A data processing method, device, apparatus, and storage medium

Abstract

Embodiments of the application disclose a data processing method, device and equipment in the field of artificial intelligence, and a storage medium, wherein the method comprises: obtaining target text of a target barrage to be identified and user barrage features corresponding to a target user sending the target barrage; the user barrage features are determined according to at least historical barrage sending conditions of the user; determining an identification result corresponding to the target barrage according to the target text and the user barrage features corresponding to the target user through a barrage identification model; and the identification result corresponding to the target barrage is used to represent a type of the target barrage. The method can accurately identify whether the barrage belongs to bad barrage, and avoid missing identification of bad barrage and misidentification of normal barrage as bad barrage.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence (AI), and in particular to a data processing method and device, equipment and a storage medium. BACKGROUND

[0002] Nowadays, many network video playing platforms have the function of bullet screen, that is, supporting users to edit and send their comments on the currently played video, that is, bullet screen, during watching the video, and users watching the video can accordingly see the bullet screen sent by other users. In actual application, the bullet screen content sent by users is mixed with good and bad content, among which there are often bullet screens including improper content, boring content, content that other users hate, and other bad content. Such bullet screen is referred to as bad bullet screen in the industry. In order to improve the video watching experience of users, some network video playing platforms identify whether the bullet screen sent by the user is a bad bullet screen, and filter the bad bullet screen, and do not show the bad bullet screen to other users.

[0003] In related technologies, whether the bullet screen belongs to a bad bullet screen is usually identified only from the content of the bullet screen itself. For example, the words included in the bullet screen are matched with the words in the pre-constructed illegal word list. If the words in the bullet screen successfully match the words in the illegal word list, it can be determined that the bullet screen belongs to a bad bullet screen. For another example, a text classification model learned through a machine learning algorithm is used to identify whether the bullet screen belongs to a bad bullet screen.

[0004] The above-mentioned bullet screen identification method only focuses on the single dimension information of the bullet screen text when identifying whether the bullet screen belongs to a bad bullet screen, and it is often difficult to accurately identify bad bullet screens, and it is easy to miss identifying bad bullet screens and misidentifying normal bullet screens as bad bullet screens. SUMMARY

[0005] The embodiments of the present application provide a data processing method, device, equipment and storage medium, which can accurately identify whether the bullet screen belongs to a bad bullet screen, and avoid missing identifying bad bullet screens and misidentifying normal bullet screens as bad bullet screens.

[0006] Therefore, the first aspect of the present application provides a data processing method, which comprises:

[0007] For a target bullet screen to be identified, a target text of the target bullet screen and a user bullet screen feature corresponding to a target user sending the target bullet screen are obtained. The user bullet screen feature is determined at least according to the historical bullet screen sending situation of the user.

[0008] The bicker recognition model is used to determine a recognition result corresponding to the target bicker according to the target text and the user bicker feature corresponding to the target user, and the recognition result corresponding to the target bicker is used to represent a type of the target bicker.

[0009] The second aspect of the present application provides a data processing apparatus, and the apparatus comprises:

[0010] The data acquisition module is configured to acquire, for a target bicker to be recognized, a target text of the target bicker and a user bicker feature corresponding to a target user sending the target bicker, and the user bicker feature is determined according to at least historical bicker sending conditions of a user.

[0011] The bicker recognition module is configured to use a bicker recognition model to determine a recognition result corresponding to the target bicker according to the target text and the user bicker feature corresponding to the target user, and the recognition result corresponding to the target bicker is used to represent a type of the target bicker.

[0012] The third aspect of the present application provides an apparatus, and the apparatus comprises a processor and a memory:

[0013] The memory is configured to store a computer program.

[0014] The processor is configured to execute steps of the data processing method according to the computer program.

[0015] The fourth aspect of the present application provides a computer readable storage medium, and the computer readable storage medium is configured to store a computer program, and the computer program is configured to execute steps of the data processing method according to the first aspect.

[0016] The fifth aspect of the present application provides a computer program product or a computer program, and the computer program product or the computer program comprises computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes steps of the data processing method according to the first aspect.

[0017] From the above technical solutions, it can be seen that the embodiments of the present application have the following advantages:

[0018] In the data processing method provided in this application embodiment, for the target bullet screen to be identified, the target text of the target bullet screen and the user bullet screen features corresponding to the target user who sent the target bullet screen are first obtained. Here, the user bullet screen features corresponding to the target user are determined based on the target user's historical bullet screen sending situation. Then, through the bullet screen recognition model, the recognition result corresponding to the target bullet screen is determined based on the target text and the user bullet screen features corresponding to the target user. The recognition result corresponding to the target bullet screen can correspondingly characterize the type of the target bullet screen, such as characterizing whether the target bullet screen belongs to the category of malicious bullet screens. Compared to related technologies that identify whether a bullet comment is inappropriate solely based on the text of the bullet comment, the data processing method provided in this application comprehensively considers both the text information of the target bullet comment itself and the user bullet comment characteristics corresponding to the target user who sent the target bullet comment when identifying whether a target bullet comment is inappropriate. Since user bullet comment characteristics are determined based on the user's historical bullet comment sending behavior, they can reflect the characteristics of the bullet comments sent by the user. Therefore, using the user's corresponding user bullet comment characteristics to assist in identifying whether the bullet comment sent by the user is inappropriate can effectively improve the accuracy of identifying inappropriate bullet comments, avoid the situation of missed or misidentified inappropriate bullet comments, and thus help to effectively filter inappropriate bullet comments, avoid showing inappropriate bullet comments to video viewers, and improve the user's video viewing experience. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating an application scenario of the data processing method provided in the embodiments of this application;

[0020] Figure 2 A flowchart illustrating the data processing method provided in an embodiment of this application;

[0021] Figure 3 A schematic diagram illustrating the working principle of a bullet screen recognition model provided in this application embodiment;

[0022] Figure 4 A schematic diagram illustrating the working principle of another bullet screen recognition model provided in this application embodiment;

[0023] Figure 5 A flowchart illustrating the method for determining the target rate of malicious bullet comments provided in this application embodiment;

[0024] Figure 6 A flowchart illustrating the method for determining the target feature vector provided in this application embodiment;

[0025] Figure 7 A schematic diagram illustrating the working principle of the user bullet screen feature modeling model provided in this application embodiment;

[0026] Figure 8A working principle schematic diagram of a user feature fusion model provided by an embodiment of the present application is shown in FIG. 1.

[0027] Figure 9 A structure schematic diagram of a first data processing device provided by an embodiment of the present application is shown in FIG. 2.

[0028] Figure 10 A structure schematic diagram of a second data processing device provided by an embodiment of the present application is shown in FIG. 3.

[0029] Figure 11 A structure schematic diagram of a third data processing device provided by an embodiment of the present application is shown in FIG. 4.

[0030] Figure 12 A structure schematic diagram of a fourth data processing device provided by an embodiment of the present application is shown in FIG. 5.

[0031] Figure 13 A structure schematic diagram of a fifth data processing device provided by an embodiment of the present application is shown in FIG. 6.

[0032] Figure 14 A structure schematic diagram of a sixth data processing device provided by an embodiment of the present application is shown in FIG. 7.

[0033] Figure 15 A structure schematic diagram of a terminal device provided by an embodiment of the present application is shown in FIG. 8.

[0034] Figure 16 A structure schematic diagram of a server provided by an embodiment of the present application is shown in FIG. 9. DETAILED DESCRIPTION

[0035] In order to make the personnel in the technical field better understand the present application scheme, the technical scheme in the present application embodiment will be described clearly and completely in the following with the drawings in the present application embodiment. Obviously, the described embodiments are only a part of the present application, not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by the person of ordinary skill in the art without making creative labor belong to the scope of protection of the present application.

[0036] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, and above-mentioned drawings (if any) are used to distinguish between similar objects, not necessarily described in a particular sequential or chronological order. It is to be understood that the use of data so designated is not intended to limit the present application to the specific examples for which they are used, but rather, are used to illustrate embodiments of the present application. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted in an overly literal sense unless expressly so defined herein.

[0037] Natural language processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science and mathematics. Therefore, the research in this field will involve natural language, i.e. the language used in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph and other technologies.

[0038] Machine learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a subject that studies how computers simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent. Its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and example-based learning.

[0039] The scheme provided by the embodiments of the present application relates to natural language processing technology and machine learning technology in the field of artificial intelligence, which is specifically explained as follows:

[0040] The data processing method provided by the embodiments of the present application can consider the text information of the barrage and the user barrage features corresponding to the user sending the barrage when identifying the type of the barrage (such as identifying whether the barrage belongs to bad barrage), thereby improving the identification accuracy of the barrage.

[0041] Specifically, in the data processing method provided in the embodiments of the present application, for a target barrage to be identified, the target text of the target barrage and the user barrage feature corresponding to the target user sending the target barrage are first obtained, where the user barrage feature corresponding to the target user is determined according to the historical barrage sending situation of the target user; then, according to the target text and the user barrage feature corresponding to the target user, the recognition result corresponding to the target barrage is determined through a barrage recognition model, and the recognition result corresponding to the target barrage can correspondingly represent the type of the target barrage, for example, whether the target barrage belongs to bad barrage.

[0042] Compared with the technical solution in the related art that only identifies whether the barrage belongs to bad barrage from the dimension of barrage text, when identifying whether the target barrage belongs to bad barrage, the data processing method provided in the embodiments of the present application comprehensively considers the text information of the target barrage itself and the user barrage feature corresponding to the target user sending the target barrage; since the user barrage feature is determined according to the historical barrage sending situation of the user, it can reflect the characteristics of the barrage sent by the user, therefore, using the user barrage feature corresponding to the user to assist in identifying whether the barrage sent by the user belongs to bad barrage can effectively improve the identification accuracy of bad barrage, avoid the situation of missing identification and misidentification of bad barrage, and further help to effectively filter bad barrage, avoid showing bad barrage to video watching users, and improve the video watching experience of users.

[0043] It should be understood that the data processing method provided in the embodiments of the present application can be applied to a device with text processing capability, such as a terminal device or a server. The terminal device can be specifically a smart phone, a computer, a tablet computer, a personal digital assistant (PDA), a vehicle-mounted terminal, etc.; the server can be specifically an application server or a Web server, and in actual deployment, can be an independent server, a cluster server or a cloud server. The related data (such as barrage text, user barrage feature, recognition result of barrage, etc.) involved in the data processing method disclosed in the embodiments of the present application can be saved on a blockchain.

[0044] In order to facilitate understanding of the data processing method provided in the embodiments of the present application, the application scenario of the data processing method will be exemplarily introduced below taking the server as an example of the execution subject of the data processing method.

[0045] Referring to Figure 1 , Figure 1 the application scenario of the data processing method provided in the embodiments of the present application. As Figure 1As shown, the application scenario includes a terminal device 110, a server 120 and a database 130; the terminal device 110 and the server 120 can communicate through a network; the server 120 can call data from the database 130 through the network, or the database 130 can be integrated in the server 120. Among them, the terminal device 110 runs a video playing application (Application, APP), and the video playing APP supports users to send a barrage during watching a video; the server 120 can be a background server of the video playing APP, which is used to execute the data processing method provided by the embodiments of the application; the database 130 is used to store the user barrage features corresponding to the users using the video playing APP.

[0046] In actual application, when a target user watches a target video through the video playing APP running on the terminal device 110, the user can use the barrage sending function provided by the video playing APP to edit and send the comment content on the video, i.e. the target barrage. For example, the video playing interface of the video playing APP includes a barrage editing bar and a barrage sending control, the user can input the target text in the barrage editing bar, and after confirming the input of the target text, click the barrage sending control to trigger the barrage sending operation; after the terminal device 110 detects that the user triggers the barrage sending operation, the terminal device 110 will generate the target barrage based on the target text input by the user in the barrage editing bar, and then send the target barrage to the server 120 through the network.

