Object classification method and device, computer equipment, storage medium and program product

Abstract

The invention relates to the technical field of computers, and discloses an object classification method and device, computer equipment, a storage medium and a program product.The object classification method comprises the steps that to-be-analyzed comments issued by to-be-classified objects are taken; the to-be-analyzed comments are analyzed based on preset association features, the to-be-classified objects are classified according to the analysis result, an object set is obtained, and the preset association features are used for indicating abnormal association existing between the to-be-analyzed comments; and performing abnormal feature analysis based on each to-be-classified object in the object set to obtain an abnormal object, and adjusting the object set according to the abnormal object to obtain a target set. According to the method, the object groups of the abnormal comments are found and divided, the confidence coefficient of the recognition result is improved, and meanwhile the recognition efficiency for the abnormal objects is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to an object classification method, apparatus, computer equipment, storage medium, and program product. Background Art

[0002] Social platforms usually support comment functions to enable users to comment on interactive content, thereby enhancing user interaction, improving content value, and optimizing user experience. However, comments often contain abnormal comments. For example, a group of users posts a large amount of homogeneous content for a certain purpose, which reduces the overall content quality of the comment area and affects the user experience of normal users.

[0003] In related solutions for identifying abnormal comment objects, network models are usually learned based on object information, UGC (User Generated Content) features, etc. to distinguish normal objects from abnormal objects and mark abnormal objects. However, this single object dimension recognition method leads to a large amount of information loss during network model training and object recognition, affecting the confidence of the recognition results and reducing the recognition efficiency of abnormal objects. Summary of the invention

[0004] In view of this, the present disclosure provides an object classification method, apparatus, computer device, storage medium and program product to solve the problem that a single recognition method for object dimensions leads to poor confidence in network model recognition results and low recognition efficiency.

[0005] In a first aspect, the present disclosure provides an object classification method, the method comprising:

[0006] Get the comments to be analyzed posted by the object to be classified;

[0007] Analyze the comments to be analyzed based on the preset correlation features, and classify the objects to be classified according to the analysis results to obtain an object collection, wherein the preset correlation features are used to indicate abnormal correlations between the comments to be analyzed;

[0008] An abnormal feature analysis is performed on each object to be classified in the object collection to obtain an abnormal object, and the object collection is adjusted according to the abnormal object to obtain a target collection.

[0009] In a second aspect, the present disclosure provides an object classification device, the device comprising:

[0010] An acquisition module is used to obtain the comments to be analyzed posted by the object to be classified;

[0011] A classification module, used to analyze the comments to be analyzed based on preset correlation features, and classify the objects to be classified according to the analysis results to obtain an object collection, wherein the preset correlation features are used to indicate abnormal correlations between the comments to be analyzed;

[0012] The analysis module is used to perform abnormal feature analysis based on each object to be classified in the object collection to obtain abnormal objects, and adjust the object collection according to the abnormal objects to obtain the target collection.

[0013] In a third aspect, the present disclosure provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the object classification method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0014] In a fourth aspect, the present disclosure provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the object classification method of the first aspect or any corresponding embodiment thereof.

[0015] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the object classification method of the first aspect or any corresponding embodiment thereof.

[0016] In the disclosed embodiment, the comments to be analyzed posted by the objects to be classified can be first obtained, and the comments to be analyzed can be analyzed based on the preset correlation features, so as to classify the objects to be classified according to the analysis results and obtain an object collection, wherein the preset correlation features are used to indicate the abnormal correlations between the comments to be analyzed. Then, abnormal feature analysis can be performed based on each object to be classified in the object collection to obtain abnormal objects, and the object collection can be adjusted according to the abnormal objects to obtain a target collection, thereby discovering and dividing the object groups of abnormal comments, improving the confidence of the recognition results, and at the same time improving the recognition efficiency for abnormal objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the specific embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 is a flowchart of an object classification method according to an embodiment of the present disclosure;

[0019] Figure 2is a schematic diagram of a process of determining an object group according to a rough call model of an embodiment of the present disclosure;

[0020] Figure 3 is a schematic diagram of a process for determining a target collection according to an embodiment of the present disclosure;

[0021] Figure 4 is a schematic diagram of a process of identifying a target object by using a second graph classification model according to an embodiment of the present disclosure;

[0022] Figure 5 is a schematic diagram of a process of training a second image classification model according to an embodiment of the present disclosure;

[0023] Figure 6 is a structural block diagram of an object classification device according to an embodiment of the present disclosure;

[0024] Figure 7 It is a schematic diagram of the hardware structure of the computer device of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0026] In conjunction with the application scenarios on which the execution of the object classification method depends, the application scenarios are described here.

