Topic material detection method and device, equipment, storage medium and product

Abstract

The invention provides a topic material detection method, device and equipment, a computer readable storage medium and a product. The method comprises the steps that topic information and material content of a first topic material to be detected and user feature information corresponding to the first topic material are obtained, and the topic information comprises a topic label carried in the first topic material; the user feature information comprises account information of an account issuing the first topic material and material features of historical issuing materials of the account; based on the topic information and the material content of the first topic material and the user feature information, detecting whether the first topic material is a hot spot material; the hot-spot material is a topic material of which the topic label and the material content are not related. According to the invention, the detection accuracy of hot spot materials can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer processing technology, and in particular to a method, device, equipment, storage medium and product for detecting topic materials. Background Art

[0002] With the continuous development of self-media, the number of users participating in the release of blog posts and other materials is also increasing. However, when releasing materials, some users will add current hot topics to materials that are unrelated to hot topics in an attempt to increase the exposure and traffic of the released materials. Such materials are called hot topics materials. Since the content of such materials is actually unrelated to the hot topics included in the materials, if such materials are recommended to users as hot topic materials, it will affect the user experience.

[0003] At present, media platforms usually judge whether a material is a hot topic based on the degree of relevance between the material and the topic it covers, but the accuracy of the judgment is low. Summary of the invention

[0004] The present application provides a method, device, equipment, computer-readable storage medium and product for detecting topic materials, which can improve the detection accuracy of hot topic materials.

[0005] In a first aspect, the present application provides a method for detecting topic materials, comprising: obtaining topic information and material content of a first topic material to be detected, and user characteristic information corresponding to the first topic material, the topic information including topic tags carried in the first topic material, and the user characteristic information including account information of an account that publishes the first topic material, and material characteristics of materials historically published by the account; based on the topic information and material content of the first topic material, and the user characteristic information, detecting whether the first topic material is a hot topic material; a hot topic material refers to a topic material where the topic tag and material content are unrelated.

[0006] In the second aspect, the present application provides a device for detecting topic materials, which includes: a first acquisition module, configured to obtain topic information and material content of a first topic material to be detected, and user characteristic information corresponding to the first topic material, the topic information including topic tags carried in the first topic material, the user characteristic information including account information of an account that publishes the first topic material, and material characteristics of materials historically published by the account; a first detection module, configured to detect whether the first topic material is a hot topic material based on the topic information and material content of the first topic material, and user characteristic information; a hot topic material refers to a topic material where the topic tag and material content are unrelated.

[0007] In a third aspect, the present application provides an electronic device, comprising: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided by the present application.

[0008] In a fourth aspect, the present application provides a computer-readable storage medium storing executable instructions for implementing the method provided in the present application when the executable instructions are executed by a processor.

[0009] In a fifth aspect, the present application provides a computer program product, including a computer program or instructions, which are used to implement the method provided by the present application when the computer program or instructions are executed by a processor.

[0010] This application has the following beneficial effects:

[0011] In this application, due to the account information of the account that publishes the first topic material, as well as the material characteristics of the materials published by the account that published the first topic material in the past, and other user characteristic information, it is possible to reflect the material publishing habits of the account owner and the types of materials that are likely to be published to a certain extent. Therefore, based on the material content and topic information of the first topic material itself, the account information of the account that published the first topic material, and the material characteristics of the materials published by the account in the past, it is determined whether the first topic material is a hot topic material, which can improve the detection accuracy of hot topic materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the architecture of a topic material detection system provided in an embodiment of the present application;

[0013] Figure 2 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0014] Figure 3 It is a flow chart of a method for detecting topic materials provided in an embodiment of the present application;

[0015] Figure 4 It is a structural diagram of a sensitive information identification module provided in an embodiment of the present application;

[0016] Figure 5 It is a flowchart of a method for determining material type based on a tree model provided in an embodiment of the present application;

[0017] Figure 6 is a flow chart of a method for determining a material type based on a second material classification model provided in an embodiment of the present application;

[0018] Figure 7 is a schematic diagram of a process for determining the material type of a second material based on a RoBERTa model provided in an embodiment of the present application;

[0019] Figure 8 It is a flowchart of a model training method provided in an embodiment of the present application;

[0020] Fig. 9 It is a flow chart of another method for detecting topic materials provided in an embodiment of the present application;

[0021] Fig.10 It is a flow chart of another method for detecting topic materials provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0023] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it can be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0024] If similar descriptions of "first / second" appear in the application documents, the following explanation is added. In the following description, the terms "first / second / third" involved are merely to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first / second / third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0026] With the continuous development of self-media, the number of users participating in the release of blog posts and other materials is also increasing. However, when releasing materials, some users will add topic tags of current hot topics to materials that are not related to hot topics in an attempt to increase the exposure and traffic of the released materials. Since the content of such materials is actually unrelated to the topic tags carried in the materials, such materials can be understood as hot topic materials.

[0027] At present, in order to filter out hot-topic materials from materials posted by users, media platforms usually determine whether a material is a hot-topic material based on the correlation between the material content and the topic tags. However, the judgment accuracy of this method is low.

[0028] To solve the above problems, the embodiments of the present application provide a method, device, equipment, computer-readable storage medium and product for detecting topic materials, which can more accurately screen out hot topic materials.

[0029] The following describes an exemplary application of the electronic device provided by the embodiment of the present application. The electronic device provided by the embodiment of the present application can be a user terminal of various types such as a laptop computer, a tablet computer, a desktop computer, a mobile device (e.g., a mobile phone, a wearable smart watch, a dedicated messaging device), and can also be implemented as a server. The following describes an exemplary application when the electronic device is implemented as a server.

[0030] See also Figure 1 , Figure 1 1 is a schematic diagram of the architecture of a topic material detection system 100 provided in an embodiment of the present application. To implement the topic material detection method in the embodiment of the present application, electronic devices 400-1 and 400-2 ( Figure 1 Only two electronic devices are shown, but this does not constitute a limitation on the number of electronic devices included in the detection system for topic materials in the embodiment of the present application) The server 200 is connected via a network 300, and the network 300 can be a wide area network or a local area network, or a combination of the two.

[0031] In one embodiment, a user may publish a first topic material on electronic devices 400-1 and 400-2. Accordingly, the server 200 may receive the first topic material sent by the electronic devices 400-1 and 400-2 based on the user input, and store the received first topic material in the database 500. Afterwards, the server 200 may obtain the topic information and material content of the first topic material, and the user characteristic information corresponding to the first topic material, wherein the topic information includes the topic tag carried in the first topic material, and the user characteristic information includes the account information of the account that publishes the first topic material and the material characteristics of the historically published materials of the account; and based on the topic information and material content of the first topic material, and the user characteristic information, detect whether the first topic material is a hot topic material, wherein the hot topic material refers to a topic material that is unrelated between the topic tag and the material content. After determining that the first topic material is a hot topic material, the server 200 may send the first topic material to the manual review platform for manual review. When the server 200 determines that the first topic material is not a hot topic material, the server 200 can send the first topic material to the electronic device 400-1 and the electronic device 400-2 through the network, so as to recommend the first topic material to the user through the electronic device 400-1 and the electronic device 400-2, so that the user can consume the first topic material through the electronic device.

[0032] In some embodiments, the server 200 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The electronic device 400 may be an electronic device with a display screen, such as a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart watch, etc., but is not limited thereto. The electronic device and the server may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present invention.

[0033] Figure 2 It is a schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present application. Figure 2 The electronic device 400 shown includes: at least one processor 410, a memory 450, at least one network interface 420 and a user interface 430. The various components in the electronic device 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 440 is not described in detail. Figure 2 Various buses are labeled as bus system 440 .