[0047] After the server 120 receives the target barrage, the server 120 will obtain the target text in the target barrage, and obtain the user barrage feature corresponding to the target user sending the target barrage from the database 130. It should be noted that the user barrage features corresponding to the users using the video playing APP stored in the database 130 are all predetermined; for example, the server 120 can determine the user barrage feature corresponding to each user using the video playing APP according to the historical barrage sending situation of the user, and then store the determined user barrage feature in the database 130; for example, the user barrage feature can be the bad barrage sending rate generated according to the historical barrage sending situation of the user, or the feature vector generated according to the historical barrage sending situation of the user, and the application does not limit the form of the user barrage feature.

[0048] Further, the server 120 can invoke the pre-trained barrage identification model 121 to determine the identification result corresponding to the target barrage according to the target text in the target barrage and the user barrage feature corresponding to the target user sending the target barrage, i.e., determine the type of the target barrage, such as whether the target barrage belongs to bad barrage. For example, when the user barrage feature corresponding to the target user is the bad barrage sending rate corresponding to the target user, the server 120 can first determine the basic probability that the target barrage belongs to bad barrage according to the target text through the barrage identification model 121, and then correct the basic probability by using the bad barrage sending rate corresponding to the target user to obtain the target probability that the target barrage belongs to bad barrage as the identification result corresponding to the target barrage. For example, when the user barrage feature corresponding to the target user is the feature vector corresponding to the target user, the server 120 can fuse the feature vector corresponding to the target user and the word feature vector corresponding to each word in the target text through the barrage identification model 121, and then determine the identification result corresponding to the target barrage based on the fused feature vector. Of course, in actual application, the server 120 can also use other ways to consider the target text and the user barrage feature corresponding to the target user to identify whether the target barrage belongs to bad barrage, and the application does not make specific limitation on the identification method of the target barrage.

[0049] If the server 120 determines that the target barrage belongs to bad barrage through the above operation, the target barrage is filtered, i.e., discarded, and the target barrage is not displayed to other viewing users of the target video. On the contrary, if the server 120 determines that the target barrage does not belong to bad barrage through the above operation, the target barrage can be normally displayed so that other viewing users of the target video can normally see the target barrage.

[0050] It should be understood that Figure 1 The application scenarios shown are only examples, and in actual application, the data processing method provided by the application embodiment can also be applied to other scenarios; for example, the terminal device 110 can also independently execute the data processing method to identify whether the target barrage sent by the target user is bad barrage, and directly intercept the target barrage identified as bad barrage; for example, the server 120 can also determine the user barrage feature of the target user in real time without deploying the database 130. The application scenario of the data processing method provided by the application embodiment is not limited in any way.

[0051] The data processing method provided by the application will be described in detail below through method embodiments.

[0052] Referring to Figure 2 , Figure 2A flowchart of a data processing method provided by an embodiment of the present application is shown. For ease of description, the following embodiment is described by taking a server as an example of an execution subject of the data processing method. As shown in Figure 2 the data processing method includes the following steps:

[0053] Step 201: For a target barrage to be identified, a target text of the target barrage and a user barrage feature corresponding to a target user sending the target barrage are obtained; the user barrage feature is determined according to at least a historical barrage sending situation of the user.

[0054] In actual application, when a user watches a video through a video playing application or a video playing website, the user can use a barrage sending function provided by the video playing application or the video playing website to trigger a barrage editing operation and a barrage sending operation. In response to the barrage sending operation triggered by the user, a terminal device supporting running of the video playing application or establishing a communication connection with the video playing website will correspondingly send a barrage edited by the user through the barrage editing operation to a background server of the video playing application or the video playing website. It should be understood that the barrage sent by the terminal device to the server includes a barrage text edited by the user through the barrage editing operation; optionally, the barrage can also include a barrage style (such as a barrage text font, a barrage text color, a barrage decoration frame, etc.) selected by the user through the barrage editing operation.

[0055] After the server receives the barrage sent by the terminal device, the server regards the barrage as a target barrage to be identified and extracts a target text from the target barrage. The target text here is a barrage text edited and sent by the user through the barrage sending function, for example, assuming that the user edits and sends a target barrage “The plot is too wonderful” through a barrage sending function provided by a video playing application, the target text extracted from the target barrage by the server is “The plot is too wonderful”.

[0056] In addition, the server also needs to obtain a user barrage feature corresponding to the target user sending the target barrage, which is determined according to at least the historical barrage sending situation of the target user, and can reflect the characteristics of the barrage sent by the target user. As an example, after receiving the target barrage sent by the terminal device, the server can obtain the user barrage feature corresponding to the user identifier of the target user sending the target barrage from the database for storing the user barrage feature, as the user barrage feature corresponding to the target user. As another example, after receiving the target barrage sent by the terminal device, the server can determine the user barrage feature corresponding to the target user in real time according to the historical barrage sending situation of the target user sending the target barrage. Of course, in actual application, the server can also obtain the user barrage feature corresponding to the target user by other ways, and the application does not limit the way of obtaining the user barrage feature.

[0057] In some cases, the user barrage feature corresponding to the target user can be the bad barrage sending rate of the target user. For example, the proportion of the bad barrage sent by the target user in the historical reference period in all the barrages sent by the target user in the historical reference period can be calculated to obtain the basic bad barrage sending rate of the target user, as the user barrage feature corresponding to the target user. For another example, the target bad barrage sending rate of the target user can also be determined on the basis of the above-mentioned basic bad barrage sending rate of the target user, combined with the basic bad barrage sending rate of other reference users having a target social relationship with the target user, as the user barrage feature corresponding to the target user, and the determination method of the user barrage feature will be described in detail in another method embodiment below.

[0058] In some cases, the user barrage feature corresponding to the target user can be a feature vector corresponding to the target user. For example, the target user can be modeled according to the historical barrage sending situation of the target user to obtain a feature vector capable of representing the barrage sending characteristics of the target user, as the user barrage feature corresponding to the target user. For another example, the target user can be modeled according to the bad barrage sending rate of the target user, the number of barrages sent in the reference period, and the personal basic information of the target user to obtain a basic feature vector capable of more comprehensively representing the barrage sending characteristics of the target user, as the user barrage feature corresponding to the target user. The determination method of this kind of user barrage feature will be described in detail below through another method embodiment. For another example, the target feature vector corresponding to the target user can also be determined based on the basic feature vector corresponding to the target user and the basic feature vectors corresponding to other reference users having a target social relationship with the target user, as the user barrage feature corresponding to the target user. The determination method of this kind of user barrage feature will be described in detail below through another method embodiment.

[0059] It should be understood that in actual application, in addition to the bad barrage sending rate, the feature vector, etc. mentioned above corresponding to the target user can be used as the user barrage feature corresponding to the target user, other forms of user barrage features can also be determined according to the historical barrage sending situation of the target user according to actual needs, and the form of the user barrage feature and the method of determining the user barrage feature are not limited in the present application.

[0060] Step 202: determining, by a barrage identification model, an identification result corresponding to the target barrage according to the target text and the user barrage feature corresponding to the target user; the identification result corresponding to the target barrage is used to represent the type of the target barrage.

[0061] After the server obtains the target text of the target barrage and the user barrage feature corresponding to the target user, the server can determine an identification result corresponding to the target barrage according to the target text and the user barrage feature corresponding to the target user by using a pre-trained barrage identification model. The identification result corresponding to the target barrage can represent the type of the target barrage, for example, whether the target barrage belongs to a bad barrage.

[0062] It should be noted that the above barrage identification model is a pre-trained text category identification model, which can identify whether the target barrage belongs to bad barrage according to the input content. As an example, the barrage identification model can be a model specialized in processing text, that is, the server can only input the target text of the target barrage into the barrage identification model, and the barrage identification model can output the preliminary identification result corresponding to the target barrage after analyzing and processing the input target text, and then the preliminary identification result corresponding to the target barrage is corrected by using the user barrage features corresponding to the target user to obtain the identification result corresponding to the target barrage. As another example, the barrage identification model can be a model for comprehensive processing of text and user barrage features, that is, the server can input the target text of the target barrage and the user barrage features corresponding to the target user into the barrage identification model, and the barrage identification model can output the identification result corresponding to the target barrage after analyzing and processing the input target text and user barrage features.

[0063] In a possible implementation, when the user barrage features corresponding to the target user are the target bad barrage sending rate of the target user, the server can first determine, by the barrage identification model, a basic probability that the target barrage belongs to bad barrage according to the target text of the target barrage; and then correct the above basic probability by using the target bad barrage sending rate of the target user to obtain a target probability that the target barrage belongs to bad barrage as the identification result corresponding to the target barrage.

[0064] For example, the working principle of the barrage identification model used in this implementation can be as shown in Figure 3 The barrage identification model includes a word feature extraction structure 301 and a word feature fusion structure 302, wherein the word feature extraction structure 301 can include a recurrent neural network (RNN) layer and a full connection layer, and the word feature fusion structure 302 can include a full connection layer, a convolution layer, a pooling layer and another full connection layer.

[0065] In specific work, after the server inputs the target text of the target barrage into the barrage identification model, the RNN layer in the word feature extraction structure 301 will perform preliminary feature extraction processing on each word in the input target text to obtain a primary word feature vector corresponding to each word; and then the full connection layer in the word feature extraction structure 301 will process the primary word feature vector corresponding to each word again to obtain a word feature vector corresponding to each word.

[0066] It should be noted that the RNN layer in the word feature extraction structure 301 can also be replaced by a Gated Recurrent Unit (GRU) or a Long Short-Term Memory (LSTM), and the specific structure can be set according to actual needs. The present application does not make any limitation on the specific structure of the word feature extraction structure 301.

[0067] After the word feature extraction structure 301 extracts the word feature vectors corresponding to each word in the target text, the word feature fusion structure 302 will determine the target text feature vector according to the word feature vectors corresponding to each word, and determine the basic probability of the target barrage belonging to bad barrage based on the target text feature vector. Specifically, the fully connected layer in the word feature fusion structure 302 will perform preliminary fusion processing on the word feature vectors corresponding to each word to obtain a preliminary fusion feature vector, and then the convolution layer and the pooling layer in the word feature fusion structure 302 will sequentially perform convolution processing and pooling processing on the preliminary fusion feature vector to obtain the target text feature vector. Then, the other fully connected layer in the word feature fusion structure 302 processes the target text feature vector to obtain the basic probability of the target barrage belonging to bad barrage.

[0068] It should be understood that Figure 3 The word feature fusion structure 302 shown is only an example. In actual applications, other structures of the word feature fusion structure 302 can also be designed according to actual application needs, and the present application does not make any limitation on the specific structure of the word feature fusion structure 302.

[0069] After the server obtains the basic probability of the target barrage belonging to bad barrage through the barrage recognition model, it can further correct the basic probability using the target bad barrage sending rate corresponding to the target user to obtain the target probability of the target barrage belonging to bad barrage. An exemplary correction method can be shown in equation (1):

[0070] Target probability = basic probability + a * target bad barrage sending rate (1)

[0071] Where a is a pre-set correction coefficient, which can be set according to actual needs. Of course, in actual applications, the server can also use other ways to correct the basic probability using the target bad barrage sending rate corresponding to the target user to obtain the target probability of the target barrage belonging to bad barrage, and the present application does not make any limitation on the above correction method.

[0072] It should be understood that if the target probability of the target barrage belonging to the bad barrage exceeds the preset probability threshold, it can be determined that the target barrage belongs to the bad barrage, and vice versa, if the target probability of the target barrage belonging to the bad barrage does not exceed the preset probability threshold, it can be determined that the target barrage does not belong to the bad barrage.

[0073] It should be noted that the target bad barrage sending rate corresponding to the target user can be determined according to the historical bad barrage sending situation of the target user. As an example, the number of bad barrages sent by the target user in the historical reference period and the number of all barrages sent by the target user in the historical reference period can be obtained, and then the proportion of the bad barrages sent by the target user in the historical reference period in all barrages is calculated to obtain the basic bad barrage sending rate corresponding to the target user. The basic bad barrage sending rate corresponding to the target user is used as the target bad barrage sending rate corresponding to the target user. As another example, the target bad barrage sending rate corresponding to the target user can be generated based on the basic bad barrage sending rate corresponding to the target user and the basic bad barrage sending rate corresponding to other reference users having a target social relationship with the target user on the target social network platform. The determination method of this target bad barrage sending rate will be described in detail in another method embodiment below.