[0027] Social platforms usually support comment functions to enable users to comment on interactive content, thereby enhancing user interaction, improving content value, and optimizing user experience. However, comments often contain abnormal comments. For example, a group of users posts a large amount of homogeneous content for a certain purpose, which reduces the overall content quality of the comment area and affects the user experience of normal users.

[0028] In related solutions for identifying abnormal comment objects, network models are usually learned based on object information, UGC (User Generated Content) features, etc. to distinguish normal objects from abnormal objects and mark abnormal objects. However, this single object dimension recognition method leads to a large amount of information loss during network model training and object recognition, affecting the confidence of the recognition results and reducing the recognition efficiency of abnormal objects.

[0029] For example, in the relevant scheme for identifying abnormal comment objects, non-graph algorithms can be used to identify single object dimensions, while ignoring the group attributes of the object, resulting in a large amount of group loss. In addition, in the current work of identifying abnormal comment objects using graph algorithms, the usual practice is to directly identify abnormal groups based on some high-standard edge relationships and various community partitioning algorithms. However, in this mode, due to the insufficient accuracy of some edge relationships, many objects in abnormal groups cannot be associated, which affects the confidence of the final recognition result.

[0030] Based on this, the disclosed embodiment provides an object classification method, which first obtains the comments to be analyzed posted by the object to be classified, and analyzes the comments to be analyzed based on preset association features, so as to classify the object to be classified according to the analysis results to obtain an object collection, wherein the preset association features are used to indicate the abnormal associations between the comments to be analyzed. Then, abnormal feature analysis can be performed based on each object to be classified in the object collection to obtain abnormal objects, and the object collection can be adjusted according to the abnormal objects to obtain a target collection, thereby discovering and dividing the object groups of abnormal comments, improving the confidence of the recognition results, and at the same time improving the recognition efficiency for abnormal objects.

[0031] According to an embodiment of the present disclosure, an object classification method embodiment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0033] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0034] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0035] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0036] It is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.

[0037] According to an embodiment of the present disclosure, an object classification method embodiment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] In this embodiment, an object classification method is provided, which can be used for a service terminal. Figure 1 is a flow chart of an object classification method according to an embodiment of the present disclosure, such as Figure 1 As shown, the process includes the following steps:

[0039] Step S101, obtaining comments to be analyzed posted by the object to be classified.

[0040] In an embodiment of the present disclosure, the object to be classified may be an object that has posted comments on a target social platform within a preset historical period. Here, the target social platform may support the creative object to post interactive content therein, and other objects within the platform may perform interactive operations on the interactive content, such as commenting.

[0041] When obtaining comments to be analyzed, the comment contents under all or part of the interactive contents in the target social platform may be obtained, and the objects to be classified corresponding to each comment content may be determined to identify whether there are any abnormalities in the objects to be classified.

[0042] Step S102 , analyzing the comments to be analyzed based on preset correlation features, and classifying the objects to be classified according to the analysis results to obtain an object collection, wherein the preset correlation features are used to indicate abnormal correlations between the comments to be analyzed.

[0043] In the disclosed embodiment, the preset association feature may be used to indicate the association feature existing between objects in an abnormal group. For example, the preset association feature may be used to indicate a group of objects that publish the same content under the same type of video.

[0044] When analyzing the comments to be analyzed based on the preset correlation features, comments to be analyzed with abnormal correlation can be obtained. For example, multiple comments to be analyzed with the same content correspond to the same object to be analyzed. Then, the comments to be analyzed with abnormal correlation can be associated with the object to be classified that posted the comments to obtain an object collection.

[0045] Step S103, performing abnormal feature analysis based on each object to be classified in the object collection to obtain abnormal objects, and adjusting the object collection according to the abnormal objects to obtain a target collection.