[0034] Processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0035] The user interface 430 includes one or more output devices 431 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0036] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0037] The memory 450 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0038] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0039] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0040] A network communication module 452, for reaching other computing devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 include: Bluetooth, wireless compatibility certification (WiFi), and universal serial bus (USB), etc.;

[0041] a presentation module 453 for enabling presentation of information via one or more output devices 431 (e.g., display screen, speaker, etc.) associated with the user interface 430 (e.g., a user interface for operating peripherals and displaying content and information);

[0042] The input processing module 454 is used to detect one or more user inputs or interactions from one of the one or more input devices 432 and translate the detected inputs or interactions.

[0043] In some embodiments, the topic material detection device 455 provided in the embodiments of the present application can be implemented in a software manner. Figure 2 The detection device 455 of topic materials stored in the memory 450 is shown, which can be software in the form of programs and plug-ins, including the following software modules: a first acquisition module 4551 and a first detection module 4552. These modules are logical, so they can be arbitrarily combined or further split according to the functions implemented. The functions of each module will be explained below.

[0044] In other embodiments, the detection device 455 of the topic material provided in the embodiment of the present application can be implemented in hardware. As an example, the detection device 455 of the topic material provided in the embodiment of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the detection method of the topic material provided in the embodiment of the present application. For example, the processor in the form of a hardware decoding processor can adopt one or more application specific integrated circuits (application specific integrated circuit, ASIC), DSP, programmable logic device (programmable logic device, PLD), complex programmable logic device (complex programmable logic device, CPLD), field programmable gate array (field-programmable gate array, FPGA) or other electronic components.

[0045] The following will describe the topic material detection method provided by the embodiment of the present application in combination with the exemplary application and implementation of the electronic device provided by the embodiment of the present application. In addition, hereinafter, for the convenience of description, the topic material detection device is referred to as the detection device.

[0046] See also Figure 3 , Figure 3 This is a flow chart of a method for detecting topic materials provided in an embodiment of the present application. Figure 3 The steps shown illustrate the method for detecting topic materials provided in the embodiments of the present application.

[0047] In step S301: the topic information and material content of the first topic material to be detected, and the user feature information corresponding to the first topic material are obtained.

[0048] It is understandable that the detection device can obtain the topic information and material content of the first topic material before recommending the first topic material to the user. In addition, the detection device can also obtain the user feature information of the user who publishes the first topic material to obtain the user feature information corresponding to the first topic material.

[0049] In some embodiments, the topic information includes but is not limited to the following included in the first topic material: topic tags, the number of topic tags, the number of historically repeated topic tags in the topic tags, the number of repeated words between different topic tags, whether the topic tags contain authentication information, etc.

[0050] It should be noted that the historical repeated topic tags may be topic tags existing in the historical published materials of the account that published the first topic material.

[0051] In some embodiments, the material content includes, but is not limited to: text and pictures included in the first topic material.

[0052] It should be noted that the pictures included in the first topic material may be pictures directly carried in the first topic material, or pictures extracted from the video carried by the first topic material, and this embodiment of the present application is not limited to this.

[0053] In some embodiments, the user characteristic information includes, but is not limited to: account information of the account that publishes the first topic material and material characteristics of the account's historically published materials.

[0054] In some embodiments, the account information of the account that publishes the first topic material includes, but is not limited to: account level, duration from account registration time to current time, account name, etc.

[0055] In some embodiments, the material characteristics of the historical published materials of the account that published the first topic material include at least one of the following: the total number of historical published materials of the account; the number of hot-spot materials in the historical published materials of the account; the number of topic materials in the historical published materials of the account; the number of ordinary materials in the historical published materials of the account, where ordinary materials are materials that do not carry topic tags; the number of repeated topic tags carried in the historical published materials of the account; the average number of topic tags for each historical published material of the account; the number of valid characters for each historical published material of the account, etc.

[0056] In some embodiments, the number of valid characters of each historical release material may be the number of characters excluding emoticons, pictures, and topic tags in each historical release material.

[0057] In some embodiments, the material characteristics of the historically published materials acquired by the detection device may be the material characteristics of the materials published by the account that published the first topic material within a preset time period of historical time.

[0058] Exemplarily, the detection device can obtain the material characteristics of all materials published by the account that published the first topic material before the current moment, and the material characteristics of materials published by the account that published the first topic material within one year, half a year or one month before the current moment.

[0059] In step S302, based on the topic information and material content of the first topic material, as well as the user characteristic information, it is detected whether the first topic material is a hot topic material.

[0060] It is understandable that after obtaining the topic information and material content of the first topic material, as well as the user characteristic information corresponding to the first topic material, the detection device can determine whether the topic tags carried in the first topic material are related to the material content based on the topic information and material content, as well as the user characteristic information corresponding to the first topic material. If the detection device determines that the topic tags carried in the first topic material are not related to the material content, then the first topic material is determined to be a hot topic material.

[0061] In some embodiments, the irrelevance between the topic tag and the material content can be understood as the correlation between the topic tag and the material content is less than the second threshold.

[0062] In some embodiments, the detection device may input the topic information and material content of the first topic material, as well as user feature information corresponding to the first topic material, into a trained material classification model to determine whether the first topic material is a hot topic material through the material classification model.

[0063] In some embodiments, the material classification model can be any network model that can determine whether a topic material is a hot topic material, and the embodiments of the present application are not limited to this.

[0064] In some embodiments, the material classification model can be a neural network model or a combination of multiple neural network models, which is not limited in the embodiments of the present application.

[0065] In some embodiments, the material classification model may include a first material classification model and a second material classification model, wherein the first material classification model is used to detect the probability that a topic material is a non-hotspot material, and the second material classification model is used to determine whether a topic material is a hotspot material or a non-hotspot material. Among them, non-hotspot materials are topic materials that are related between material content and topic tags. Based on this, the detection device can first determine the probability that the first topic material is a non-hotspot material based on the first material classification model. When the detection device determines that the first topic material does not meet the set hotspot material probability threshold based on the probability output by the first material classification model, it can further determine whether the first topic material is a hotspot material based on the second material classification model.

[0066] In some embodiments, when the detection device determines that the first topic material is a hot topic material, the detection device sends the first topic material to a manual review platform for manual review of the first topic material.

[0067] In some embodiments, when the detection device determines that the first topic material is not a hot topic material, it can recommend the first topic material to the user.

[0068] In the embodiment of the present application, due to the account information of the account that publishes the first topic material, and the material characteristics of the materials published by the account that publishes the first topic material in the past, etc., the user characteristic information can reflect the material publishing habits of the account owner and the types of materials that are likely to be published to a certain extent. Therefore, based on the material content and topic information of the first topic material itself, the account information of the account that publishes the first topic material, and the material characteristics of the materials published by the account in the past, it is determined whether the first topic material is a hot topic material, which can improve the detection accuracy of hot topic materials.

[0069] In some embodiments, before step S301, the method further includes: detecting whether sensitive information exists in the first topic material; and confirming that sensitive information does not exist in the first topic material.

[0070] It is understandable that before the detection device determines whether the first topic material is a hot topic material based on the topic information and material content of the first topic material, as well as the user characteristic information corresponding to the first topic material, it may first determine whether there is sensitive information in the first topic material. If the detection device confirms that there is sensitive information in the first topic material, it means that the first topic material is not suitable for recommendation to the user. At this time, the detection device may send the first topic material to the manual review platform to manually review whether there is sensitive information in the first topic material. If the detection device confirms that there is no sensitive information in the first topic material, it may determine whether the first topic material is a hot topic material based on the topic information and material content of the first topic material, as well as the user characteristic information corresponding to the first topic material.

[0071] In some embodiments, the sensitive information in the first topic material includes, but is not limited to, any words, phrases, or pictures in the first topic material that violate the law and / or violate morality.

[0072] Exemplarily, the sensitive information in the first topic material may be words and phrases involving pornography, gambling, etc. in the text included in the first topic material and images involving pornography, gambling, etc. in the pictures included in the first topic material.

[0073] In some embodiments, after obtaining the text included in the first topic material, the detection device may determine whether the text included in the first topic material contains words or phrases in the sensitive word library. When the detection device determines that the text included in the first topic material contains words or phrases in the sensitive word library, it confirms that sensitive information exists in the first topic material.