[0074] Optionally, in order to improve the model performance of the barrage identification model used in this implementation mode, the server can also update and train the barrage identification model using the barrages generated in actual application and the target probability of the barrage belonging to the bad barrage.

[0075] That is, the server can obtain the barrages generated in the first preset period as the first training barrage, and obtain the first training text of the first training barrage and the target probability of the first training barrage belonging to the bad barrage. Then, the first training sample is constructed using the first training text and the target probability of the first training barrage belonging to the bad barrage. The above barrage identification model is updated and trained based on the first training sample.

[0076] For example, the server can obtain each barrage generated in the past week as the first training barrage. For each first training barrage, the server can extract the barrage text as the first training text, and obtain the target probability of the first training barrage belonging to the bad barrage. The first training sample corresponding to the first training barrage is constructed using the first training text and the target probability of the first training barrage belonging to the bad barrage. After constructing the first training sample corresponding to each first training barrage in the above manner, the barrage identification model can be supervised and updated and trained using the constructed first training samples, thereby improving the identification performance of the barrage identification model.

[0077] It should be understood that the first training barrage is a barrage used when training the above barrage identification model; the first training sample is a sample constructed based on the second training text in the second training barrage and the target probability that the first training barrage belongs to bad barrage, which is a training sample used when training the above barrage identification model. The first preset period can be set according to actual needs, for example, one week, one month, etc., and the present application does not make any limitation on the first preset period.

[0078] In another possible implementation, when the user barrage feature corresponding to the target user is the target feature vector corresponding to the target user, the server can perform feature extraction processing on each word in the target text through the word feature extraction structure in the barrage identification model, to obtain the primary word feature vector corresponding to each word in the target text; and then, through the fusion feature processing structure in the barrage identification model, determine the recognition result corresponding to the target barrage according to the primary word feature vector corresponding to each word in the target text and the target feature vector corresponding to the target user.

[0079] For example, the working principle of the barrage identification model used in this implementation can be as shown in Figure 4 The barrage identification model includes a word feature extraction structure 401 and a fusion feature processing structure 402, wherein the word feature extraction structure 401 can include an RNN layer, and the fusion feature processing structure 402 can include two full-connection layers, a convolution layer, a pooling layer and another full-connection layer.

[0080] In specific work, the server can input the target text of the target barrage and the target feature vector corresponding to the target user into the barrage identification model; correspondingly, the RNN layer in the word feature extraction structure 401 in the barrage identification model will perform feature extraction processing on each word in the target text, to obtain the primary word feature vector corresponding to each word in the target text. The RNN layer in the word feature extraction structure 401 can also be replaced by GRU or LSTM, which can be set according to actual needs, and the present application does not make any limitation on the specific structure of the word feature extraction structure 401.

[0081] After the primary word feature vectors corresponding to each wordpiece in the target text are extracted by the word feature extraction structure 401, the fusion feature processing structure 402 in the barrage recognition model will determine the recognition result corresponding to the target barrage according to the primary word feature vectors corresponding to each wordpiece in the target text and the target feature vector corresponding to the target user. For example, the target feature vector corresponding to the target user can be taken as the first input of the fusion feature processing structure 402, and the primary feature vectors corresponding to each wordpiece are taken as subsequent inputs in sequence according to the order of the wordpieces in the target text; the first full-connection layer in the fusion feature processing structure 402 will adjust the feature vectors corresponding to each wordpiece in the target text based on the target feature vector corresponding to the target user to obtain the word feature vectors corresponding to each wordpiece in the target text; then, the second full-connection layer in the fusion feature processing structure 402 will preliminarily fuse the word feature vectors corresponding to each wordpiece to obtain a preliminary fusion feature vector, and then the convolution layer and the pooling layer in the fusion feature processing structure 402 are used to sequentially perform convolution processing and pooling processing on the preliminary fusion feature vector to obtain a target text feature vector, and finally the third full-connection layer in the fusion feature processing structure 402 is used to process the target text feature vector to obtain the recognition result corresponding to the target barrage.

[0082] It should be understood that Figure 4 It should be understood that the fusion feature processing structure 402 shown is only an example, and in actual applications, other structures of the fusion feature processing structure 402 can also be designed according to actual application requirements, and the specific structure of the fusion feature processing structure 402 is not limited in the present application.

[0083] It should be understood that the recognition result of the target barrage output by the above barrage recognition model can be the probability that the target barrage belongs to bad barrage, and if the probability exceeds the preset probability threshold, it can be determined that the target barrage belongs to bad barrage, otherwise, if the probability does not exceed the preset probability threshold, it can be determined that the target barrage does not belong to bad barrage. Alternatively, the recognition result of the target barrage can also be 1 or 0, and if the recognition result is 1, it indicates that the target barrage belongs to bad barrage, and if the recognition result is 0, it indicates that the target barrage does not belong to bad barrage. The recognition result output by the barrage recognition model is not limited in the present application.

[0084] It should be noted that the target feature vector corresponding to the target user can be obtained by modeling the historical barrage sending situation of the target user, and can reflect the barrage sending characteristics of the target user. As an example, a basic feature vector corresponding to the target user can be obtained by modeling the bad barrage sending rate of the target user, the number of barrages sent in a reference period, and the personal basic information of the target user, as the target feature vector corresponding to the target user. The determination method of the target feature vector will be described in detail below by another method embodiment. As another example, on the basis of the basic feature vector corresponding to the target user, the basic feature vector corresponding to other reference users having a target social relationship with the target user on the target social network platform can be combined to generate the target feature vector corresponding to the target user. The determination method of the target feature vector will be described in detail below by another method embodiment.

[0085] Optionally, in order to improve the model performance of the barrage identification model used in this implementation manner, the server can also update and train the barrage identification model by using the barrages generated in actual application and the determined identification results corresponding to the barrages.

[0086] That is, the server can obtain the barrages generated in a third preset period as second training barrages, and obtain the second training texts of the second training barrages, the target feature vectors corresponding to the users sending the second training barrages, and the identification results corresponding to the second training barrages; then, the second training samples are constructed by using the second training texts, the target feature vectors corresponding to the users sending the second training barrages, and the identification results corresponding to the second training barrages; and then, the barrage identification model is updated and trained based on the second training samples.

[0087] For example, the server can obtain each barrage generated in the past week as a second training barrage. For each second training barrage, the server can extract the barrage text therefrom as a second training text, and obtain the target feature vector corresponding to the user sending the second training barrage, and the identification result corresponding to the second training barrage; and then, the second training sample corresponding to the second training barrage is constructed by using the second training text, the target feature vector corresponding to the user sending the second training barrage, and the identification result corresponding to the second training barrage. After constructing the second training sample corresponding to each second training barrage by the above method, the barrage identification model can be supervised and updated and trained by using the constructed second training samples, so as to improve the identification performance of the barrage identification model.

[0088] It should be understood that the second training barrage is a barrage used when updating the training of the above barrage identification model; the second training sample is a sample constructed based on a second training text in the second training barrage, a target feature vector corresponding to a user sending the second training barrage, and an identification result corresponding to the second training barrage, which is a training sample used when updating the training of the above barrage identification model. The above third preset time period can be set according to actual needs, for example, set to one week, one month, etc., and the present application does not make any limitation on the third preset time period.

[0089] The data processing method provided by the above embodiments of the present application comprehensively considers the text information of the target barrage itself and the user barrage features corresponding to the target user sending the target barrage when identifying whether the target barrage belongs to bad barrage; since the user barrage features are determined according to the historical barrage sending situation of the user, they can reflect the characteristics of the barrage sent by the user, therefore, using the user barrage features corresponding to the user to assist in identifying whether the barrage sent by the user belongs to bad barrage can effectively improve the identification accuracy of bad barrage, avoid the situation of missing identification and misidentification of bad barrage, and further, help to effectively filter bad barrage, avoid showing bad barrage to video watching users, and improve the video watching experience of users.

[0090] Figure 2 It is mentioned in the embodiments shown that in a possible implementation manner provided by the embodiments of the present application, the server can correct the basic probability that the target barrage belongs to bad barrage output by the barrage identification model by using the target bad barrage sending rate corresponding to the target user, so as to obtain the target probability that the target barrage belongs to bad barrage as the identification result corresponding to the target barrage. The determination manner of the target bad barrage sending rate corresponding to the target user will be described in detail below by means of method embodiments.

[0091] Referring to Figure 5 , Figure 5 The flowchart of the method for determining the target bad barrage sending rate provided by the embodiments of the present application is shown. In order to facilitate description, the following embodiments are described by taking the server as an example of the execution subject of the method for determining the target bad barrage sending rate. As Figure 5 shown, the method for determining the target bad barrage sending rate includes the following steps:

[0092] Step 501: determining a user having a target social relationship with the target user on a target social network platform as a reference user associated with the target user.

[0093] Before determining the target bad comment sending rate of the target user, the server can first determine the users having a target social relationship with the target user on the target social network platform as the reference users associated with the target user. In many cases, the comment sending characteristics of such reference users associated with the target user can have certain influence on the comment sending characteristics of the target user itself.

[0094] It should be noted that the target social network platform is a network platform supporting social interaction between the target user and other users. The network platform can support various network service activities based on the Internet technology. In the embodiments of the present application, the target social network platform can be a video playing platform supporting user social interaction. For example, after a user logs in the video playing platform, the user can follow other users on the video playing platform, and even chat with other users on the video playing platform. In addition, the target social network platform can also be a social network platform independent of the video playing platform. The social network platform and the video playing platform can pre-agree a cooperation relationship, and allow the video playing platform to obtain relevant information of users from the social network platform, such as the social relationship of the user on the social network platform. The present application does not make any specific limitation on the above target social network platform.

[0095] It should be noted that the target social relationship can be a pre-set association relationship between users. If the relationship between a user on the target social network platform and the target user is the target social relationship, the user can be determined as a reference user associated with the target user. For example, the target social relationship can include at least one of the following: a one-way following relationship (e.g., the target user one-way follows a user), a mutual following relationship (e.g., the target user and a user mutually follow each other), a relationship having a common followed content, a friend relationship, a relationship having a common user group, and the present application does not make any specific limitation on the above target social relationship.

[0096] In a possible implementation, in order to more intuitively represent the social relationship of the target user on the target social network platform, and represent the association degree between the target user and the reference users associated with the target user, the embodiments of the present application can construct a social relationship graph corresponding to the target user, and configure a corresponding association weight for each edge in the social relationship graph.

[0097] That is, the server can construct a social relationship graph corresponding to the target user according to the target user and the reference users associated with the target user; the social relationship graph includes a target node corresponding to the target user and reference nodes corresponding to the reference users. Then, for the edge between the target node and each reference node in the social relationship graph, the associated weight of the edge is configured according to the association degree between the target user and the reference user corresponding to the reference node, as the associated weight corresponding to the reference user.

[0098] Specifically, the server can create a target node corresponding to the target user, and reference nodes corresponding to each reference user associated with the target user, and then connect the target node with each reference node to obtain a social relationship graph corresponding to the target user. For the edge between the target node and each reference node in the social relationship graph, the server can determine the associated weight corresponding to the edge according to the association degree between the reference user corresponding to the reference node and the target user according to the pre-set weight configuration rule, and take the associated weight corresponding to the edge as the associated weight corresponding to the reference user. The associated weight can represent the closeness of the association relationship between the reference user and the target user. The greater the associated weight corresponding to the reference user, the closer the association relationship between the reference user and the target user, and the smaller the associated weight corresponding to the reference user, the more distant the association relationship between the reference user and the target user.

[0099] As an example, the server can configure the associated weight of the edge between the target node and the reference node according to the association degree between the target user and the reference user corresponding to the reference node in any of the following ways or any combination of multiple ways:

[0100] The associated weight of the edge between the target node and the reference node is configured according to the attention relationship between the target user and the reference user corresponding to the reference node. For example, if the target user unilaterally follows the reference user, the associated weight corresponding to the edge between the target node and the reference node can be configured as 1; if the reference user unilaterally follows the target user, the associated weight corresponding to the edge between the target node and the reference node can be configured as 0.5; if the target user and the reference user follow each other, the associated weight corresponding to the edge between the target node and the reference node can be configured as 2.

