions is usually very similar to ordinary chat.

[0051] Therefore, we cannot simply label the entire conversation as online fraud or a specific type of online fraud. Instead, we need to use a sequence labeling scheme to label each message individually. See Table 1 below:

[0052]

[0053] Table 1 Sequence Labeling Instructions

[0054] For example, the characteristic tag for a video request is "video". When a message requests a video chat, this message is labeled B-video; when a message indicates that the video chat has ended, this message is labeled E-video; and multiple messages in between are labeled I-video. At this point, the entire guiding action is complete, and the fraudsters have completed a phase of their mission.

[0055] Based on the above rules, chat content is labeled, with each message corresponding to a tag. The labeling method is shown in Table 2 below:

[0056]

[0057] Table 2 Data Labeling Method

[0058] Each chat scenario typically contains hundreds of messages, including spoken content, system messages, and shared links. We treat all the information in a chat scenario as a single data point and input it into the model. Finally, the model outputs the label corresponding to each message.

[0059] Furthermore, the model structure in the embodiments of the present invention will be described in detail below. Based on the specific characteristics of the task scenario, the embodiments of the present invention design a special deep learning model, which is divided into two main parts: a sentence encoding part and a sequence labeling part.

[0060] 1) Sentence encoding part

[0061] The embodiments of this invention mainly adopt a network structure based on BERT (Sentence BERT). Sentence BERT consists of three parts: a pre-trained BERT model, a max-pooling layer, and a sentence representation output layer.

[0062] The BERT pre-trained model employs a multi-layer transformer encoder, with each encoder consisting of a multi-head self-attention mechanism and a feedforward network. The output of each encoder serves as the input to the next encoder. The final output matrix of this part has dimensions [B,L,H], where B represents the number of sentences imported into the model in a batch, L represents the maximum length of the input sentences, and H represents the dimension of the encoder.

[0063] The main function of the max pooling layer is to sample the output of the BERT pre-trained layer, retain the most significant semantic features, reduce the number of parameters, and prevent the model from overfitting. At the same time, the max pooling layer can transform the feature matrix of each sentence from two dimensions to one dimension, specifically [B,L].

[0064] The main function of the sentence vector output layer is to transform the output vector of the max pooling layer into a vector of a specified dimension as the final sentence vector. The final output matrix has dimensions [B, E], where E represents the dimension of the sentence vector.

[0065] Sentence Bert's model structure is as follows: Figure 2 As shown, Sentence BERT is used to convert each statement into a sentence vector of a specified dimension, and then the sentence vector is used as input for the sequence labeling part.

[0066] 2) Sequence labeling section

[0067] In this embodiment, the traditional convolutional neural network model is modified to enable it to perform entity annotation tasks.

[0068] Traditional convolutional neural networks (CNNs) are designed for classification tasks. Their primary function is to extract local features from data using numerous convolutional kernels, then reduce the dimensionality of these features to decrease the resources required for model inference. In a traditional CNN structure, after all convolutions and pooling are completed, the model simply concatenates the feature vectors obtained from convolution and pooling before entering a fully connected layer, ultimately outputting a one-dimensional vector.

[0069] Therefore, convolutional kernels are continued for semantic feature extraction, and then the convolutional pooling results at the same position are concatenated. The dimension of each feature matrix is ​​[B, S, C]. B represents the amount of data imported into the model in batches. We define the online chat between two users as one dialogue scenario. If B equals 128, it means that 128 dialogue scenarios are imported in batches for parallel analysis. S represents the input length of each scenario. If S equals 1024, it means that a maximum of 1024 messages are imported into each scenario for analysis. C represents the number of convolutional kernels in each layer of the convolutional neural network. As a hyperparameter of the model, the number of convolutional kernels in each layer can be set freely.

[0070] After multiple convolutional pooling operations, the feature matrix dimensions are adjusted to [B, L, N] via a fully connected layer, where N is equal to the number of network fraud features, resulting in the most likely label for each sentence. The feature annotation network structure diagram is shown below. Figure 3 As shown.