[0074] In some embodiments, after obtaining the pictures included in the first topic material, the detection device may identify the text in the pictures included in the first topic material, and determine whether the identified text contains words or phrases in the sensitive word library. When the detection device determines that the identified text contains words or phrases in the sensitive word library, it confirms that sensitive information exists in the first topic material. In addition, the detection device may also identify the pictures included in the first topic material to determine whether the pictures involve pornography. When the material processing device determines that the pictures included in the first topic material involve pornography, it confirms that sensitive information exists in the first topic material.

[0075] In some embodiments, upon confirming that sensitive information exists in the first topic material, the detection device may send the first topic material to a manual review platform for manual review.

[0076] In some embodiments, the method for detecting whether sensitive information is present in the material content of the first topic material also includes: identifying the pictures included in the first topic material based on the trained image recognition model to obtain a picture recognition result, and the picture recognition result is used to indicate whether sensitive information is present in the pictures included in the first topic material; identifying the text included in the first topic material based on the trained text recognition model to obtain a text recognition result, and the text recognition result is used to indicate whether sensitive information is present in the text included in the first topic material; based on the picture recognition result and the text recognition result, determining whether sensitive information is present in the first topic material.

[0077] It can be understood that after obtaining the text of the first topic material, the detection device can also input the obtained text into the trained text recognition model, so as to identify whether there is sensitive information in the text included in the first topic material through the trained text recognition model, and output the recognition result. Based on this, the detection device can determine whether there is sensitive information in the text of the first topic material based on the recognition result output by the trained text recognition model. In addition, when there is a picture in the first topic material, the detection device can also input the picture in the first topic material into the trained image recognition model, so as to identify whether there is sensitive information in the picture in the first topic material through the trained image recognition model, and obtain the recognition result. Based on this, the detection device can also determine whether there is sensitive information in the picture in the first topic material based on the recognition result output by the trained image recognition model.

[0078] In some embodiments, when there are no pictures in the first topic material, the detection device can determine whether there is sensitive information in the first topic material based on the output results of the trained text recognition model. When there are pictures in the first topic material, the detection device can determine whether there is sensitive information in the first topic material based on the output results of the trained text recognition model and the output results of the trained image recognition model.

[0079] In some embodiments, before the detection device determines whether there is sensitive information in the first topic material based on the trained text recognition model and the trained image recognition model, it can also train the text recognition model and the image recognition model so that the text recognition model has the ability to recognize sensitive information from text, and the image recognition model has the ability to recognize sensitive information from pictures.

[0080] In some embodiments, the image recognition model may be a trained ViT model or other network model, which is not limited in the embodiments of the present application.

[0081] In some embodiments, the text recognition model can be a BERT model, a RoBERTa model, etc., or it can be other network models, and the embodiments of the present application are not limited to this.

[0082] In some embodiments, reference Figure 4 , Figure 4It is a structural diagram of a sensitive information identification module provided in an embodiment of the present application. Among them, the sensitive information identification module 40 includes a BERT model 401 and a ViT model 402. Based on this, the detection device can obtain pictures in the sample material to obtain sample pictures 403 for training the ViT model 402. After obtaining the sample picture 403, the detection device can input the obtained sample picture 403 into the ViT model 402 to obtain the recognition result output by the ViT model 402. Afterwards, the detection device can classify the recognition result output by the ViT model 402 through a softmax classifier, and calculate the loss value through a loss function 404 according to the category of the sample picture 403 determined by the softmax classifier and the category indicated by the label on the sample picture 403, and adjust the model parameters of the ViT model 402 according to the calculated loss value to obtain the trained ViT model.

[0083] Exemplarily, after obtaining the sample image, the detection device may adjust the dimension of the sample image to 224*224*3, and then divide the sample image into 16*16 tiles. At this time, the detection device can obtain (224*224) / (16*16)=196 tiles. Afterwards, the detection device may input the obtained tiles into the embedding layer of the ViT model to obtain a feature matrix of (196+1)*(16*16*3)=(196+1)*768 output by the embedding layer of the ViT model. Among them, +1 is the feature vector of the identifier [CLS] added to the tile header. Afterwards, the detection device may input the feature matrix of (196+1)*(16*16*3)=(196+1)*768 into the encoding (Encoder) layer of the ViT model for encoding. At this time, the detection device can obtain a feature vector with a dimension of 197*768 output by the encoding layer of the ViT model. The detection device can input the feature vector with a dimension of 1*768 at the first position in the feature vector with a dimension of 197*768 into the softmax classifier connected to the ViT model to obtain the classification result of the sample image output by the softmax classifier.

[0084] refer to Figure 4, the detection device can obtain the sample material and the text in the sample material to obtain the sample text 405 for training the image recognition model. After obtaining the sample text 405, the detection device can input the obtained sample text 405 into the BERT model 401 to obtain the recognition result output by the BERT model 401. Afterwards, the detection device can classify the output result of the BERT model 401 through the softmax classifier, and calculate the loss value through the loss function 406 according to the category of the sample text 405 determined by the softmax classifier and the category indicated by the label on the sample text 405, and adjust the model parameters of the BERT model 401 according to the calculated loss value to obtain the trained BERT model 401.

[0085] Exemplarily, the detection device inputs the sample text into the BERT model, and can obtain a feature vector with a dimension of 512*768 output by the BERT model. Afterwards, the detection device can input the feature vector with a dimension of 512*768 into a softmax classifier connected to the BERT model, and obtain the category of the sample text output by the softmax classifier. The type of the sample text output by the softmax classifier is used to indicate whether there is sensitive information in the sample text.

[0086] In some embodiments, the loss function used to train the image recognition model and the text recognition model includes, but is not limited to: a binary cross entropy loss function, a negative log-likelihood function, and the like.

[0087] In some embodiments, reference Figure 4 After completing the training of the BERT model 401 and the ViT model 402 included in the sensitive information identification module 40, the detection device can deploy the BERT model 401 and the ViT model 402 in the ONNX model format. Afterwards, when the detection device obtains the first topic material, it can call the sensitive information identification module 40 to detect the pictures and / or texts included in the first topic material to obtain a detection result indicating whether sensitive information exists in the text and / or pictures included in the first topic material.

[0088] In some embodiments, step S302 includes: in the case where the material content includes text and pictures, determining the probability that the first topic material is a non-hotspot material based on the topic information and material content in the first topic material, and user characteristic information through the trained first material classification model; in the case where it is determined based on probability that the first topic material does not meet the set hotspot material probability threshold, determining whether the first topic material is a hotspot material based on the text and pictures in the first topic material through the second material classification model.

[0089] It can be understood that the trained first material classification model can output the probability of whether a topic material is a non-hotspot material. The trained second material classification model can be a neural network model that can identify hotspot materials and non-hotspot materials. Based on this, after obtaining the topic information and material content of the first topic material, and the user feature information corresponding to the first topic material, the detection device can input the topic information and material content of the first topic material, and the user feature information corresponding to the first topic material into the trained first material classification model to detect the probability of the first topic material being a non-hotspot material through the first material classification model. At this time, the detection device can obtain the probability that the first topic material is a non-hotspot material.

[0090] When the detection device determines that the first topic material does not meet the set probability threshold of the hot topic material based on the obtained probability, the text and pictures included in the first topic material can be input into the second material classification model to detect whether the first topic material is a hot topic material through the second material classification model.

[0091] In some embodiments, when the detection device determines that the first topic material meets the set hot spot material probability threshold based on the probability that the first topic material is a non-hot spot material, it determines that the first topic material is a hot spot material and sends the first topic material to the manual review platform for manual review of the first topic material.

[0092] It should be noted that the probability threshold of hot spot materials can be set as needed, and the embodiments of the present application do not limit this.