[0101] According to the degree of coincidence between the attention content of the target user and the attention content of the reference user corresponding to the reference node, the edge between the target node and the reference node is configured with an associated weight. For example, if the degree of coincidence between the attention content of the target user and the attention content of the reference user is higher than 30%, the associated weight corresponding to the edge between the target node and the reference node can be configured as 1; if the degree of coincidence between the attention content of the target user and the attention content of the reference user is higher than 60%, the associated weight corresponding to the edge between the target node and the reference node can be configured as 2; if the degree of coincidence between the attention content of the target user and the attention content of the reference user is higher than 90%, the associated weight corresponding to the edge between the target node and the reference node can be configured as 3; here, the proportion of the attention content common to the target user and the reference user in all the attention content of the target user and the reference user can be calculated as the above-mentioned degree of coincidence. It should be understood that the above-mentioned attention content includes but is not limited to bloggers, sites, video uploaders, video communities, etc.

[0102] According to the interaction frequency between the target user and the reference user corresponding to the reference node, the edge between the target node and the reference node is configured with an associated weight. For example, if the interaction frequency between the target user and the reference user is higher than 2 times / day, the associated weight corresponding to the edge between the target node and the reference node can be configured as 1; if the interaction frequency between the target user and the reference user is higher than 3 times / day, the associated weight corresponding to the edge between the target node and the reference node can be configured as 2; if the interaction frequency between the target user and the reference user is higher than 4 times / day, the associated weight corresponding to the edge between the target node and the reference node can be configured as 3. It should be understood that the above-mentioned interaction includes but is not limited to comments, likes, private messages, etc.

[0103] According to the degree of coincidence between the group to which the target user belongs and the group to which the reference user corresponding to the reference node belongs, the edge between the target node and the reference node is configured with an associated weight. For example, if the target user and the reference user commonly belong to two groups, the associated weight corresponding to the edge between the target node and the reference node can be configured as 1; if the target user and the reference user commonly belong to three groups, the associated weight corresponding to the edge between the target node and the reference node can be configured as 2; if the target user and the reference user commonly belong to four groups, the associated weight corresponding to the edge between the target node and the reference node can be configured as 3.

[0104] It should be understood that, in actual applications, in addition to the four ways of configuring the association weight described above, other ways of configuring the association weight between the target node and the reference node can also be used according to actual needs, and the application does not make any limitation on the configuration manner of the association weight. In addition, if the server uses multiple ways to configure the association weight between the target node and the reference node at the same time, the association weights determined by each way can be superimposed, for example, the association weight determined by the server according to the attention relationship between the target user and the reference user is 1, and the association weight determined by the server according to the coincidence degree between the attention content of the target user and the attention content of the reference user is 2, then the association weight between the target node and the reference node corresponding to the reference user is 3.

[0105] Step 502: Obtain the basic bad bullet screen sending rate corresponding to the target user, and the basic bad bullet screen sending rate corresponding to each reference user associated with the target user; the basic bad bullet screen sending rate is the proportion of the bad bullet screen sent by the user in the historical reference period in all the bullet screens sent by the user in the historical reference period.

[0106] After the server determines the reference users associated with the target user on the target social network platform, the server can obtain the basic bad bullet screen sending rate corresponding to the target user, and the basic bad bullet screen sending rate corresponding to each reference user. As an example, the server can obtain the basic bad bullet screen sending rate corresponding to the target user and the basic bad bullet screen sending rate corresponding to each reference user from the database for storing the basic bad bullet screen sending rate according to the user identifier corresponding to the target user and the user identifier corresponding to the reference user. As another example, the server can also calculate the basic bad bullet screen sending rate corresponding to the target user and the basic bad bullet screen sending rate corresponding to each reference user in real time. The application does not make any limitation on the way in which the server obtains the basic bad bullet screen sending rate.

[0107] It should be noted that the basic bad bullet screen sending rate is the proportion of the bad bullet screen sent by the user in the historical reference period in all the bullet screens sent by the user in the historical reference period. For example, the proportion of the bad bullet screen sent by the user in the last month in all the bullet screens sent by the user in the last month can be calculated as the basic bad bullet screen sending rate corresponding to the user; for another example, the period from the registration time of the user to the current time can be taken as the historical reference period, and the proportion of all the bad bullet screens sent by the user before this time in all the bullet screens sent by the user before this time can be calculated as the basic bad bullet screen sending rate corresponding to the user. The application does not make any limitation on the historical reference period.

[0108] Step 503: determining the target bad bullet screen sending rate corresponding to the target user according to the basic bad bullet screen sending rate corresponding to the target user and the basic bad bullet screen sending rate corresponding to each reference user associated with the target user.

[0109] After the server obtains the basic bad bullet screen sending rate corresponding to the target user and the basic bad bullet screen sending rate corresponding to each reference user associated with the target user, the server can further determine the target bad bullet screen sending rate corresponding to the target user according to the basic bad bullet screen sending rate corresponding to the target user and the basic bad bullet screen sending rate corresponding to each reference user associated with the target user, so as to correct the basic probability that the target bullet screen output by the bullet screen recognition model belongs to a bad bullet screen by using the target bad bullet screen sending rate corresponding to the target user, and obtain a more standard recognition result corresponding to the target bullet screen.

[0110] As an example, the server can first calculate the sum of the basic bad bullet screen sending rates corresponding to each reference user associated with the target user, and divide the sum by the number of the reference users associated with the target user to obtain a bad bullet screen sending rate correction value; and then calculate the sum of the basic bad bullet screen sending rate corresponding to the target user and the bad bullet screen sending rate correction value as the target bad bullet screen sending rate corresponding to the target user.

[0111] It should be noted that if the association weight corresponding to each reference user is configured according to the association degree between the target user and the reference user, the server can determine the target bad bullet screen sending rate corresponding to the target user according to the basic bad bullet screen sending rate corresponding to the target user, the basic bad bullet screen sending rate corresponding to each reference user associated with the target user, and the association weight corresponding to each reference user.

[0112] As an example, the server can calculate, for each reference user, the product of the basic bad bullet screen sending rate corresponding to the reference user and the association weight corresponding to the reference user as the reference bad bullet screen sending rate corresponding to the reference user. Then, the server calculates the sum of the reference bad bullet screen sending rates corresponding to each reference user, and divides the sum by the sum of the association weights corresponding to each reference user to obtain a bad bullet screen sending rate correction value. Then, the server calculates the sum of the basic bad bullet screen sending rate corresponding to the target user and the bad bullet screen sending rate correction value as the target bad bullet screen sending rate corresponding to the target user.

[0113] It should be understood that in actual applications, the server can also calculate the target bad bullet screen sending rate corresponding to the target user in other ways, and the present application does not limit the calculation method of the target bad bullet screen sending rate corresponding to the target user.

[0114] Optionally, considering that the sending characteristics of the target user may change over time, the technical solution provided by the embodiments of the present application can also update the target undesirable comment sending rate corresponding to the target user; that is, the server can obtain the comments sent by the target user in a second preset time period as first update reference comments, and obtain the respective recognition results corresponding to each of the first update reference comments, and then update the target undesirable comment sending rate corresponding to the target user according to the respective recognition results corresponding to each of the first update reference comments.

[0115] Specifically, the server can obtain each comment sent by the target user in the past month as a first update reference comment, and obtain the respective recognition results corresponding to each of the first update reference comments; then, the basic undesirable comment sending rate corresponding to the target user is updated in combination with the respective recognition results corresponding to each of the first update reference comments. Similarly, for each reference user associated with the target user, the server also needs to update the basic undesirable comment sending rate corresponding to each reference user through the above-mentioned manner. Then, the server can use the updated basic undesirable comment sending rate corresponding to the target user and the updated basic undesirable comment sending rate corresponding to each reference user to recalculate the target undesirable comment sending rate corresponding to the target user. Of course, in this process, the server can also update the association weight corresponding to the reference user.

[0116] The target undesirable comment sending rate corresponding to the target user determined in the above manner is used as the user comment feature corresponding to the target user. The user comment feature corresponding to the target user thus determined fully integrates the undesirable comment sending characteristics of other reference users associated with the target user, so that the user comment feature can comprehensively reflect the undesirable comment sending characteristics of the target user; and the user comment feature is comprehensively considered in identifying whether the comment belongs to undesirable comment, which can effectively improve the recognition accuracy of undesirable comment.

[0117] Figure 2 It is mentioned in the illustrated embodiments that, in a possible implementation manner provided by the embodiments of the present application, the comment recognition model can comprehensively process the target text of the target comment and the target feature vector corresponding to the target user to obtain the recognition result corresponding to the target comment. The determination manner of the target feature vector corresponding to the target user will be described in detail below through method embodiments.

[0118] Referring to Figure 6 , Figure 6 The flowchart of the determination method of the target feature vector provided by the embodiments of the present application is shown. For ease of description, the following embodiments are described by taking the server as an example of the execution subject of the determination method of the target feature vector. As Figure 6 shown, the determination method of the target feature vector includes the following steps:

[0119] Step 601: obtaining the bad comment sending rate corresponding to the target user, the number of comment sending times of the target user in a reference period, and the personal basic information of the target user.

[0120] Before modeling the target feature vector corresponding to the target user, the reference data required for modeling the target feature vector corresponding to the target user, such as the bad comment sending rate corresponding to the target user, the number of comment sending times of the target user in a reference period, and the personal basic information of the target user, need to be obtained.

[0121] As an example, the bad comment sending rate corresponding to the target user can include a basic bad comment sending rate corresponding to the target user and a target bad comment sending rate. Among them, the basic bad comment sending rate corresponding to the target user is the ratio of the number of bad comments sent by the target user in the historical reference period to the total number of comments sent by the target user in the historical reference period, which is mentioned in the embodiment shown in the above Figure 5 The target bad comment sending rate corresponding to the target user is the ratio of the number of bad comments sent by the target user in the historical reference period to the total number of comments sent by the target user in the historical reference period, which is mentioned in the embodiment shown in the above Figure 5 The target bad comment sending rate corresponding to the target user is the ratio of the number of bad comments sent by the target user in the historical reference period to the total number of comments sent by the target user in the historical reference period, which is mentioned in the embodiment shown in the above Figure 5 The determination method of the target bad comment sending rate corresponding to the target user can be referred to in detail in the embodiment shown in the above

[0122] Of course, in actual application, when modeling the target feature vector corresponding to the target user, only one of the basic bad comment sending rate corresponding to the target user and the target bad comment sending rate can be used, and the application does not make any limitation on the bad comment sending rate used when modeling the target feature vector.

[0123] As an example, the number of comment sending times of the target user in the reference period can include the number of comment sending times of the target user in multiple reference periods. For example, when modeling the target feature vector corresponding to the target user, the number of comment sending times in the last one day, the number of comment sending times in the last one week, and the number of comment sending times in the last thirty days can be obtained.

[0124] It should be understood that the above-mentioned reference period can be set according to actual needs, and the reference period to which the required number of comments belongs can also be set according to actual needs, and the application does not make any limitation on the reference period according to which the target feature vector is modeled. Of course, in actual application, when modeling the target feature vector corresponding to the target user, only the number of comment sending times in one reference period can be obtained.

[0125] As an example, the personal basic information of the target user includes at least two of the following: account creation time of the target user, the number of friends of the target user, the amount of content followed by the target user, and the activity level of the target user. The number of friends of the target user can be the number of users that the target user is mutually followed by. The content followed by the target user can include but is not limited to bloggers, sites, video uploaders, video communities, etc. The activity level of the target user can be determined according to the interaction frequency between the target user and other users, the frequency of sending bullet screen comments of the target user, and other factors.

[0126] It should be understood that when modeling the target feature vector corresponding to the target user, in addition to the above-mentioned personal basic information, other personal basic information can also be comprehensively referred to according to actual needs, and the present application does not make any limitation on the personal basic information used when modeling the target feature vector. In addition, when modeling the target feature vector corresponding to the target user, the server can also use only one kind of personal basic information, and the present application does not make any limitation on the number of personal basic information used when modeling the target feature vector.