[0071] Furthermore, by using models to quickly identify content related to online fraud from chat data and arranging the relevant content in chronological order, it is possible to determine whether the chat content belongs to a certain type of fraud, and also to accurately determine whether the fraudulent operation was successful and whether the victim suffered economic losses.

[0072] For example, if the model identifies information such as ambiguous topics, video chat, threats, and money transfers, it can determine that the suspect has committed nude chat fraud against the victim, and the victim is likely to suffer financial losses.

[0073] The outcome processing is freely editable. When new scams emerge, new outcome processing templates can be created, enabling rapid transfer of model capabilities. For example, romance scams often involve ambiguous topics and money transfers. Compared to nude chat scams, money transfers are more frequent, and video chats and threats are absent. Simply add the romance scam template to the outcome processing template; no model adjustments are needed.

[0074] Specifically, online fraud is characterized by a large amount of chat content, relatively dispersed criminal characteristics, a certain degree of deception, and frequently updated scripts. To address these challenges, the technical solution of this invention incorporates the following innovations:

[0075] 1. By introducing Sentence Bert to extract semantic features from chat messages, and using sentences as semantic units input into the model, the length limit of the feature annotation network is greatly expanded.

[0076] 2. By using feature annotation methods, we can identify common behaviors in online fraud. Even if the script is updated, the key inducement behaviors will not change significantly.

[0077] 3. By modifying the convolutional neural network model, the model is able to perform sequence labeling tasks while retaining the high inference performance of the convolutional neural network.

[0078] Through the above adjustments, the model can effectively adapt to the analysis needs of massive amounts of chat data and quickly identify chat content with characteristics of online fraud. This model not only adapts to the ever-evolving nature of online fraud topics but also accurately determines whether fraudulent activities have caused substantial harm to the individuals involved.

[0079] Secondly, embodiments of the present invention also disclose a chat topic recognition system based on text tag prediction, such as... Figure 4 As shown, the system includes: an acquisition module 41, a deep learning model module 42, and a processing module 43.

[0080] In one specific embodiment, the acquisition module 41 is configured to acquire the chat data information to be processed and preprocess it; the deep learning model module 42 is configured to use a trained deep learning model to process the preprocessed chat data information, including sentence encoding and sequence labeling; and the processing module 43 is configured to identify specific content from the chat data information and arrange it in chronological order, further determine the specific category of the chat content and predict the result.

[0081] The following is for reference. Figure 5 It illustrates an electronic device suitable for implementing embodiments of the present invention (e.g., Figure 1 The diagram shows the structure of a computer device 600 (a server or terminal device). Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0082] like Figure 5 As shown, the computer device 600 includes a central processing unit (CPU) 601 and a graphics processing unit (GPU) 602, which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 603 or programs loaded from storage section 609 into random access memory (RAM) 604. The RAM 604 also stores various programs and data required for the operation of the device 600. The CPU 601, GPU 602, ROM 603, and RAM 604 are interconnected via a bus 605. An input / output (I / O) interface 606 is also connected to the bus 605.

[0083] The following components are connected to I / O interface 606: an input section 607 including a keyboard, mouse, etc.; an output section 608 including an LCD, speakers, etc.; a storage section 609 including a hard disk, etc.; and a communication section 610 including a network interface card, such as a LAN card or modem. The communication section 610 performs communication processing via a network such as the Internet. A drive 611 may also be connected to I / O interface 606 as needed. A removable medium 612, such as a hard disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 611 as needed so that computer programs read from it can be installed into storage section 609 as needed.

[0084] In particular, according to embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 610, and / or installed from removable medium 612. When the computer program is executed by central processing unit (CPU) 601 and graphics processing unit (GPU) 602, the functions defined in the methods of this invention are performed.