[0093] In some embodiments, the first material classification model and the second material classification model are different. For example, the second material classification model is better than the first material classification model in identifying hot spot materials.

[0094] In some embodiments, the trained first material classification model may be a trained tree model. For example, the trained first material classification model may be an ensemble tree model, an extreme gradient boosting (XGboost) tree model, or the like.

[0095] In some embodiments, the second material classification model may be any other type of neural network model. Exemplarily, the second material classification model may be any type of deep neural network model or convolutional neural network model. For example, the second material classification model may be a multimodal large model, which is not limited in the embodiments of the present application.

[0096] In some embodiments, after obtaining the topic information and material content included in the first topic material, and the user feature information corresponding to the first topic material, the detection device can process the topic information and material content of the first topic material, and the user feature information corresponding to the first topic material through feature engineering to obtain material features that can be recognized by the trained first material classification model. Afterwards, the detection device can input the obtained material features into the trained first material classification model to detect the probability that the first topic material is a non-hotspot material through the trained first material classification model.

[0097] In some embodiments, the detection device processes the topic information and material content included in the first topic material, and the user feature information corresponding to the first topic material through feature engineering, including but not limited to: the detection device discretizes, buckets, and performs feature cross-processing on the topic information and material content included in the first topic material, and the user feature information corresponding to the first topic material.

[0098] In some embodiments, after obtaining the historical published materials of the account that published the first topic material, the detection device may also segment the time between the registration of the account and the publishing time of the first topic material to obtain multiple time periods. Afterwards, the detection device may obtain the historical published materials of the account in each time period, and process the historical published materials of the account in each time period in turn based on feature engineering to obtain the material features corresponding to each time period in the multiple time periods.

[0099] In some embodiments, after obtaining the material characteristics corresponding to each time period, the detection device can input all the obtained material characteristics, the topic information and material content of the first topic material, and the account information of the account that publishes the first topic material into the trained first material classification model to detect the probability that the first topic material is a non-hotspot material through the trained first material classification model.

[0100] In some embodiments, before the detection device detects the probability that the first topic material is a non-hotspot material through the first material classification model, the detection device may also train the first material classification model so that the first material classification model has the ability to detect the probability that a topic material is a non-hotspot material.

[0101] Exemplarily, the detection device can obtain the topic information and material content of the sample material, as well as the user characteristic information corresponding to the sample material, wherein the sample material includes hot topic sample materials and non-hot topic sample materials, the topic information of the sample material includes the topic tags carried in the sample material, and the user characteristic information corresponding to the sample material includes the account information of the account that publishes the sample material, and the material characteristics of the historically published materials of the account that publishes the sample material. Afterwards, the detection device can train the first material classification model based on the topic information and material content, as well as the user characteristic information corresponding to the sample material, to obtain the trained first material classification model.

[0102] In some embodiments, the topic information of the sample material also includes: the topic tags of the sample material, the number of topic tags of the sample material, the number of historically repeated topic tags in the topic tags of the sample material, the number of repeated words between different topic tags of the sample material, whether the topic tags of the sample material contain authentication information, etc.

[0103] It should be noted that historical repeated topic tags can be understood as topic tags that the account that published the sample material had in the materials published in the historical time.

[0104] In some embodiments, the material content of the sample material includes, but is not limited to: text and pictures included in the sample material.

[0105] It should be noted that the pictures included in the sample material may be pictures directly carried in the sample material, or may be pictures extracted from a video carried by the sample material, and this embodiment of the present application is not limited to this.

[0106] In some embodiments, the account information of the account that publishes the sample material includes but is not limited to: the account level of the account that publishes the sample material, the length of time between the registration time of the account that publishes the sample material and the current time, the account name of the account that publishes the sample material, etc.

[0107] In some embodiments, material characteristics of historical materials released by the account that releases sample materials include but are not limited to: the number of historical materials released by the account that releases sample materials, the number of historical hot materials released by the account that releases sample materials, the number of topic materials and the number of ordinary materials released by the account that releases sample materials in historical times, the number of repeated topic tags included in the topic materials released by the account that releases sample materials in historical times, the number of occurrences of each repeated topic tag in the historical materials released by the account that releases sample materials, the average number of topic tags for each material released by the account that releases sample materials in historical times, the number of valid characters in each historical material released by the account that releases sample materials, etc.

[0108] In some embodiments, the number of valid characters in a material may be the number of characters in the material excluding emoticons, pictures, and topic tags.

[0109] In some embodiments, the detection device obtains the material characteristics of the historically released materials of the account that released the sample material, which may be the material characteristics of the historically released materials of the account that released the sample material within a preset time period.

[0110] Exemplarily, the detection device may obtain material characteristics of all materials published by the account that published the sample material before the current moment, and materials published by the account that published the sample material within one year, six months, or one month before the current moment.

[0111] In some embodiments, the detection device may input the material characteristics of the sample material, the account information of the account that released the sample material, and the material characteristics of the historically released materials of the account into the first material classification model, and obtain the detection result output by the first material classification model. Afterwards, the detection device may determine the loss value based on the obtained detection result, and adjust the model parameters of the first material classification model based on the loss value until the first material classification model converges or the number of training times of the first material classification model reaches a preset number of times.

[0112] refer to Figure 5 , Figure 5 is a flow chart of a method for determining material type based on a tree model provided in an embodiment of the present application. Figure 5 The steps in this paper introduce the implementation process of determining whether the first topic material is a hot topic material based on the tree model.

[0113] In step S501, the detection device obtains a training set, which includes hot spot sample materials and non-hot spot sample materials, account information of the account that publishes the sample materials, material characteristics of historically published materials of the account that publishes the sample materials, and material characteristics of the hot spot materials in the historically published materials of the account that publishes the sample materials.

[0114] In step S502, the detection device processes the hot spot sample materials and non-hot spot sample materials, the account information of the account that publishes the sample materials, the material characteristics of the historically published materials of the account that publishes the sample materials, and the material characteristics of the hot spot materials in the historically published materials of the account that publishes the sample materials based on feature engineering to obtain sample material characteristics.

[0115] In step S503, based on the sample material features, the tree model is trained so that the tree model can detect the probability that a topic material is a non-hotspot material.

[0116] In step S504, based on the trained tree model, the probability that the first topic material is a non-hotspot material is determined, and based on the probability, it is determined whether the first topic material meets the set hotspot material probability threshold.

[0117] In some embodiments, the second material classification model includes: a large language model; wherein, through the second material classification model, based on the text and pictures in the first topic material, determining whether the first topic material is a hot topic material, including: generating relevant prompt words (prompt) indicating that the large language model determines whether the first topic material is a hot topic material; encoding the picture in the first topic material to obtain the picture features of the picture; aligning the picture features of the picture with the text features; inputting the prompt words, the text in the first topic material, and the image features aligned with the text features into the large language model to determine whether the first topic material is a hot topic material through the large language model.

[0118] It is understandable that the second material classification model may include a large language model. Based on this, when the detection device determines that the first topic material is not a hot topic material based on the first material classification model, it can obtain a picture in the first topic material and encode the picture in the first topic material to obtain the picture features of the picture, and align the picture features of the picture with the text features. After obtaining the image features aligned with the text features, the detection device can input the image features aligned with the text features, the text included in the first topic material, and the prompt words indicating the large language model to determine whether the first topic material is a hot topic material into the large language model to determine whether the first topic material is a hot topic material through the large language model.

[0119] In some embodiments, the second material classification mode further includes an image coding model, and the detection device can input the picture in the first topic material into the image coding model to encode the picture in the first topic material through the image coding model. At this time, the detection device can obtain the image features of the picture in the first topic material.

[0120] In some embodiments, after obtaining the image features aligned with the text features, the detection device can splice the image features aligned with the text features with the text features of the text included in the first topic material, and input the spliced ​​features into the large language model to determine whether the first topic material is a hot topic material through the large language model.