[0127] Step 602: determining the basic feature vector corresponding to the target user according to the bad bullet screen comment sending rate of the target user, the number of times of sending bullet screen comments of the target user in the reference period, and the personal basic information of the target user through the user bullet screen comment feature modeling model.

[0128] After the server obtains the bad bullet screen comment sending rate of the target user corresponding to the target feature vector, the number of times of sending bullet screen comments of the target user in the reference period, and the personal basic information of the target user, the server can input the bad bullet screen comment sending rate of the target user, the number of times of sending bullet screen comments of the target user in the reference period, and the personal basic information of the target user into the pre-trained user bullet screen comment feature modeling model. The user bullet screen comment feature modeling model analyzes and processes the input bad bullet screen comment sending rate, the number of times of sending bullet screen comments, and the personal basic information, and outputs the basic feature vector corresponding to the target user. The basic feature vector can reflect the bullet screen comment sending characteristics of the target user.

[0129] It should be noted that when the bad bullet screen comment sending rate of the target user corresponding to the target user obtained through step 601 includes the basic bad bullet screen comment sending rate of the target user and the target bad bullet screen comment sending rate, and the number of times of sending bullet screen comments of the target user in the reference period obtained includes the number of times of sending bullet screen comments of the target user in multiple reference periods, and the personal basic information of the target user obtained includes multiple cases, the user bullet screen comment feature modeling model can determine the basic feature vector corresponding to the target user in the following manner:

[0130] The first feature fusion structure in the user barrage feature modeling model is used to determine a first feature vector according to the basic bad barrage sending rate corresponding to the target user and the target bad barrage sending rate.

[0131] As an example, Figure 7 The working principle of the exemplary user barrage feature modeling model provided by the embodiments of the present application is shown in the figure. Figure 7 As shown in the figure, the user barrage feature modeling model includes a first feature fusion structure 701, a second feature fusion structure 702, a third feature fusion structure 703, and a feature modeling structure 704.

[0132] In the case where the obtained bad barrage sending rate corresponding to the target user includes the basic bad barrage sending rate corresponding to the target user and the target bad barrage sending rate, the first feature fusion structure 701 will process the basic bad barrage sending rate corresponding to the target user and the target bad barrage sending rate through two branches each including a full connection layer (FC) and a nonlinear function (NL), and then fuse the feature vectors output by the two branches through another full connection layer and another nonlinear function to obtain the first feature vector.

[0133] In the case where the obtained number of barrage sending times of the target user in multiple reference periods includes the number of barrage sending times of the target user in the last 30 days and the number of barrage sending times of the target user in the last 1 day, the second feature fusion structure 702 will process the number of barrage sending times of the target user in the last 30 days and the number of barrage sending times of the target user in the last 1 day through two branches each including a full connection layer (FC) and a nonlinear function (NL), and then fuse the feature vectors output by the two branches through another full connection layer and another nonlinear function to obtain the second feature vector.

[0134] In a case where the acquired personal basic information of the target user includes the account creation time of the target user, the number of friends of the target user, and the amount of content followed by the target user, the third feature fusion structure 703 will process the account creation time of the target user, the number of friends of the target user, and the amount of content followed by the target user respectively through three branches each including a full connection layer (FC) and a nonlinear function (NL), and then fuse the feature vectors output by the three branches through another full connection layer and another nonlinear function to obtain a third feature vector.

[0135] The feature modeling structure 704 (which can be a full connection layer, for example) in the user barrage feature modeling model will finally determine the basic feature vector corresponding to the target user according to the first feature vector, the second feature vector, and the third feature vector.

[0136] It should be understood that Figure 7 The model structure and working principle of the user barrage feature modeling model shown are only examples. In actual applications, other structure models can also be used as the user barrage feature modeling model, and the user barrage feature modeling model will correspondingly determine the basic feature vector corresponding to the target user based on other working principles. The model structure and working principle of the user barrage feature modeling model are not limited in this application.

[0137] Step 603: Taking the basic feature vector corresponding to the target user as the target feature vector corresponding to the target user.

[0138] In a possible implementation, the server can perform step 603, i.e., directly taking the basic feature vector corresponding to the target user determined through step 602 as the target feature vector corresponding to the target user; so as to use the barrage identification model to comprehensively consider the target text of the target barrage and the target feature vector corresponding to the target user to determine whether the target barrage belongs to bad barrage.

[0139] Step 604: Determining the target feature vector corresponding to the target user according to the basic feature vector corresponding to the target user and the basic feature vectors corresponding to the reference users associated with the target user respectively through the user feature fusion model.

[0140] In another possible implementation, the server can perform step 604 to determine the users having a target social relationship with the target user on the target social network platform as the reference users associated with the target user; then acquire the basic feature vectors corresponding to the reference users associated with the target user respectively; and then determine the target feature vector corresponding to the target user according to the basic feature vector corresponding to the target user and the basic feature vectors corresponding to the reference users associated with the target user respectively through the pre-trained user feature fusion model.

[0141] The way of determining the reference user associated with the target user is the same as the way of determining the reference user associated with the target user in the embodiment shown in the above Figure 5 The detailed content can be referred to the related content in the embodiment shown in the above Figure 5 The way of determining the reference user corresponding to the basic feature vector is the same as the way of determining the target user corresponding to the basic feature vector, that is, through the user barrage feature modeling model, according to the bad barrage sending rate corresponding to the reference user, the number of barrage sending of the reference user in the reference period, and the personal basic information of the reference user, the specific way of determining the reference user corresponding to the basic feature vector can be referred to the related content of step 602 in the above.

[0142] After the server obtains the basic feature vector corresponding to the target user and the basic feature vectors corresponding to each reference user associated with the target user, the server can input the basic feature vector corresponding to the target user and the basic feature vectors corresponding to each reference user associated with the target user into the pre-trained user feature fusion model. After the user feature fusion model analyzes and processes the basic feature vector corresponding to the target user and the basic feature vectors corresponding to each reference user associated with the target user, the target feature vector corresponding to the target user is output.

[0143] As an example, Figure 8 The working principle schematic diagram of the exemplary user feature fusion model provided by the embodiment of the present application is shown. Assuming that the reference users associated with the target user include reference user 1, reference user 2, reference user 3 and reference user 4, the user feature fusion model can process the basic feature vectors corresponding to the target user, reference user 1, reference user 2, reference user 3 and reference user 4 respectively to obtain the target feature vector corresponding to the target user. The user feature fusion model used here can be a graph convolutional network (GCN), of course, in actual application, other structure models can also be used as the user feature fusion model according to actual needs, and the model structure of the user feature fusion model is not limited in the present application.

[0144] Optionally, considering that the barrage sending characteristics of the target user can change over time, the technical solution provided by the embodiment of the present application can also update the target feature vector corresponding to the target user; that is, the server can obtain the barrage sent by the target user in the fourth predetermined period as the second update reference barrage, and obtain the recognition result corresponding to the second update reference barrage; according to the recognition result corresponding to the second update reference barrage, update the target feature vector corresponding to the target user.

[0145] Exemplarily, the server can acquire each piece of the barrage sent by the target user in the last month as the second update reference barrage, and acquire the recognition result corresponding to each piece of the second update reference barrage; then, the server updates the basic bad barrage sending rate of the target user and the target bad barrage sending rate of the target user in combination with the recognition result corresponding to each piece of the second update reference barrage; meanwhile, the server also needs to update the number of barrages sent by the target user in the reference period and the personal basic information of the target user. Further, the server generates the updated basic feature vector corresponding to the target user according to the updated basic bad barrage sending rate and the target bad barrage sending rate of the target user, the number of barrages sent by the target user in the reference period after being updated, and the personal basic information of the target user after being updated, through the user barrage feature modeling model.

[0146] In a possible implementation, the updated basic feature vector corresponding to the target user can be directly used as the updated target feature vector corresponding to the target user.

[0147] In another possible implementation, the server also needs to acquire the updated basic feature vector corresponding to each reference user associated with the target user, and generate the updated target feature vector corresponding to the target user according to the updated basic feature vector corresponding to the target user and the updated basic feature vector corresponding to each reference user associated with the target user, through the user feature fusion model.

[0148] The target feature vector corresponding to the target user determined in the above manner is used as the user barrage feature of the target user. The target feature vector corresponding to the target user determined in this way fuses information that can reflect the barrage sending characteristics of the target user in multiple aspects, and even fuses information that can reflect the barrage sending characteristics of other reference users associated with the target user, so that the target feature vector can more accurately reflect the barrage sending characteristics of the target user. Accordingly, the target feature vector can be comprehensively considered to identify whether the barrage belongs to a bad barrage, which can effectively improve the identification accuracy of bad barrages.

[0149] In order to facilitate further understanding of the technical solutions provided by the embodiments of the present application, the overall implementation process of the data processing method provided by the embodiments of the present application is exemplarily introduced below. The overall implementation process of the data processing method provided by the embodiments of the present application mainly includes the following modules: 1, supervised training module based on bad text labeling data; 2, user behavior collection module; 3, user modeling based on multi-feature fusion of social network; 4, processing result correction module 1 based on the processing result of module 2 or module 3.

[0150] The four modules are introduced respectively as follows.

[0151] 1. Supervised training module based on bad text annotation data

[0152] This module is a supervised task based on strong annotation. The annotation data can be obtained by manual annotation. That is, the annotator can annotate which of the barrage belongs to bad barrage and which of the barrage belongs to normal barrage according to a batch of barrage. The model structure of the barrage identification model trained can be as shown in Figure 3 When training the barrage identification model, the trained barrage identification model can identify whether the input barrage text belongs to bad barrage, and output the prediction result accordingly. Then, the prediction result is compared with the annotation result corresponding to the barrage text, and the barrage identification model is updated according to the difference between the prediction result and the annotation result.

[0153] 2. User behavior collection module

[0154] 2.1 User behavior collection based on bad barrage identification result

[0155] For the barrage sent by the user in the past, the bad barrage included in the barrage can be identified by the barrage identification model trained by module 1. Then, the proportion of the bad barrage sent by the user in all the barrage sent by the user is calculated as the basic bad barrage sending rate corresponding to the user.

[0156] 2.2 Construction of user social relationship graph

[0157] The user social relationship graph can be constructed according to the relationship between users on the target social network platform. In the user social relationship graph, each user corresponds to a node in the social relationship graph, and the edge between the nodes corresponding to two users can reflect the relationship between the two users. If there is no edge between the nodes corresponding to two users, it means that there is no relationship between the two users. The association weight configured on the edge can reflect the closeness of the user relationship (the closer the relationship, the greater the association weight). The relationship between the nodes corresponding to the users established by the application is as follows:

[0158] a) Establish an edge according to the attention relationship between users, that is, an edge can be established between the nodes corresponding to two users as long as the two users have an attention relationship.

[0159] b) If two users follow each other, the association weight of the edge between the nodes corresponding to the two users is configured as 2; if a user follows another user unilaterally, the association weight of the edge between the nodes corresponding to the two users is configured as 1.

[0160] c) If the overlap of the bloggers / sites / video upload users followed by the two users is higher than 30%, an edge is established between the nodes corresponding to the two users. If the overlap of the bloggers / sites / video upload users followed by the two users is higher than 30%, the association weight of the edge between the nodes corresponding to the two users is configured to be 1; if the overlap of the bloggers / sites / video upload users followed by the two users is higher than 60%, the association weight of the edge between the nodes corresponding to the two users is configured to be 2; if the overlap of the bloggers / sites / video upload users followed by the two users is higher than 90%, the association weight of the edge between the nodes corresponding to the two users is configured to be 3.

[0161] The association weights configured in the above manner can be superimposed, for example, if the two users follow each other and the overlap of the bloggers / sites / video upload users followed by the two users is higher than 90%, the association weight of the edge between the nodes corresponding to the two users should be 2+3=5.

[0162] The calculation method of the above overlap is as follows:

[0163] The overlap of the contents followed by users A and B = the amount of contents followed by both users A and B / (the amount of contents followed by user A + the amount of contents followed by user B)

[0164] 2.3 Summary of user behavior combined with social network

[0165] For each user, the present application can summarize the basic bad comment sending rate of a certain user and the basic bad comment sending rate of the reference users corresponding to the adjacent nodes of the node corresponding to the user based on the social relationship graph constructed in 2.2 to obtain the target bad comment sending rate of the user.