[0046] In an embodiment of the present disclosure, an object collection may be used to indicate an abnormal object group, for example, the abnormal object group publishes a large amount of homogeneous content for the purpose of "drawing traffic" to guide other objects to browse or purchase something.

[0047] After the object collection is determined through rough summoning, the object collection can be purified to accurately identify abnormal objects, thereby filtering out objects that are mistakenly recalled due to normal interactive behaviors. Specifically, the object features corresponding to each object to be classified in the collection can be introduced to further classify the object to be classified according to the object features using the high-precision model to obtain abnormal objects.

[0048] After identifying abnormal objects in the object collection, non-abnormal objects can be screened out, and the associations between abnormal objects in the object collection can be adjusted according to the screening results to obtain a target collection. It should be understood that abnormal objects in the social platform can be further mined according to the target collection to improve the efficiency of identifying abnormal objects.

[0049] It can be seen from the above description that in the disclosed embodiment, the comments to be analyzed posted by the objects to be classified can be first obtained, and the comments to be analyzed can be analyzed based on the preset association features, so as to classify the objects to be classified according to the analysis results and obtain an object collection, wherein the preset association features are used to indicate the abnormal associations between the comments to be analyzed. Then, abnormal feature analysis can be performed based on each object to be classified in the object collection to obtain abnormal objects, and the object collection can be adjusted according to the abnormal objects to obtain a target collection, thereby discovering and dividing the object groups of abnormal comments, improving the confidence of the recognition results, and at the same time improving the recognition efficiency for abnormal objects.

[0050] In some optional implementations, the above step S102 analyzes the comments to be analyzed based on the preset correlation features, and classifies the objects to be classified according to the analysis results to obtain an object collection, including:

[0051] Step S1021, analyzing the comments to be analyzed based on preset correlation features, obtaining the comments to be analyzed that hit the preset correlation features, and constructing a heterogeneous graph based on the objects to be classified corresponding to the comments to be analyzed, wherein the nodes in the heterogeneous graph are used to indicate the objects to be classified, and the edges between the nodes are used to indicate the preset correlation features between the nodes.

[0052] Step S1022, classify the heterogeneous graph according to the preset comment scenario based on the first graph classification model to obtain multiple homogeneous graphs, and determine the object collection based on the homogeneous graphs, wherein the nodes in the homogeneous graph are used to indicate the corresponding objects to be classified in the current comment scenario, and the edges between the nodes are used to indicate the preset association features between the nodes.

[0053] In the embodiment of the present disclosure, the first graph classification model may be a rough recall model, which may use the above-mentioned preset association features to recall the abnormal group and obtain an object collection. Here, Figure 2 The figure shows a flow chart of the rough recall model for determining the object group, wherein each preset comment scenario can be recalled separately to obtain the corresponding object collection.

[0054] Here, first, a heterogeneous graph of objects to be classified for input into the first graph classification model may be constructed. Specifically, the above step S1021 analyzes the comments to be analyzed based on the preset correlation features to obtain the comments to be analyzed that hit the preset correlation features, and constructs a heterogeneous graph based on the objects to be classified corresponding to the comments to be analyzed, including:

[0055] Step a1, based on the comments to be analyzed, determining the interactive link characteristics of the object to be classified, wherein the interactive link characteristics are used to indicate the operation link of the object to be classified to publish the comments to be analyzed.

[0056] Step a2, obtaining the comment content features of the comments to be analyzed and the object information features corresponding to the objects to be classified, and determining the target comments that hit the preset association features in the comments to be analyzed based on the interactive link features, comment content features and object information features.

[0057] Step a3, based on the preset association features corresponding to the target comments, the objects to be classified are associated to obtain a heterogeneous graph.

[0058] In the embodiment of the present disclosure, the feature types that may have an associated relationship may be preset, for example, interactive link features, comment content features, and object information features. Specifically, the above step a2, based on the interactive link features, comment content features, and object information features, determines the target comment that hits the preset associated feature in the comments to be analyzed, including:

[0059] Based on the comment content features, target comments whose content and / or abnormal label similarity meet the first association condition are determined among the comments to be analyzed, and based on the interaction link features, target comments with the same comment source and / or the number of similar comments reaching a threshold are determined among the comments to be analyzed, and based on the object information features, target comments whose object display information similarity meets the second association condition are determined among the comments to be analyzed.