[0121] In some embodiments, when the detection device determines that the first topic material is a hot topic material based on the recognition result output by the large language model, the first topic material is sent to the manual review platform for manual review of the first topic material. When the detection device determines that the first topic material is not a hot topic material based on the recognition result output by the large language model, the first topic material can be recommended to the user.

[0122] In some embodiments, the image coding model may be a vision transformer (ViT), a contrastive language-image pre-training (CLIP) model, or other image coding models, which is not limited in the embodiments of the present application.

[0123] In some embodiments, the image coding model includes a coding layer and a multilayer perceptron (MLP) layer, and the detection device can input the picture included in the first topic material into the image coding model to encode the picture included in the first topic material through the coding layer of the image coding model. At this time, the detection device can obtain a feature vector with a dimension of m*768, that is, obtain the image features of the picture. Afterwards, the detection device can input the feature vector with a dimension of m*768 into the multilayer perceptron to align the image features with the text features. Wherein, m is the number of blocks segmented from the picture included in the first topic material.

[0124] It should be noted that the text in the text feature is a general term and does not refer to the text in the first topic material.

[0125] In some embodiments, before inputting the pictures in the first topic material into the image coding model, the detection device may also obtain the pictures included in the sample material and the description text of the pictures. Afterwards, the detection device may train the image coding model based on the pictures included in the sample material and the description text of the pictures, so that the image coding model has image coding capabilities.

[0126] refer to Figure 6 , Figure 6 is a flow chart of a method for determining a material type based on a second material classification model provided in an embodiment of the present application. Figure 6 , the implementation process of determining whether the first topic material is a hot topic material based on the second material classification model is introduced. Among them, the second material classification model 60 includes a ViT model 605 and a large language model 606. The detection device obtains the picture ( Figure 6The video cover 601 and accompanying picture 602), topic label 603 and text 604 in the first topic material are input into the ViT model to obtain the image features of the picture through the ViT model. Afterwards, the detection device can align the image features with the text features to obtain the image features aligned with the text features. Afterwards, the detection device can input the image features aligned with the text features, the topic label 603, the text 604 and the prompt into the large language model 606 to obtain the classification result output by the large language model 606. Among them, the classification result is used to indicate whether the first topic material is a hot topic material.

[0127] In an embodiment of the present application, the detection device detects whether the first topic material is a hot topic material based on the pictures and texts included in the first topic material through a second material classification model. Compared with detecting whether the first topic material is a hot topic material only through the text included in the first topic material, the detection accuracy of the hot topic material can be further improved.

[0128] In some embodiments, before executing step S302, the method includes: obtaining the correlation between the topic tag and the material content in the first topic material; and determining that the correlation is between a first threshold and a second threshold.

[0129] It is understandable that the correlation between topic tags and material content can reflect to a certain extent whether a topic material is a hot topic material. Therefore, after obtaining the first topic material, the detection device can calculate the correlation between the topic tags and the material content in the first topic material. When the detection device determines that the correlation between the topic tags carried by the first topic material and the material content in the first topic material is between the first threshold and the second threshold, it means that it is impossible to determine whether the first topic material is a hot topic material through the correlation between the topic tags and the material content. At this time, the detection device can determine whether the first topic material is a hot topic material based on the above step S302.

[0130] In some embodiments, after the detection device obtains the correlation between the topic tag and the material content in the first topic material, the method also includes: when the correlation is greater than a first threshold, determining that the first topic material is a non-hotspot material; when the correlation is less than a second threshold, determining that the first topic material is a hotspot material.

[0131] It can be understood that when the detection device determines that the correlation between the topic tag carried by the first topic material and the material content is greater than the first threshold, it can determine that the correlation between the topic tag and the material content is large. At this time, the detection device can determine that the first topic material is a non-hotspot material. When the detection device determines that the correlation between the topic tag carried by the first topic material and the material content is less than the second threshold, it means that the correlation between the topic tag carried by the first topic material and the material content is small or irrelevant. At this time, the detection device can determine that the first topic material is a hotspot material.

[0132] In some embodiments, when the detection device determines that the first topic material is not a hot topic material, it can recommend the first topic material to the user. When the detection device determines that the first topic material is a hot topic material, it can send the first topic material to the manual review platform for manual review of the first topic material.

[0133] In some embodiments, non-hotspot materials can be topic materials other than hotspot materials. Exemplarily, non-hotspot materials can be topic materials whose material content is unrelated to topic tags.

[0134] In some embodiments, after obtaining the first topic material, the detection device may input the first topic material into a trained third material classification model to determine the correlation between the topic tag in the first topic material and the material content through the trained third material classification model.

[0135] In some embodiments, the detection device may input the first topic material into the trained third material classification model to determine the position of the topic tag in the first topic material through the trained third material classification model, and extract the topic tag from the first topic material. Afterwards, the third material classification model may determine the correlation between the topic tag carried by the first topic material and the material content, and based on the correlation between the topic tag carried by the first topic material and the material content, determine whether the first topic material is a hot topic material.

[0136] In some embodiments, the third material classification model can be a BERT (bidirectional encoder representations from transformers) model or a RoBERTa model (robustly optimized bidirectional encoder representations from transformers approach) plus a fully connected layer, and the third material classification model can also be other types of machine learning models. This embodiment of the application is not limited to this.

[0137] In some embodiments, when the third material classification model is a RoBERTa model plus a fully connected layer, the detection device can input the first topic material into the RoBERTa model to encode the first topic material through the encoding layer of the RoBERTa model to obtain a feature vector with a dimension of 512*768. Afterwards, the RoBERTa model can also obtain the feature vector 1*768 of the topic tag from the feature vector with a dimension of 512*768 according to the position of the topic tag carried by the first topic material in the first topic material. Afterwards, the RoBERTa model can convert the 1*768 feature vector into a 1*2 feature vector through a fully connected layer, and normalize and classify the 1*2 feature vector into a correlation through a softmax classifier. After the detection device obtains the correlation between the topic tag and the material content, the RoBERTa model can output a detection result of whether the first topic material is a hot topic material based on the obtained correlation.

[0138] In some embodiments, when the correlation degree is greater than a first threshold, the RoBERTa model outputs a detection result indicating that the first topic material is a non-hotspot material; when the correlation degree is less than a second threshold, the RoBERTa model outputs a detection result indicating that the first topic material is a hotspot material; when the correlation degree is between the first threshold and the second threshold, the RoBERTa model outputs a detection result that the material type of the first topic material cannot be determined.

[0139] In some embodiments, the second threshold is less than the first threshold. For example, the first threshold may be any value between 0.8 and 1, and the second threshold may be any value between 0 and 0.1.

[0140] In some embodiments, Figure 7 1 is a flow chart of determining the material type of the first topic material based on the RoBERTa model provided by an embodiment of the present application. In the following, taking the third material classification model as the RoBERTa model and the first topic material as "#This generation of young people love to drink tea#I like both pure tea and milk tea!" as an example, Figure 7 , the implementation process of determining the material type of the first topic material based on the RoBERTa model is introduced.

[0141] In step S701, the detection device inputs the material "#This generation of young people love to drink tea# They like both pure tea and milk tea!" to the RoBERTa model.

[0142] In step S702, the detection device extracts the topic tag "This generation of young people loves to drink tea", and isolates the topic tag and the material content through the character [SEP] to obtain "[CLS]#This generation of young people loves to drink tea#[SEP]#I like both pure tea and milk tea![SEP]", and masks the topic tag through [MASK] to obtain "#[MASK]#I like both pure tea and milk tea![SEP]".

[0143] In step S703, the detection device uses the embedding layer ( Figure 7 The material “[CLS]#This generation of young people likes to drink tea#[SEP]#They like both pure tea and milk tea! [SEP]” is encoded in the Embedding in [CLS] to obtain the feature vector [101, 108, 6821, 2237, 2399, 6768, 782, 4263, 1600, 5763, 108, 102, 108, 103, 108, 5283, 5763, 1959, 5763, 6963, 1599, 3614, 8013, 102].