[0166] The summary method is as follows:

[0167] The target bad comment sending rate of the user = the basic bad comment sending rate of the user + (the basic bad comment sending rate of the user corresponding to the adjacent node (the association weight of the edge between the current node and the adjacent node i)) / (the association weight of the edge between the current node and the adjacent node i in the social relationship graph)

[0168] Among them, indicates the meaning of summation.

[0169] 3, User modeling combined with multi-feature fusion of social network

[0170] 3.1 Multi-feature fusion module

[0171] ​The module uses the calculated basic bad bullet screen sending rate and target bad bullet screen sending rate as features, and uses the number of bullet screen sending in the last 30 days, the number of bullet screen sending in the last 1 day, account creation time, number of mutual attention friends, amount of attention content, etc. as features. The neural network model fuses these features to obtain the corresponding basic feature vector of the user. The model structure for generating the basic feature vector can be as shown in Figure 7

[0172] 3.2 User representation evolution module combined with social network

[0173] The module evolves the basic feature vector learned in 3.1 according to the user social relationship graph constructed in 2.2. The evolution method is to use a graph convolutional neural network to generate a target feature vector of the current node corresponding to the user based on the basic feature vectors of the users corresponding to adjacent nodes in the user social relationship graph and the basic feature vector of the user corresponding to the current node.

[0174] 4. Correct the processing result of module 1 based on the processing result of module 2 or module 3

[0175] 4.1 Correct the processing result of module 1 based on the target bad bullet screen sending rate of the user generated by module 2

[0176] The target bad bullet screen sending rate calculated by module 2.3 corrects the basic probability of the bullet screen belonging to bad bullet screen output by module 1. The correction method is as follows:

[0177] Target probability = basic probability + a * target bad bullet screen sending rate

[0178] Where a is a correction coefficient set by the developer.

[0179] 4.2 Correct the processing result of module 1 based on the target feature vector of the user generated by module 3

[0180] Input the target feature vector output by module 3.2 into module 1 before the primary word feature vector corresponding to the first word, and then retrain the model to obtain the final prediction result.

[0181] 4.3 Iterative update

[0182] Module 1, module 2 or module 3 can correct each other, so that module 1 can produce better prediction results, and modules 2 and 3 can generate more accurate user representations.

[0183] ​It should be noted that the data processing method provided by the embodiments of the present application can be applied to various application scenarios with bullet screens, for example, identifying bullet screens in various live video streams (such as game live, shopping live, etc.), identifying bullet screens in video resources (such as TV series, movies, variety shows, etc.), and the like. The present application does not make any limitation on the application scenarios of the data processing method.

[0184] In order to facilitate understanding of the data processing method provided by the embodiments of the present application, the overall implementation process of the data processing method will be exemplarily introduced below by taking an example of the data processing method being applied to identifying bad bullet screens in game live.

[0185] In the process of game live, the background server of the live platform can receive the bullet screen sent by the watching user in the game live room, and regard it as the target bullet screen to be identified, and extract the target text from the target bullet screen. In addition, the server can also regard the user sending the target bullet screen in the game live room as the target user, and obtain the user bullet screen feature corresponding to the target user.

[0186] In one possible case, the user bullet screen feature corresponding to the target user can be the target bad bullet screen sending rate corresponding to the target user. Specifically, when the server determines the target bad bullet screen sending rate corresponding to the target user, it can first determine the target social relationship user on the game live platform as the reference user associated with the target user; for example, it can determine the anchor followed by the target user on the game live platform as the reference user associated with the target user, it can also determine the user mutually followed by the target user on the game live platform as the reference user associated with the target user, and it can also determine the user having more common follow content with the target user on the game live platform as the reference user associated with the target user; then, the server can construct a social relationship graph corresponding to the target user according to the target user and the reference user associated with the target user, and configure an association weight between the edge between the target node (corresponding to the target user) and the reference node (corresponding to the reference user) in the social relationship graph as the association weight corresponding to the reference user according to the association degree between the target user and the reference user (which can be determined according to the factors such as the follow relationship, the coincidence degree of the follow content, and the interaction frequency between the target user and the reference user). Further, the server can determine the target bad bullet screen sending rate corresponding to the target user according to the basic bad bullet screen sending rate corresponding to the target user, the basic bad bullet screen rate corresponding to each reference user associated with the target user, and the association weight corresponding to each reference user; here, the basic bad bullet screen sending rate is the proportion of the bad bullet screen sent by the corresponding user on the game live platform in the historical reference period (such as in the past month) in all the bullet screens sent by the corresponding user.

[0187] In this case, the server can first determine the basic probability that the target barrage belongs to the bad barrage according to the target text of the target barrage by using the pre-trained barrage identification model, and then correct the basic probability by using the target bad barrage sending rate of the target user to obtain the identification result corresponding to the target barrage.

[0188] In another possible case, the user barrage feature corresponding to the target user can be a target feature vector corresponding to the target user. Specifically, the server can first generate a basic feature vector corresponding to the target user by using a user barrage feature modeling model according to the basic bad barrage sending rate and the target bad barrage sending rate of the target user, the number of barrage sending times of the target user in the last week and the number of barrage sending times of the target user in the last month on the game live streaming platform, and the personal basic information of the target user (such as the account creation time of the target user on the game live streaming platform, the number of anchors followed by the target user, the number of users mutually followed by the target user, the activity of the target user, and the like); similarly, for each reference user associated with the target user, a similar manner is also used to generate a basic feature vector corresponding to each reference user; and then, by using a user feature fusion model, a target feature vector corresponding to the target user is generated according to the basic feature vector corresponding to the target user and the basic feature vectors corresponding to each reference user associated with the target user.

[0189] In this case, the server can perform feature extraction processing on each word in the target text by using a word feature extraction structure in the barrage identification model to obtain a primary word feature vector corresponding to each word in the target text; and then, by using a fusion feature processing structure in the barrage identification model, the identification result corresponding to the target barrage is determined according to the primary word feature vector corresponding to each word in the target text and the target feature vector corresponding to the target user, that is, whether the target barrage belongs to the bad barrage is determined.

[0190] After the above processing, if the server determines that the target barrage belongs to the bad barrage, the target barrage will be filtered out, that is, the watching users in the game live streaming room will not see the target barrage; if the server determines that the target barrage does not belong to the bad barrage, the server can normally play the target barrage so that the watching users in the game live streaming room can see the target barrage.

[0191] For the data processing method described above, the application also provides a corresponding data processing device to make the above data processing method applicable and realized in practice.

[0192] Referring to Figure 9 , Figure 9 is a structural schematic diagram of a data processing device 900 corresponding to the data processing method shown in the above Figure 2 . As shown in FIG. 9, the data processing device 900 comprises a server 910 and a database 920.Figure 9 As shown in the figure, the data processing apparatus 900 comprises:

[0193] The data acquisition module 901 is configured to acquire, for a target barrage to be identified, target text of the target barrage and user barrage features corresponding to a target user sending the target barrage; the user barrage features are determined according to at least historical barrage sending conditions of the user;

[0194] The barrage identification module 902 is configured to determine, by a barrage identification model, an identification result corresponding to the target barrage according to the target text and the user barrage features corresponding to the target user; the identification result corresponding to the target barrage is used to represent a type of the target barrage.

[0195] Optionally, on the basis of the data processing apparatus as shown in the figure, when the user barrage features corresponding to the target user are a target bad barrage sending rate corresponding to the target user, the barrage identification module 902 is specifically configured to: Figure 9

[0196] Determine, by the barrage identification model, a basic probability that the target barrage belongs to a bad barrage according to the target text;

[0197] Correct the basic probability by using the target bad barrage sending rate corresponding to the target user to obtain a target probability that the target barrage belongs to a bad barrage as the identification result corresponding to the target barrage.

[0198] Optionally, on the basis of the data processing apparatus as shown in the figure, refer to Figure 9 , Figure 10 , Figure 10 Another structure diagram of a barrage identification apparatus 1000 provided by an embodiment of the present application is shown in the figure. As shown in the figure, the barrage identification apparatus further comprises a bad barrage sending rate determination module 1001; the bad barrage sending rate determination module 1001 is configured to: Figure 10

[0199] Determine a user having a target social relationship with the target user on a target social network platform as a reference user associated with the target user;

[0200] Acquire a basic bad barrage sending rate corresponding to the target user and a basic bad barrage sending rate corresponding to each reference user associated with the target user; the basic bad barrage sending rate is a proportion of bad barrages sent by a user in a historical reference period in all barrages sent by the user in the historical reference period;

[0201] ​​According to the basic bad comment sending rate corresponding to the target user and the basic bad comment sending rates corresponding to the reference users respectively, the target bad comment sending rate corresponding to the target user is determined.

[0202] Optionally, on the basis of the data processing apparatus shown in Figure 10 The bad comment sending rate determination module 1001 is specifically configured to:

[0203] According to the target user and the reference users associated with the target user, a social relationship graph corresponding to the target user is constructed; the social relationship graph includes a target node corresponding to the target user and reference nodes corresponding to the reference users;

[0204] For an edge between the target node and each reference node in the social relationship graph, according to the association degree between the target user and the reference user corresponding to the reference node, an association weight is configured for the edge as the association weight corresponding to the reference user;

[0205] According to the basic bad comment sending rate corresponding to the target user, the basic bad comment sending rates corresponding to the reference users respectively, and the association weights corresponding to the reference users respectively, the target bad comment sending rate corresponding to the target user is determined.

[0206] Optionally, on the basis of the data processing apparatus shown in Figure 10 The bad comment sending rate determination module 1001 is specifically configured to configure the association weight for the edge by at least one of the following manners:

[0207] According to the attention relationship between the target user and the reference user, the association weight is configured for the edge;

[0208] According to the coincidence degree between the attention content of the target user and the attention content of the reference user, the association weight is configured for the edge;

[0209] According to the interaction frequency between the target user and the reference user, the association weight is configured for the edge;

[0210] According to the coincidence degree between the group to which the target user belongs and the group to which the reference user belongs, the association weight is configured for the edge.

[0211] Optionally, on the basis of the data processing apparatus shown in Figure 9 The bad comment sending rate determination module 1001 is specifically configured to configure the association weight for the edge by at least one of the following manners: Figure 11 , Figure 11 Another data processing apparatus 1100 provided by the embodiment of the application is shown in a structural schematic diagram. As Figure 11 shown, the data processing apparatus further includes:

[0212] The first model updating training module 1101 is configured to obtain barrage generated in a first preset period as first training barrage, obtain first training text of the first training barrage and a target probability that the first training barrage belongs to bad barrage, construct a first training sample by using the first training text and the target probability that the first training barrage belongs to bad barrage, and perform updating training on the barrage identification model based on the first training sample.

[0213] The bad barrage sending rate updating module 1102 is configured to obtain barrage sent by the target user in a second preset period as first updating reference barrage, obtain an identification result corresponding to the first updating reference barrage, and update the target bad barrage sending rate corresponding to the target user according to the identification result corresponding to the first updating reference barrage.

[0214] Optionally, based on the data processing apparatus shown in Figure 9 When the user barrage feature corresponding to the target user is a target feature vector corresponding to the target user, the barrage identification module 902 is specifically configured to:

[0215] The feature extraction structure in the barrage identification model is used to perform feature extraction processing on each word segmentation in the target text, to obtain a primary word feature vector corresponding to each word segmentation in the target text;

[0216] The fusion feature processing structure in the barrage identification model is used to determine an identification result corresponding to the target barrage according to the primary word feature vector corresponding to each word segmentation in the target text and the target feature vector corresponding to the target user.

[0217] Optionally, based on the data processing apparatus shown in Figure 9 Referring to Figure 12 , Figure 12 FIG. 12 is a structural schematic diagram of another data processing apparatus 1200 provided by the embodiments of the present application. As shown in Figure 12 The data processing apparatus further includes a first feature vector determination module 1201, which is configured to:

[0218] obtain a bad barrage sending rate corresponding to the target user, a barrage sending frequency of the target user in a reference period, and personal basic information of the target user;

[0219] The user barrage feature modeling model is used to determine a basic feature vector corresponding to the target user as a target feature vector corresponding to the target user according to the bad barrage sending rate corresponding to the target user, the barrage sending frequency of the target user in the reference period, and the personal basic information of the target user.