[0085] It should be noted that the computer-readable medium described in this invention can be a computer-readable signal medium, a computer-readable medium, or any combination thereof. A computer-readable medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, apparatus, or any combination thereof. More specific examples of a computer-readable medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution apparatus, device, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than a computer-readable medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution device, apparatus, or apparatus. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0086] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0087] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using dedicated hardware-based devices that perform the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0088] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be located in a processor.

[0089] In another aspect, the present invention also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: acquire and preprocess chat data information to be processed; process the preprocessed chat data information using a trained deep learning model, including sentence encoding and sequence labeling; and identify specific content required from the chat data information using the deep learning model and perform further processing.

[0090] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A chat topic identification method based on text tag prediction, characterized in that, The method includes the following steps: In response to acquiring and preprocessing the chat data information that needs to be processed; The pre-processed chat data information is processed using a trained deep learning model, including sentence encoding and sequence labeling; The deep learning model identifies the specific content required from the chat data and performs further processing. The sentence encoding process includes: using the BERT-based network structure Sentence BERT, which consists of three parts: a pre-trained BERT model, a max-pooling layer, and a sentence vector output layer (Sentencerepresentation). Each speech content is converted into a sentence vector of a specified dimension, and then the sentence vector is used as the input for the sequence labeling part. Specifically, this includes: The pre-trained BERT model employs a multi-layer transformer encoder, with each encoder consisting of a multi-head self-attention mechanism and a feedforward network. The output of each encoder serves as the input to the next encoder, resulting in a final output matrix with dimensions [B, L, H], where B represents the number of sentences imported into the model in a batch, L represents the maximum length of the input sentences, and H represents the encoder dimension; the max-pooling layer samples the output of the BERT pre-trained layers, retaining the most significant semantic features, reducing the number of parameters, and preventing model overfitting, transforming the feature matrix of each sentence from two dimensions to one dimension, specifically represented by dimensions [B, L]; the sentence representation layer transforms the output vector of the max-pooling layer into a vector of a specified dimension, serving as the final sentence vector, with the final output matrix having dimensions [B, E], where E represents the dimension of the sentence vector; Sequence labeling processing includes: modifying traditional convolutional neural network models to enable them to perform entity labeling tasks; extracting semantic features using convolutional kernels, concatenating the convolutional pooling results at the same positions, with the feature matrix of each layer having dimensions [B, S, C]; where S represents the input length of each scene and C represents the number of convolutional kernels in each layer of the convolutional neural network; after multiple convolutional pooling operations, the feature matrix dimensions are finally adjusted to [B, L, N] through a fully connected layer, where N is equal to the number of features required for the specific content, resulting in the most likely label for each sentence.

2. The chat topic recognition method based on text tag prediction according to claim 1, characterized in that, Each layer of the feature matrix has dimensions [B, S, C], specifically including: B represents the amount of data imported into the model in batches. If the online chat between two users is defined as one dialogue scenario, then if B equals 128, it means that 128 dialogue scenarios are imported in each batch for parallel analysis. S represents the input length of each scenario. If S is defined as 1024, it means that a maximum of 1024 messages can be imported for analysis in each scenario. C represents the number of convolutional kernels in each layer of the convolutional neural network. As a hyperparameter of the model, the number of convolutional kernels in each layer can be freely set.

3. The chat topic recognition method based on text tag prediction according to claim 1, characterized in that, Identifying the specific content required and performing further processing includes: The relevant content is arranged in chronological order to further determine the specific category of the chat content and predict the outcome.

4. A chat topic recognition system based on text tag prediction, characterized in that, The system, applicable to the method as described in any one of claims 1-3, comprises: The acquisition module is configured to acquire and preprocess the chat data information that needs to be processed. A deep learning model module is configured to process the preprocessed chat data information using a trained deep learning model, including sentence encoding and sequence labeling. The processing module is configured to identify specific content from the chat data and arrange it in chronological order, further determine the specific category of the chat content, and predict the result.

5. An electronic device, comprising: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 3.

Images