[0144] In step S704, the detection device encodes the feature vector corresponding to the position of the topic tag through the RoBERTa model to obtain a vector [-1.3198768, 1.2342917].

[0145] In step S705, the detection device performs probability normalization calculation on the vector [-1.3198768, 1.2342917] through a softmax classifier to obtain a vector [0.07214694, 0.92785305].

[0146] In step S706, the detection device obtains a correlation degree of 0.927 output by the RoBERTa model.

[0147] In some embodiments, if the first threshold is 0.8, the detection device determines that "#This generation of young people love to drink tea# They like both pure tea and milk tea!" is not a hot topic material.

[0148] In some embodiments, before determining the correlation between the topic tag carried by the first topic material and the material content in the first topic material through the third material classification model, the detection device may also train the third material classification model so that the third material classification model has the ability to determine the correlation between the topic tag carried by the first topic material and the material content in the first topic material.

[0149] Exemplarily, the detection device may obtain sample materials, which include hot spot materials and non-hot spot materials. Afterwards, the detection device may input the obtained sample materials into a third material classification model to train the third material classification model to obtain a trained third material classification model.

[0150] In some embodiments, the ratio of hotspot sample materials and non-hotspot sample materials used to train the third material classification model can be 1:1.

[0151] It should be noted that the proportion and quantity of hot spot sample materials and non-hot spot sample materials in the sample materials used to train the third material classification model can also be determined as needed, and the embodiments of the present application do not limit this.

[0152] In the embodiment of the present application, since the difference between hot topic materials and non-hot topic materials lies in the different correlations between the topic tags carried in the materials and the material content, and most of the materials posted by users are non-hot topic materials. Therefore, the detection device first determines whether the first topic material is a hot topic material based on the correlation between the topic tags carried by the first topic material and the material content. When the detection device cannot determine whether the first topic material is a hot topic material based on the correlation between the topic tags and the material content, it determines whether the first topic material is a hot topic material based on the above steps S301 and S302. This can reduce the overhead of determining whether a topic material is a hot topic material through the above steps S301 and S302.

[0153] refer to Figure 8 , Figure 8 is a flow chart of a model training method provided in an embodiment of the present application. Figure 8 Taking the third material classification model as an example, the process of training the third material classification model is introduced.

[0154] In step S801, the detection device obtains sample materials to obtain a training set, wherein the sample materials include hot spot materials and non-hot spot materials.

[0155] In step S802, the detection device inputs the sample material into the RoBERTa model, encodes the sample material, and obtains a 512*768 feature vector.

[0156] In step S803, the detection device obtains a feature vector 1*768 of the location of the topic tag based on the feature vector 512*768 through the RoBERTa model.

[0157] In step S804, the detection device obtains the degree of association based on the feature vector 1*768 of the location of the topic tag through the connection layer.

[0158] In step S805, the detection device adjusts and optimizes the model parameters of the third material classification model based on the correlation degree.

[0159] In step S806, the detection device evaluates whether the third material classification model converges.

[0160] In step S807, when determining that the third material classification model has converged, the detection device deploys the third material classification model.

[0161] It should be noted that the sample materials used to train the first material classification model, the second material classification model, and the third material classification model may be the same or different, and the embodiments of the present application are not limited to this.

[0162] In some embodiments, the detection device may also determine the correlation between the topic tag and the material content based on the similarity between the topic tag carried by the first topic material and the material content. The correlation between the topic tag and the material content is proportional to the similarity between the topic tag and the material content.

[0163] In some embodiments, the detection device can determine the similarity between the topic tag carried by the first topic material and the material content based on methods such as cosine similarity and Euclidean distance.

[0164] In some embodiments, the first topic material may carry one or more topic tags. When the detection device determines that the first topic material carries more than one topic tag, the detection device may determine that the first topic material is a non-hotspot material when the correlation between all topic tags carried by the first topic material and the material content is greater than a first threshold.

[0165] In some embodiments, the first topic material carries multiple topic tags, and the multiple topic tags include first-category topic tags and / or second-category topic tags, wherein the correlation between the first-category topic tags and the material content is greater than a first threshold, and the correlation between the second-category topic tags and the material content is less than the first threshold; when the correlation is greater than the first threshold, determining that the first topic material belongs to a non-hotspot material, including: when multiple topic tags are all first-category topic tags, determining that the first topic material belongs to a non-hotspot material; when multiple topic tags include first-category topic tags and second-category topic tags, retrieving a second topic material that has multiple topic tags at the same time; when the number of second topic materials is greater than or equal to a preset number, determining that the first topic material is a non-hotspot material; when the number of second materials is less than a preset number, determining that the first topic material is a hotspot material.

[0166] It can be understood that when there are multiple topic tags carried by the first topic material, the correlation between all topic tags carried in the first topic material and the material content may be greater than the first threshold, or the correlation between some topic tags and the material content may be greater than the first threshold, and the correlation between another part of the topic tags and the material content may be less than the first threshold. In the following, for the convenience of explanation, the topic tags with a correlation between the material content greater than the first threshold may be referred to as the first type of topic tags, and the topic tags with a correlation between the material content less than the first threshold may be referred to as the second type of topic tags. Based on this, when the detection device determines that the multiple topic tags carried by the first topic material are all the first type of topic tags, it determines that the first topic material is a non-hotspot material, and the first topic material can be recommended to the user. When the detection device determines that the multiple topic tags include the first type of topic tags and the second type of topic tags, the second topic material with multiple topic tags can be retrieved at the same time. When the detection device determines that the number of the second topic material is greater than the preset number, it determines that the first topic material is a non-hotspot material, and the first topic material can be recommended to the user. When the detection device determines that the quantity of the second topic material is less than or equal to the preset quantity, it determines that the first topic material is a hot topic material, and the first topic material can be sent to the manual review platform for manual review.

[0167] In some embodiments, when determining that the plurality of topic tags include the first category topic tags and the second category topic tags, the detection device may retrieve the second topic material that has the plurality of topic tags carried in the first topic material within a preset period before the current moment, and determine the number of the second topic materials. Exemplarily, the detection device may retrieve the second topic material that has the plurality of topic tags carried in the first topic material within 24 hours before the current moment, and determine the number of the second topic materials on the network that have the plurality of topic tags carried in the first topic material within 24 hours before the current moment.

[0168] In some embodiments, before executing step 102, the method further includes: when confirming that the correlation degree is between the first threshold and the second threshold, determining whether the topic tag in the first topic material belongs to a preset field; and confirming that the topic tag does not belong to the preset field.

[0169] It is understandable that when the detection device determines that the correlation between the topic tag carried by the first topic material and the material content is between the first threshold and the second threshold, it can determine whether the topic tag belongs to the preset field. When the detection device determines that the topic tag does not belong to the preset field, the above steps S301 and S302 can be performed.

[0170] In some embodiments, the preset fields include but are not limited to: film and television field, operation field, generalization field and other fields.

[0171] In some embodiments, the TV drama field can be understood as the topic tags carried by the topic material being the topic tags related to TV dramas or movies. The operation topic-related field can be understood as the topic tags carried by the topic material being the topic tags related to operation. The generalization field can be understood as the topic tags carried by the topic material being the topic tags related to emotions, comedy, etc.

[0172] In some embodiments, after obtaining the topic tag carried by the first topic material, the detection device can determine whether the topic tag carried by the first topic material belongs to a topic tag in the film and television field topic library, the operation field topic library, or the generalization field topic library. When the detection device determines that the topic tag does not belong to a topic tag in the film and television field topic library, the operation field topic library, or the generalization field topic library, it determines that the topic tag carried by the first topic material does not belong to the preset field.

[0173] In some embodiments, before executing step S302, the method further includes: when the topic tag belongs to a preset field, determining whether the topic tag and the material content in the first topic material belong to the same field; and confirming that the field to which the topic tag belongs is different from the field to which the material content belongs.