[0220] Optionally, based on the data processing apparatus shown in Figure 12 The target user corresponding bad barrage sending rate includes a target user corresponding basic bad barrage sending rate and a target bad barrage sending rate. The basic bad barrage sending rate is the proportion of bad barrage sent by the user in the historical reference period in all barrage sent by the user in the historical reference period. The target bad barrage sending rate is determined according to the basic bad barrage sending rate corresponding to the user and the basic bad barrage sending rate corresponding to each reference user having a target social relationship with the user on the target social network platform.

[0221] The number of barrage sent by the target user in the reference period includes the number of barrage sent by the target user in multiple reference periods.

[0222] The personal basic information of the target user includes at least two of the following: account creation time of the target user, number of friends of the target user, amount of content followed by the target user, and activity level of the target user.

[0223] The first feature vector determination module 1201 is specifically configured to:

[0224] According to the target user corresponding basic bad barrage sending rate and the target bad barrage sending rate, the first feature vector is determined through the first feature fusion structure in the user barrage feature modeling model.

[0225] According to the number of barrage sent by the target user in multiple reference periods, the second feature vector is determined through the second feature fusion structure in the user barrage feature modeling model.

[0226] According to the personal basic information of the target user, the third feature vector is determined through the third feature fusion structure in the user barrage feature modeling model.

[0227] According to the first feature vector, the second feature vector and the third feature vector, the target user corresponding basic feature vector is determined through the feature modeling structure in the user barrage feature modeling model.

[0228] Optionally, based on the data processing apparatus shown in Figure 9 According to the data processing apparatus shown in Figure 13 , Figure 13 Another data processing apparatus 1300 provided by the embodiment of the present application is shown in the structural schematic diagram. As Figure 13 The data processing apparatus also includes a second feature vector determination module 1301. The second feature vector determination module 1301 is configured to:

[0229] determine a user having a target social relationship with the target user on a target social network platform as a reference user associated with the target user;

[0230] obtain a basic feature vector corresponding to the target user and a basic feature vector corresponding to each reference user associated with the target user; the basic feature vector is determined by a user barrage feature modeling model according to a bad barrage sending rate of a user, a barrage sending frequency of the user in a reference period, and personal basic information of the user;

[0231] determine a target feature vector corresponding to the target user according to the basic feature vector corresponding to the target user and the basic feature vector corresponding to each reference user associated with the target user by a user feature fusion model.

[0232] Optionally, on the basis of the data processing apparatus shown in Figure 9 , see Figure 14 , Figure 14 Another structure schematic diagram of a barrage identification device 1400 provided by an embodiment of the present application is shown in the figure. Figure 14 As shown in the figure, the barrage identification device further comprises:

[0233] The second model update training module 1401 is configured to obtain barrages generated in a third preset period as second training barrages, obtain a second training text of the second training barrages, a target feature vector corresponding to a user sending the second training barrages, and an identification result corresponding to the second training barrages, construct a second training sample by using the second training text, the target feature vector corresponding to the user sending the second training barrages, and the identification result corresponding to the second training barrages, and perform update training on the barrage identification model based on the second training sample.

[0234] The feature vector update module 1402 is configured to obtain barrages sent by the target user in a fourth preset period as second update reference barrages, obtain an identification result corresponding to the second update reference barrages, and update the target feature vector corresponding to the target user according to the identification result corresponding to the second update reference barrages.

[0235] When the data processing apparatus provided in the embodiments of the present application identifies whether the target barrage belongs to the bad barrage, the text information of the target barrage and the user barrage feature corresponding to the target user sending the target barrage are comprehensively considered; since the user barrage feature is determined according to the historical barrage sending situation of the user, the user barrage feature can reflect the characteristics of the barrage sent by the user, therefore, using the user barrage feature corresponding to the user to assist in identifying whether the barrage sent by the user belongs to the bad barrage can effectively improve the identification accuracy of the bad barrage, avoid the situation of missing identification and misidentification of the bad barrage, and further, help to effectively filter the bad barrage, avoid showing the bad barrage to the video watching user, and improve the video watching experience of the user.

[0236] The embodiments of the present application also provide a device for identifying a barrage, and the device can be a terminal device or a server. In the following, the terminal device and the server provided in the embodiments of the present application will be introduced from the perspective of hardware embodiment.

[0237] Referring to Figure 15 , Figure 15 is a structural schematic diagram of a terminal device provided in the embodiments of the present application. As Figure 15 indicated, only parts related to the embodiments of the present application are shown for the convenience of description, and specific technical details not disclosed are referred to the method part of the embodiments of the present application. The terminal can be any terminal device including a mobile phone, a tablet computer, a personal digital assistant (English full name: Personal Digital Assistant, English abbreviation: PDA), a point of sales (English full name: Point of Sales, English abbreviation: POS), a vehicle-mounted computer, etc. Taking the terminal as a computer for example:

[0238] Figure 15 shown is a block diagram of part of the structure of the computer related to the terminal provided in the embodiments of the present application. Referring to Figure 15 , the computer includes a radio frequency (English full name: Radio Frequency, English abbreviation: RF) circuit 1510, a memory 1520, an input unit 1530 (including a touch panel 1531 and other input devices 1532), a display unit 1540 (including a display panel 1541), a sensor 1550, an audio circuit 1560 (which can be connected to a speaker 1561 and a microphone 1562), a wireless fidelity (English full name: wireless fidelity, English abbreviation: WiFi) module 1570, a processor 1580, and a power supply 1590, etc. Those skilled in the art can understand Figure 15 that the computer structure shown in the figure does not constitute a limitation to the computer, and can include more or fewer components than the figure, or combine certain components, or different component arrangements.

[0239] The memory 1520 can be used to store software programs and modules, and the processor 1580 executes various functions and processes of the computer by running the software programs and modules stored in the memory 1520. The memory 1520 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the computer (such as audio data, a telephone book, etc.), and the like. In addition, the memory 1520 can include a high-speed random access memory, and can also include a nonvolatile memory such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device.

[0240] The processor 1580 is the control center of the computer, connects all parts of the computer through various interfaces and lines, and executes various functions and processes data of the computer by running or executing the software programs and / or modules stored in the memory 1520 and calling the data stored in the memory 1520. Optionally, the processor 1580 can include one or more processing units; preferably, the processor 1580 can integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, and an application program, and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 1580.

[0241] In the embodiment of the present application, the processor 1580 included in the terminal further has the following functions:

[0242] For the target barrage to be identified, the target text of the target barrage and the user barrage feature corresponding to the target user sending the target barrage are acquired; the user barrage feature is determined at least according to the historical barrage sending situation of the user;

[0243] Through the barrage identification model, the target text and the user barrage feature corresponding to the target user are used to determine the identification result corresponding to the target barrage; the identification result corresponding to the target barrage is used to represent the type of the target barrage.

[0244] Optionally, the processor 1580 is further configured to execute the steps of any one of the implementation manners of the data processing method provided in the embodiments of the present application.

[0245] Referring to Figure 16 , Figure 16A structural schematic diagram of a server 1600 is provided in embodiments of the present application. The server 1600 can have great differences due to different configurations or performances, and can include one or more central processing units (CPU) 1622 (for example, one or more processors) and a memory 1632, one or more storage media 1630 (for example, one or more mass storage devices) storing application programs 1642 or data 1644. The memory 1632 and the storage media 1630 can be temporary storage or persistent storage. The programs stored in the storage media 1630 can include one or more modules (not shown in the figure), each of which can include a series of instruction operations in the server. Further, the central processing unit 1622 can be configured to communicate with the storage media 1630 and execute the series of instruction operations in the storage media 1630 on the server 1600.

[0246] The server 1600 can also include one or more power supplies 1626, one or more wired or wireless network interfaces 1650, one or more input / output interfaces 1658, and / or one or more operating systems, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0247] The steps performed by the server in the above embodiments can be based on the Figure 16 The server structure is shown.

[0248] The CPU 1622 is configured to perform the following steps:

[0249] For a target barrage to be identified, a target text of the target barrage and a user barrage feature corresponding to a target user sending the target barrage are obtained; the user barrage feature is determined at least according to a historical barrage sending situation of the user;

[0250] A recognition result corresponding to the target barrage is determined by a barrage recognition model according to the target text and the user barrage feature corresponding to the target user; the recognition result corresponding to the target barrage is used to represent a type of the target barrage.

[0251] Optionally, the CPU 1622 can also be configured to perform the steps of any one of the implementation manners of the data processing method provided in embodiments of the present application.

[0252] Embodiments of the present application also provide a computer readable storage medium for storing a computer program, which is used to execute any one of the implementation manners of the data processing method described in the foregoing embodiments.

[0253] The embodiments of the present application further provide a computer program product or computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes any one of the data processing methods according to the foregoing embodiments.

[0254] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the system, device and unit described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0255] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented by other manners. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0256] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0257] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically independently, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of software functional unit.

[0258] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (English full name: Read-Only Memory, English abbreviation: ROM), a random access memory (English full name: Random Access Memory, English abbreviation: RAM), a magnetic disk or an optical disk, and various computer program storage media.

[0259] It should be understood that in the present application, "at least one" means one or more, and "multiple" means two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three kinds of relationships, for example, "A and / or B" can represent: only A, only B, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally represents that the associated objects before and after are in an "or" relationship. "At least one of the following" or the like means any combination of these items, including any combination of single item or multiple items. For example, at least one of a, b or c, can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0260] The above-described and above-mentioned embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A data processing method, characterized by, The method comprises: For a target barrage to be identified, obtaining target text of the target barrage and user barrage features corresponding to a target user sending the target barrage; the user barrage features are determined at least according to historical barrage sending conditions of the user; According to the target text and the user barrage features corresponding to the target user, determining, by a barrage identification model, an identification result corresponding to the target barrage; the identification result corresponding to the target barrage is used to represent a type of the target barrage; When the user barrage features corresponding to the target user are a target bad barrage sending rate corresponding to the target user, the determination of the identification result corresponding to the target barrage by the barrage identification model according to the target text and the user barrage features corresponding to the target user comprises: According to the target text, determining, by the barrage identification model, a basic probability that the target barrage belongs to a bad barrage; Using the target bad barrage sending rate corresponding to the target user to correct the basic probability to obtain a target probability that the target barrage belongs to a bad barrage as the identification result corresponding to the target barrage; The target bad barrage sending rate corresponding to the target user is determined by: Determining users having a target social relationship with the target user on a target social network platform as reference users associated with the target user; Obtaining a basic bad barrage sending rate corresponding to the target user and a basic bad barrage sending rate corresponding to each reference user associated with the target user; the basic bad barrage sending rate is a proportion of bad barrages sent by a user in a historical reference period in all barrages sent by the user in the historical reference period; According to the basic bad barrage sending rate corresponding to the target user and the basic bad barrage sending rate corresponding to each reference user, determining a target bad barrage sending rate corresponding to the target user.

2. The method of claim 1, wherein, The method further comprises: According to the target user and the reference users associated with the target user, constructing a social relationship graph corresponding to the target user; the social relationship graph comprises a target node corresponding to the target user and reference nodes corresponding to the reference users; For an edge between the target node and each reference node in the social relationship graph, according to an association degree between the target user and the reference user corresponding to the reference node, configuring an association weight for the edge between the target node and the reference node as an association weight corresponding to the reference user; According to the basic bad barrage sending rate corresponding to the target user and the basic bad barrage sending rate corresponding to each reference user, determining a target bad barrage sending rate corresponding to the target user comprises: According to the basic bad barrage sending rate corresponding to the target user, the basic bad barrage sending rate corresponding to each reference user, and the association weight corresponding to each reference user, determining a target bad barrage sending rate corresponding to the target user.