[0060] In the disclosed embodiment, the target comment whose content similarity satisfies the first association condition can be used to indicate that the comment content similarity of the object to be classified under the same interactive content or the same type of interactive content is high. For example, a certain object to be classified has posted multiple comment texts with similar content texts in the same video. The abnormal label similarity can be used to indicate that the comment content of the object to be classified under the same interactive content or the same type of interactive content is marked with an abnormal label, and the similarity of the abnormal label is high. For example, a certain object to be classified has posted multiple abnormal labels marked with malicious attacks on others in the same video.

[0061] In addition, the target comment with the same comment source can be used to indicate the starting operation in the operation link of the object to be classified entering the interactive content and making comments. For example, the object to be classified jumps to the interactive content through the jumpable identifier of the interactive content X, and publishes the comment to be analyzed in the interactive content as an operation link, and the comment source of the comment to be analyzed is the interactive content X. When detecting whether the number of similar comments corresponding to the comment to be analyzed reaches a threshold, the object that posted the comment to be analyzed can be determined, and similar comments corresponding to the comment to be analyzed posted by the object can be determined among all the comments to be analyzed, and it is determined whether the number of similar comments reaches the threshold.

[0062] In addition, the object display information may include the public object display information of the object to be analyzed, such as account registration time, account nickname and other object information. On this basis, the comments whose object display information similarity in the comments to be analyzed meets the second association condition can be determined as target comments, such as the account registration time is similar and the account nickname format is the same.

[0063] In the embodiment of the present disclosure, the comments to be analyzed containing multiple comment scenes can be screened and associated based on the preset association features, so as to construct the following Figure 2 The heterogeneous graph shown in provides a technical basis for the coarse-recall model to capture links based on the heterogeneous graph through local similarity to obtain a homogeneous graph in each comment scenario.

[0064] In some optional implementations, the above step S1022, classifying the heterogeneous graphs according to the preset comment scenario based on the first graph classification model to obtain a plurality of homogeneous graphs, and determining the object collection according to the homogeneous graphs, includes:

[0065] Based on the first graph classification model, the heterogeneous graphs are classified according to the comment scenarios to obtain homogeneous graphs corresponding to the various comment scenarios, wherein the comment scenarios include at least one of the following: the same interactive content comment scenarios, the same type of interactive content comment scenarios, and the object self-created content comment scenarios.

[0066] In the embodiments of the present disclosure, Figure 2 As shown in , the nodes in the heterogeneous graph can be classified for each review scenario to obtain a homogeneous graph under the review scenario. Specifically, the nodes whose corresponding associated features match the current review scenario can be screened out based on the relationship edges of the heterogeneous graph, and the homogeneous graph corresponding to the current review scenario can be generated based on the nodes and the associated features.

[0067] Here, the above-mentioned comment scenarios may include: the same interactive content comment scenario, the same type of interactive content comment scenario, and the object self-created content comment scenario, wherein the same interactive content comment scenario can be used to indicate the scenario of commenting on a single video on a social platform, the same type of interactive content comment scenario can be used to indicate the scenario of commenting on videos with the same type label, for example, the comment scenario for videos with electronic product type labels, and the object self-created content comment scenario can be used to indicate the scenario of the object to be classified commenting on the self-created and published video.

[0068] For example, when determining the homogeneous graph corresponding to the same interactive content comment scene, a homogeneous graph can be constructed for objects to be classified that satisfy the same video and similar comment text content, the same video and the same comment text sentiment tendency, the same video and the comment text hit the same abnormal label, the similar comment interaction link, and the number of comments under similar videos exceeds the threshold. In addition, when determining the same type of interactive content comment scene, a homogeneous graph can be constructed for objects to be classified that satisfy the same type of video and similar comment text content, the same type of video and the same comment text sentiment tendency, the same type of video and the same comment text hit the abnormal label, the similar comment interaction link, and the number of comments under similar videos exceeds the threshold. In addition, when determining the homogeneous graph corresponding to the object self-created content comment scene, a homogeneous graph can be constructed for objects to be classified that satisfy the similarity of the submission title or label, the posting of comments under the self-submitted video, and the similarity of the comment text between multiple objects to be classified.