[0174] It is understandable that when the detection device determines that the topic tag carried by the first topic material belongs to a preset field, it can further determine whether the topic tag and the field to which the material content of the first topic material belongs are the same. When the detection device determines that the field to which the material content in the first topic material belongs is different from the field to which the topic tag belongs, based on the above steps S301 and S302, it determines whether the first topic material is a hot topic material.

[0175] In some embodiments, when the detection device determines that the first topic material does not belong to a topic tag in any topic library among the film and television field topic library, the operation field topic library, and the generalization field topic library, it determines that the topic tag carried by the first topic material belongs to a preset field.

[0176] For example, Fig. 9 Schematic diagram of another method for detecting topic materials provided in an embodiment of the present application. Fig. 9 , the detection device can collect historical topic materials in a preset field, and build a topic library 901 based on the topic tags carried by the collected historical topic materials. Afterwards, the detection device can build a topic tree 902 based on the topic tags in the topic library 901. In addition, the detection device can also establish a topic association library 903 based on multiple topic tags carried by historical topic materials. Among them, the topic tag association relationship in the topic association library 903 is the association relationship between multiple topic tags carried by a historical topic material. Based on this, the detection device processes the first topic material based on the following rules:

[0177] Correlation rule 904: When the correlation between all topic tags carried by the first topic material and the material content is greater than the first threshold, the first topic material is determined to be a non-hotspot material, and the first topic material is recommended to the user; when the correlation between all topic tags carried by the first topic material and the material content is less than the first threshold, the first topic material is determined to be a hotspot material, and the first topic material is sent to the manual review platform for manual review.

[0178] Field screening rule 905: When the topic tag carried by the first topic material belongs to a topic tag of a preset field, and the field to which the material content included in the first topic material belongs is the same as the field to which the topic tag belongs, the first topic material is determined to be a non-hotspot material, and the first topic material is recommended to the user; when the field to which the topic tag carried by the first topic material belongs is not the preset field, further determine whether the first topic material is a hotspot material through the above steps S301 and S302; when the field to which the topic tag carried by the first topic material belongs is the preset field, and the field to which the topic tag carried by the first topic material belongs is different from the field to which the material content belongs, further determine whether the first topic material is a hotspot material through the above steps S301 and S302.

[0179] Related topic screening rule 906: When multiple topic tags carried by the first topic material are all first-category topic tags, determine that the first topic material is a non-hotspot material, and recommend the first topic material to the user; when multiple topic tags include first-category topic tags and second-category topic tags, based on the historical topic materials in the topic association relationship library 903, determine the number of second topic materials that have multiple topic tags at the same time, and when the number of second topic materials is greater than a preset number, determine that the first topic material is a non-hotspot material, and recommend the first topic material to the user; when the number of first topic materials is less than or equal to the preset number, further determine the type of the first topic material through the above steps S301 and S302.

[0180] Hereinafter, the implementation process of the topic material detection method provided in the above embodiment in actual application will be described in detail.

[0181] In some embodiments, Fig.10 FIG. 1 is a flow chart of another method for detecting topic materials provided in an embodiment of the present application. Fig.10 The detection method of the above-mentioned topic material is explained.

[0182] In the first step, the detection device obtains multiple topic tags and material content carried by the first topic material.

[0183] In the second step, the detection device inputs the first topic material into the sensitive information identification module 1001 to determine whether there is sensitive information in all the topic tags carried by the first topic material and the material content through the content security model a included in the sensitive information identification module 1001. If so, execute the third step; if not, execute the fourth step.

[0184] It should be noted that the content security model a may include the above-mentioned image recognition model and text recognition model.

[0185] In the third step, the detection device sends the first topic material to the manual review platform for manual review.

[0186] In the fourth step, the detection device inputs the first topic material into the non-hotspot identification module 1002 to determine the correlation between each topic tag carried by the first topic material and the material content through the third material classification model b included in the non-hotspot identification module 1002.

[0187] In the fifth step, when the correlation between all topic tags carried by the first topic material and the material content is greater than the first threshold, the detection device determines that the first topic material is a non-hotspot material and pushes the first topic material to the user.

[0188] In the sixth step, when the first topic material carries the second type of topic tag, the detection device determines whether the second type of topic tag with a correlation less than the first threshold belongs to the preset field based on the topic matching model c included in the non-hotspot identification module 1002. If so, execute the seventh step, if not, execute the eighth step.

[0189] In the seventh step, when the detection device determines that the second category topic tag belongs to the preset field, and the field to which the material content included in the first topic material belongs is the same as the field to which the second category topic tag belongs, the detection device determines that the first topic material is a non-hotspot material and pushes the first topic material to the user.

[0190] In the eighth step, when the detection device determines that the second category topic tag carried by the first topic material does not belong to the topic tag of the preset field; or the second category topic tag belongs to the preset field, but the field to which the material content in the first topic material belongs is different from the field to which the second category topic tag belongs, the first topic material is input into the hot spot identification module 1003.

[0191] In the ninth step, the detection device determines the correlation between all topic tags carried by the first topic material and the material content based on the third material classification model d included in the hot spot identification module 1003.

[0192] It should be noted that the third material classification model d and the third material classification model b may be the same model or different models, and this embodiment of the present application does not limit this.

[0193] In the tenth step, when the detection device determines that the correlation between all topic tags carried by the first topic material and the material content is less than the second threshold based on the third material classification model d, it determines that the first topic material is a hot topic material and sends the first topic material to the manual review platform for manual review.

[0194] In the eleventh step, when the first topic material includes the first category of topics and the second category of topics, the detection device detects whether the first topic material is a hot topic material based on the first material classification model d, according to the topic information and material content of the first topic material, and user characteristic information. If so, execute the twelfth step, if not, execute the thirteenth step.

[0195] In the twelfth step, when the detection device determines that the first topic material is a hot topic material, the first topic material is sent to the manual review platform for manual review.

[0196] In the thirteenth step, the detection device determines whether the first topic material is a hot topic material based on the multimodal large model 1004, and when it is determined that the first topic material is a hot topic material, the first topic material is sent to the manual review platform for manual review, and when it is determined that the first topic material is not a hot topic material, the first topic material is pushed to the user.

[0197] The following is a description of an exemplary structure of the topic material detection device 455 provided in the embodiment of the present application implemented as a software module. In some embodiments, for example Figure 2 As shown, the software modules in the detection device 455 of the topic material stored in the memory 450 may include: a first acquisition module 4551, configured to obtain the topic information and material content of the first topic material to be detected, and user characteristic information corresponding to the first topic material, the topic information includes the topic tag carried in the first topic material, the user characteristic information includes the account information of the account that publishes the first topic material, and the material characteristics of the historically published materials of the account; a first detection module 4552, configured to detect whether the first topic material is a hot topic material based on the topic information and material content of the first topic material, and the user characteristic information; the hot topic material refers to a topic material where the topic tag and the material content are unrelated.

[0198] In some embodiments, the first detection module 4552 is configured as follows: when the material content includes text and pictures, based on the topic information and material content in the first topic material, and user characteristic information, the probability of the first topic material being a non-hotspot material is determined through the first material classification model; when it is determined based on probability that the first topic material does not meet the set hotspot material probability threshold, based on the text and pictures in the first topic material, whether the first topic material is a hotspot material is determined through the second material classification model.

[0199] In some embodiments, the second material classification model includes: a large language model; wherein the first detection module 4552 is configured to: generate relevant prompt words indicating the large language model to determine whether the first topic material is a hot topic material; encode the picture included in the first topic material to obtain the image features of the picture; align the image features with the text features; input the prompt words, the text included in the first topic material, and the image features aligned with the text features into the large language model, and determine whether the first topic material is a hot topic material through the large language model.