3. The method of claim 2, wherein, The association weight of the edge is configured according to at least one of the following: The association weight of the edge is configured according to a following relationship between the target user and the reference user: The association weight of the edge is configured according to a coincidence degree between the attention content of the target user and the attention content of the reference user; The association weight of the edge is configured according to an interaction frequency between the target user and the reference user; The association weight of the edge is configured according to a coincidence degree between a group to which the target user belongs and a group to which the reference user belongs.

4. The method of claim 1, wherein, The method further comprises: obtaining a first training text and a target probability that the first training text belongs to bad bullet screen in a first preset period as a first training bullet screen; constructing a first training sample by using the first training text and the target probability that the first training text belongs to bad bullet screen; and updating and training the bullet screen identification model based on the first training sample. obtaining a first update reference bullet screen sent by the target user in a second preset period as a first update reference bullet screen, and obtaining an identification result corresponding to the first update reference bullet screen; and updating a target bad bullet screen sending rate corresponding to the target user according to the identification result corresponding to the first update reference bullet screen.

5. The method of claim 1, wherein, When the user bullet screen feature corresponding to the target user is a target feature vector corresponding to the target user, the identification result corresponding to the target bullet screen is determined by the bullet screen identification model according to the target text and the user bullet screen feature corresponding to the target user, comprising: performing feature extraction processing on each word in the target text by a word feature extraction structure in the bullet screen identification model to obtain a primary word feature vector corresponding to each word in the target text; determining the identification result corresponding to the target bullet screen according to the primary word feature vector corresponding to each word in the target text and the target feature vector corresponding to the target user by a fusion feature processing structure in the bullet screen identification model.

6. The method of claim 5, wherein, The target feature vector corresponding to the target user is determined by: obtaining a target bad bullet screen sending rate corresponding to the target user, a number of bullet screens sent by the target user in a reference period, and personal basic information of the target user; determining a basic feature vector corresponding to the target user as the target feature vector corresponding to the target user by a user bullet screen feature modeling model according to the target bad bullet screen sending rate corresponding to the target user, the number of bullet screens sent by the target user in the reference period, and the personal basic information of the target user.

7. The method of claim 6, wherein: the target bad bullet screen sending rate corresponding to the target user comprises a basic bad bullet screen sending rate and a target bad bullet screen sending rate corresponding to the target user; and the basic bad bullet screen sending rate is a proportion of bad bullet screens sent by a user in a historical reference period in all bullet screens sent by the user in the historical reference period. The target undesirable barrage sending rate is determined according to a basic undesirable barrage sending rate corresponding to the user and a basic undesirable barrage sending rate corresponding to each reference user having a target social relationship with the user on the target social network platform; The barrage sending frequency of the target user in the reference period includes barrage sending frequencies of the target user in multiple reference periods; The personal basic information of the target user includes at least two of the following: account creation time of the target user, number of friends of the target user, amount of content followed by the target user, and activity level of the target user; Then, the basic feature vector corresponding to the target user is determined by a user barrage feature modeling model according to the undesirable barrage sending rate corresponding to the target user, the barrage sending frequency of the target user in the reference period, and the personal basic information of the target user, including: A first feature vector is determined by a first feature fusion structure in the user barrage feature modeling model according to the basic undesirable barrage sending rate corresponding to the target user and the target undesirable barrage sending rate; A second feature vector is determined by a second feature fusion structure in the user barrage feature modeling model according to the barrage sending frequencies of the target user in multiple reference periods; A third feature vector is determined by a third feature fusion structure in the user barrage feature modeling model according to the personal basic information of the target user; The basic feature vector corresponding to the target user is determined by a feature modeling structure in the user barrage feature modeling model according to the first feature vector, the second feature vector, and the third feature vector.

8. The method of claim 5, wherein, The target feature vector corresponding to the target user is determined in the following way: Users having a target social relationship with the target user on a target social network platform are determined as reference users associated with the target user; The basic feature vector corresponding to the target user and the basic feature vector corresponding to each reference user associated with the target user are obtained; the basic feature vector is determined by a user barrage feature modeling model according to the undesirable barrage sending rate corresponding to the user, the barrage sending frequency of the user in the reference period, and the personal basic information of the user; The target feature vector corresponding to the target user is determined by a user feature fusion model according to the basic feature vector corresponding to the target user and the basic feature vector corresponding to each reference user associated with the target user.

9. The method of claim 5, wherein, The method further includes: The barrage generated in a third predetermined period is obtained as second training barrage, the second training text of the second training barrage, the target feature vector corresponding to the user sending the second training barrage, and the recognition result corresponding to the second training barrage are obtained; the second training text, the target feature vector corresponding to the user sending the second training barrage, and the recognition result corresponding to the second training barrage are used to construct a second training sample; based on the second training sample, the barrage recognition model is updated and trained; Acquire the barrage sent by the target user in a fourth preset time period as second updated reference barrage, and acquire the recognition result corresponding to the second updated reference barrage; update the target feature vector corresponding to the target user according to the recognition result corresponding to the second updated reference barrage.

10. A data processing apparatus, characterized by, The apparatus comprises: a data acquisition module configured to, for a target barrage to be recognized, acquire target text of the target barrage and a user barrage feature corresponding to a target user sending the target barrage; the user barrage feature being determined according to at least historical barrage sending conditions of a user; a barrage recognition module configured to determine, by a barrage recognition model, a recognition result corresponding to the target barrage according to the target text and the user barrage feature corresponding to the target user; the recognition result corresponding to the target barrage being used to represent a type of the target barrage; wherein, when the user barrage feature corresponding to the target user is a target bad barrage sending rate corresponding to the target user, the barrage recognition module is specifically configured to: determine, by the barrage recognition model, a basic probability that the target barrage belongs to a bad barrage according to the target text; correct the basic probability by using the target bad barrage sending rate corresponding to the target user to obtain a target probability that the target barrage belongs to a bad barrage as the recognition result corresponding to the target barrage; the apparatus further comprises a bad barrage sending rate determination module, the bad barrage sending rate determination module is configured to: determine a user having a target social relationship with the target user on a target social network platform as a reference user associated with the target user; acquire a basic bad barrage sending rate corresponding to the target user and a basic bad barrage sending rate corresponding to each reference user associated with the target user; the basic bad barrage sending rate being a proportion of bad barrages sent by a user in a historical reference period in all barrages sent by the user in the historical reference period; determine a target bad barrage sending rate corresponding to the target user according to the basic bad barrage sending rate corresponding to the target user and the basic bad barrage sending rate corresponding to each reference user.

11. The apparatus of claim 10, wherein, the bad barrage sending rate determination module is specifically configured to: construct a social relationship graph corresponding to the target user according to the target user and the reference user associated with the target user; the social relationship graph comprising a target node corresponding to the target user and a reference node corresponding to the reference user; for an edge between the target node and each reference node in the social relationship graph, configure an association weight between the target node and the reference node as an association weight corresponding to the reference user according to an association degree between the target user and the reference user corresponding to the reference node; determine the target bad barrage sending rate corresponding to the target user according to the basic bad barrage sending rate corresponding to the target user, the basic bad barrage sending rate corresponding to each reference user, and the association weight corresponding to each reference user.

12. The apparatus of claim 11, wherein, The bad barrage sending rate determination module is specifically configured to configure an association weight for the edge by at least one of the following manners: configuring the association weight for the edge according to a following relationship between the target user and the reference user: a following relationship between the target user and the reference user; a following relationship between the target user and the reference user; a following relationship between the target user and the reference user; 13. The apparatus of claim 10, wherein, a following relationship between the target user and the reference user. The device further comprises: The first model update training module is configured to obtain barrages generated in a first preset period as first training barrages, obtain first training texts of the first training barrages and target probabilities that the first training barrages belong to bad barrages, construct first training samples by using the first training texts and the target probabilities that the first training barrages belong to bad barrages, and perform update training on the barrage identification model based on the first training samples.

14. The apparatus of claim 10, wherein, The bad barrage sending rate update module is configured to obtain barrages sent by the target user in a second preset period as first update reference barrages, obtain identification results corresponding to the first update reference barrages, and update the target bad barrage sending rate corresponding to the target user according to the identification results corresponding to the first update reference barrages. When the user barrage feature corresponding to the target user is a target feature vector corresponding to the target user, the barrage identification module is specifically configured to: perform feature extraction processing on each word in the target text by using a word feature extraction structure in the barrage identification model to obtain a primary word feature vector corresponding to each word in the target text; 15. The apparatus of claim 14, wherein, determine an identification result corresponding to the target barrage according to the primary word feature vector corresponding to each word in the target text and the target feature vector corresponding to the target user by using a fusion feature processing structure in the barrage identification model. The device further comprises a first feature vector determination module, which is configured to: obtain a bad barrage sending rate corresponding to the target user, a barrage sending frequency of the target user in a reference period, and personal basic information of the target user; 16. The apparatus of claim 15, wherein, determine a basic feature vector corresponding to the target user as the target feature vector corresponding to the target user by using a user barrage feature modeling model according to the bad barrage sending rate corresponding to the target user, the barrage sending frequency of the target user in the reference period, and the personal basic information of the target user. The bad barrage sending rate corresponding to the target user comprises a basic bad barrage sending rate corresponding to the target user and a target bad barrage sending rate; the basic bad barrage sending rate is a proportion of bad barrages sent by a user in a historical reference period in all barrages sent by the user in the historical reference period. The target bad barrage sending rate is determined according to a basic bad barrage sending rate corresponding to the user and basic bad barrage sending rates corresponding to respective reference users having a target social relationship with the user on a target social network platform; The barrage sending frequency of the target user in a reference period includes barrage sending frequencies of the target user in multiple reference periods; The personal basic information of the target user includes at least two of the following: an account creation time of the target user, a number of friends of the target user, a content amount followed by the target user, and an activity level of the target user; The first feature vector determination module is specifically configured to: determine a first feature vector according to the basic bad barrage sending rate corresponding to the target user and the target bad barrage sending rate through a first feature fusion structure in the user barrage feature modeling model; determine a second feature vector according to the barrage sending frequencies of the target user in multiple reference periods through a second feature fusion structure in the user barrage feature modeling model; determine a third feature vector according to the personal basic information of the target user through a third feature fusion structure in the user barrage feature modeling model; determine the basic feature vector corresponding to the target user according to the first feature vector, the second feature vector, and the third feature vector through a feature modeling structure in the user barrage feature modeling model.

17. The apparatus of claim 14, wherein, The device further includes a second feature vector determination module configured to: determine users having a target social relationship with the target user on a target social network platform as reference users associated with the target user; obtain a basic feature vector corresponding to the target user and basic feature vectors corresponding to respective reference users associated with the target user; the basic feature vector is determined by a user barrage feature modeling model according to a bad barrage sending rate corresponding to a user, a barrage sending frequency of the user in a reference period, and personal basic information of the user; determine a target feature vector corresponding to the target user according to the basic feature vector corresponding to the target user and the basic feature vectors corresponding to respective reference users associated with the target user through a user feature fusion model.

18. The apparatus of claim 14, wherein, The device further includes: a second model update training module configured to obtain barrages generated in a third preset period as second training barrages, obtain second training texts of the second training barrages, target feature vectors corresponding to users sending the second training barrages, and recognition results corresponding to the second training barrages, construct a second training sample by using the second training texts, the target feature vectors corresponding to the users sending the second training barrages, and the recognition results corresponding to the second training barrages, and update train the barrage recognition model based on the second training sample. The feature vector updating module is configured to acquire a barrage sent by the target user in a fourth preset time period as second update reference barrage, acquire an identification result corresponding to the second update reference barrage, and update a target feature vector corresponding to the target user according to the identification result corresponding to the second update reference barrage.

19. An apparatus, comprising: The device comprises a processor and a memory; The memory is configured to store a computer program; The processor is configured to execute the data processing method according to any one of claims 1 to 9 according to the computer program.

20. A computer-readable storage medium, characterized in that, The computer readable storage medium is configured to store a computer program, and the computer program is configured to execute the data processing method according to any one of claims 1 to 9.

21. A computer program product, characterised in that, The computer program product comprises computer instructions, and a processor of a computer device executes the computer instructions, so that the computer device executes the data processing method according to any one of claims 1 to 9.

Images