[0069] In the disclosed embodiment, a homogeneous graph may be constructed for different comment scenarios, so that abnormal groups can be discovered based on the homogeneous graph to obtain a collection of objects, thereby improving the efficiency of discovering abnormal objects and better maintaining the environment of the social platform.

[0070] In some optional implementations, the above step S103, performing abnormal feature analysis based on each object to be classified in the object collection to obtain an abnormal object, includes:

[0071] Step S1031, determining input features according to content display features, interactive link features, and object information features corresponding to the objects to be classified in the object collection, wherein the content display features are used to indicate content displayed to the objects to be classified within a preset period of time.

[0072] Step S1032: input the input features and the object collection into the second graph classification model to obtain abnormal objects.

[0073] In an embodiment of the present disclosure, the second graph classification model may be a refined calling model. For example, the refined calling model may be a shallow multi-head graph attention neural network GAT (Graph Attention Networks).

[0074] Here, the objects to be classified in each object collection can be classified according to the precision recall model to obtain target objects with abnormal characteristics, thereby filtering out the misrecalled objects in each object collection. For example, if the object collection includes abnormal objects that publish a large amount of similar traffic-generating content, and also includes non-abnormal objects that interact normally with other users by copying traffic-generating content, the abnormal objects can be filtered out by the precision recall model.

[0075] Specifically, Figure 3 The figure shows a schematic diagram of determining a target collection, wherein the input features including content display features, interactive link features, and object information features can be input into the precision calling model to obtain a target collection including the target object. Here, the target object can be an object in which at least some of the input features are abnormal. For example, the abnormality indicated by the object display feature can be that the videos displayed to the object are highly homogenized and the abnormal labels of the videos are highly similar. The abnormality indicated by the interactive link feature can be that the number of executions corresponding to the diversion comment link and the advertising comment link is high. The abnormality indicated by the object information feature can be that the registration time of the object is short, the number of followers and the number of followers are small or even empty.

[0076] In some optional implementations, the above step S103 further includes:

[0077] Step S11, obtaining positive samples and negative samples corresponding to each input feature to obtain sample data.

[0078] Step S12: adjusting the parameters of the second image classification model based on the sample data.

[0079] In the disclosed embodiment, the refined model can be fine-tuned through the non-graph network model after the refined model is launched to improve the recognition accuracy of the refined model. Here, considering that the sample data is often limited and cannot reflect the real data, and the number of layers of the refined model is shallow, it is easy to produce catastrophic forgetting, therefore, the GATencoder of the refined model can be frozen, and only the MLP classifier can be trained to optimize the overall accuracy, precision and recall of the refined model.

[0080] Specifically, first, the positive samples and negative samples corresponding to the input features can be obtained to determine the sample data, wherein the positive samples can be abnormal users and the negative samples can be non-abnormal users. It should be understood that in the early stage of model launch, there may be zero samples. Therefore, the isolation forest algorithm can be used to select a part of black samples (i.e., sample data that may be negative samples) according to the output results of the rough call model to generate sample data based on the black samples.

[0081] Here, a schematic diagram of identifying abnormal objects corresponding to interactive link features is shown, Figure 4 The figure shows a flow chart of identifying the target object through the second graph classification model, wherein the historical interactive link features of the objects to be classified in each object collection can be extracted first, for example, the number of soft advertisement comments, high-risk diversion comments, other diversion comments, and advertising risk video comments of the objects to be classified in the past 30 days can be targeted. Next, outlier detection can be performed through the isolation forest algorithm to obtain a high-risk object collection, and the collection objects can be submitted for review to determine the target objects in each object collection.

[0082] In addition, in the embodiment of the present disclosure, the second graph classification model can adopt the Encoder-Decoder structure of GraphMAE, the refined model can be pre-trained in advance, and the refined model that meets the accuracy conditions can be put online.