[0200] In some embodiments, before detecting whether the first topic material is a hot topic material based on the topic information and material content of the first topic material, as well as user characteristic information, the device also includes: a second acquisition module, configured to obtain the correlation between the topic tag in the first topic material and the material content; a first confirmation module, configured to confirm that the correlation is between a first threshold and a second threshold; the first threshold is greater than the second threshold; the device also includes: a first determination module, configured to determine that the first topic material is a non-hot topic material when the correlation is greater than the first threshold; a second determination module, configured to determine that the first topic material is a hot topic material when the correlation is less than the second threshold.

[0201] In some embodiments, before detecting whether the first topic material is a hot topic material based on the topic information and material content of the first topic material and user characteristic information, the device also includes: a first judgment module, configured to confirm whether the topic tag in the first topic material belongs to a preset field when the correlation degree is between the first threshold and the second threshold; a second confirmation module, configured to confirm that the topic tag does not belong to the preset field.

[0202] In some embodiments, before detecting whether the first topic material is a hot topic material based on the topic information and material content of the first topic material, as well as user characteristic information, the device also includes: a second judgment module, configured to judge whether the topic tag and the material content in the first topic material belong to the same field when the topic tag belongs to a preset field; a third confirmation module, configured to confirm that the field to which the topic tag belongs is different from the field to which the material content belongs; the device also includes: a third determination module, configured to determine that the first topic material is a non-hot topic material when the field to which the topic tag belongs is the same as the field to which the material content belongs.

[0203] In some embodiments, the first topic material carries multiple topic tags, and the multiple topic tags include first-category topic tags and / or second-category topic tags, wherein the correlation between the first-category topic tags and the material content is greater than a first threshold, and the correlation between the second-category topic tags and the material content is less than the first threshold; the first determination module is configured as: when multiple topic tags are all first-category topic tags, determining that the first topic material belongs to a non-hotspot material; when the multiple topic tags include first-category topic tags and second-category topic tags, retrieving a second topic material having multiple topic tags at the same time; when the number of second topic materials is greater than or equal to a preset number, determining that the first topic material is a non-hotspot material; when the number of second topic materials is less than a preset number, determining that the first topic material is a hotspot material.

[0204] In some embodiments, before obtaining the topic information and material content of the first topic material to be detected, and user characteristic information corresponding to the topic material, the device also includes: a second detection module, configured to detect whether sensitive information exists in the first topic material; and a fourth confirmation module, configured to confirm that no sensitive information exists in the first topic material.

[0205] In some embodiments, the second detection module is configured to: identify the pictures included in the first topic material based on the trained image recognition model to obtain picture recognition results, and the picture recognition results are used to indicate whether sensitive information exists in the pictures included in the first topic material; and / or identify the text included in the first topic material based on the trained text recognition model to obtain text recognition results, and the text recognition results are used to indicate whether sensitive information exists in the text included in the first topic material; determine whether sensitive information exists in the first topic material based on the picture recognition results and the text recognition results.

[0206] The embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-mentioned topic material detection method of the embodiment of the present application.

[0207] The present application embodiment provides a computer-readable storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will execute the topic material detection method provided in the present application embodiment, for example, Figure 3 The detection method of the topic material is shown.

[0208] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EP ROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0209] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0210] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file storing other programs or data, such as in one or more scripts in a hypertext markup language HTML document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0211] As an example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one location, or on multiple computing devices distributed at multiple locations and interconnected by a communication network. The above is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of the present application are included in the scope of protection of the present application.

Claims

1. A method for detecting topic materials, characterized in that: The method comprises: Acquire topic information and material content of a first topic material to be detected, and user characteristic information corresponding to the first topic material; the topic information includes a topic tag carried in the first topic material, and the user characteristic information includes account information of an account that publishes the first topic material, and material characteristics of materials published historically by the account; Based on the topic information and material content of the first topic material and the user characteristic information, detect whether the first topic material is a hot topic material; the hot topic material refers to a topic material in which the topic tag and the material content are unrelated.

2. The method according to claim 1, characterized in that The detecting whether the first topic material is a hot topic material based on the topic information and material content in the first topic material and the user characteristic information includes: In the case where the material content includes text and pictures, determining the probability that the first topic material is a non-hotspot material based on the topic information and material content in the first topic material and the user characteristic information through the first material classification model; When it is determined based on the probability that the first topic material does not meet the set probability threshold of hot topic materials, a second material classification model is used to determine whether the first topic material is a hot topic material based on the text and pictures in the first topic material.

3. The method according to claim 2, characterized in that The second material classification model includes: a large language model; The determining whether the first topic material is a hot topic material based on the text and pictures in the first topic material by using the second material classification model includes: Generate a related prompt word that instructs the large language model to determine whether the first topic material is a hot topic material; Encoding the pictures included in the first topic material to obtain image features of the pictures; Aligning the image features with the text features; The prompt word, the text in the first topic material, and the image features aligned with the text features are input into the large language model, and the large language model is used to determine whether the first topic material is a hot topic material.

4. The method according to claim 1, characterized in that: Before detecting whether the first topic material is a hot topic material based on the topic information and material content of the first topic material and the user characteristic information, the method further includes: Obtaining the correlation between the topic tag in the first topic material and the material content; confirming that the degree of association is between a first threshold and a second threshold; and that the first threshold is greater than the second threshold; The method further comprises: When the correlation degree is greater than the first threshold, it is determined that the first topic material is a non-hot topic material; When the degree of association is less than the second threshold, it is determined that the first topic material is a hot topic material.

5. The method according to claim 4, characterized in that Before detecting whether the first topic material is a hot topic material based on the topic information and material content of the first topic material and the user characteristic information, the method further includes: When the correlation degree is between the first threshold and the second threshold, it is determined whether the topic tag in the first topic material belongs to a preset field; and it is confirmed that the topic tag does not belong to the preset field.

6. The method according to claim 5, characterized in that Before detecting whether the first topic material is a hot topic material based on the topic information and material content of the first topic material and the user characteristic information, the method further includes: In the case where the topic tag belongs to the preset field, determining whether the topic tag and the material content in the first topic material belong to the same field; and confirming that the field to which the topic tag belongs is different from the field to which the material content belongs; The method further comprises: When the field to which the topic tag belongs is the same as the field to which the material content belongs, it is determined that the first topic material is a non-hot topic material.

7. The method according to claim 4, characterized in that The first topic material carries a plurality of topic tags, the plurality of topic tags include first-category topic tags and / or second-category topic tags, wherein the correlation between the first-category topic tags and the material content is greater than the first threshold, and the correlation between the second-category topic tags and the material content is less than the first threshold; When the correlation degree is greater than a first threshold, determining that the first topic material is a non-hotspot material includes: When the multiple topic tags are all first-category topic tags, determining that the first topic material is a non-hotspot material; When the plurality of topic tags include first-category topic tags and second-category topic tags, retrieving second topic materials having the plurality of topic tags at the same time; When the quantity of the second topic material is greater than or equal to a preset quantity, determining that the first topic material is a non-hot topic material; When the quantity of the second topic material is less than the preset quantity, it is determined that the first topic material is a hot topic material.

8. The method according to claim 1, characterized in that Before acquiring the topic information and material content of the first topic material to be detected, and the user characteristic information corresponding to the topic material, the method further includes: Detecting whether there is sensitive information in the first topic material; Confirm that the sensitive information does not exist in the first topic material.

9. The method according to claim 8, characterized in that The detecting whether there is sensitive information in the first topic material includes: Recognize the pictures included in the first topic material based on the trained image recognition model to obtain a picture recognition result, where the picture recognition result is used to indicate whether the sensitive information exists in the pictures included in the first topic material; Recognize the text included in the first topic material based on the trained text recognition model to obtain a text recognition result, where the text recognition result is used to indicate whether the sensitive information exists in the text included in the first topic material; Based on the image recognition result and the text recognition result, determine whether the sensitive information exists in the first topic material.

10. An electronic device, characterized in that: The electronic device comprises: A memory for storing executable instructions; A processor, configured to implement the method according to any one of claims 1 to 9 when executing the executable instructions or computer programs stored in the memory.

Images