[0083] Specifically, Figure 5 Figure 1 is a flow chart of training the second graph classification model, in which randomly masking (Mask) object node features and learning is a common training strategy. Specifically, GraphEncoder is a two-layer multi-head graph attention network (GAT), which encodes the graph with some node features randomly masked off so that each node can be integrated into the local information of the neighborhood. Then, the masked nodes are masked again (Re-Mask) to increase the learning difficulty and improve the robustness of the model. Finally, a single-layer GAT is used as a decoder to reconstruct the masked node features, and the error with the original features is calculated as the loss for learning. Specifically, when performing loss learning, the loss function can be used to scale the cosine error and calculate the reconstructed feature zi With the original feature x i The error, the loss function can be expressed as:

[0084] Here, the training data for pre-training may include an unlabeled object collection, content display features, interactive link features, and object information features of each object in the object collection.

[0085] It should be understood that after the pre-training of the second graph classification model is completed, the second graph classification model can be fine-tuned. Specifically, the fine-tuning data can be a partially labeled coarse group, and the fine-tuning strategy can be to freeze the GATencoder of the fine-recall model and only train the MLP classifier to optimize the overall accuracy, precision and recall of the fine-recall model, wherein the GATencoder can be used to generate object node representations, and the MLP classifier can be used to predict the probability that the object belongs to the target collection.

[0086] In the disclosed embodiment, the refined calling model can be fine-tuned through a non-graph network model after the refined calling model is launched to improve the recognition accuracy of the refined calling model, thereby achieving real-time optimization of the refined calling model and enhancing the adaptability of the refined calling model.

[0087] In some optional embodiments, the above Figure 1 The corresponding embodiments also include:

[0088] After obtaining the target collection, an abnormal identification is determined based on the abnormal features of the target objects in the target collection, and the target collection is marked according to the abnormal identification.

[0089] In the disclosed embodiment, the abnormal label corresponding to each abnormal object in the target collection can be obtained. If it is not obtained, an abnormal identification can be generated according to the abnormal features corresponding to each abnormal object, and the abnormal label of the target collection can be determined according to the abnormal identification, so as to identify the abnormal objects matching the target collection according to the abnormal label, and cluster the abnormal objects. For example, the abnormal feature corresponding to the target object in the object collection can be the posting of multiple drainage comments with similar content, and the abnormal identification can be determined as the drainage group.

[0090] In summary, in the disclosed embodiment, the comments to be analyzed posted by the objects to be classified can be first obtained, and the comments to be analyzed can be analyzed based on the preset correlation features, so as to classify the objects to be classified according to the analysis results to obtain an object collection, wherein the preset correlation features are used to indicate the abnormal correlations between the comments to be analyzed. Then, abnormal feature analysis can be performed based on each object to be classified in the object collection to obtain abnormal objects, and the object collection can be adjusted according to the abnormal objects to obtain a target collection, thereby discovering and dividing the object groups of abnormal comments, improving the confidence of the recognition results, and at the same time improving the recognition efficiency for abnormal objects.

[0091] In this embodiment, an object classification device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0092] This embodiment provides an object classification device, such as Figure 6 As shown, including:

[0093] An acquisition module 601 is used to acquire comments to be analyzed posted by the object to be classified;

[0094] A classification module 602 is used to analyze the comments to be analyzed based on preset correlation features, and classify the objects to be classified according to the analysis results to obtain an object collection, wherein the preset correlation features are used to indicate abnormal correlations between the comments to be analyzed;

[0095] The analysis module 603 is used to perform abnormal feature analysis based on each object to be classified in the object collection to obtain abnormal objects, and adjust the object collection according to the abnormal objects to obtain a target collection.

[0096] In some optional implementations, the classification module 602 is further configured to:

[0097] Analyze the comments to be analyzed based on the preset correlation features to obtain the comments to be analyzed that hit the preset correlation features, and construct a heterogeneous graph based on the objects to be classified corresponding to the comments to be analyzed, wherein the nodes in the heterogeneous graph are used to indicate the objects to be classified, and the edges between the nodes are used to indicate the preset correlation features between the nodes;

[0098] Based on the first graph classification model, the heterogeneous graphs are classified according to the preset comment scenario to obtain multiple homogeneous graphs, and an object collection is determined based on the homogeneous graphs, wherein the nodes in the homogeneous graphs are used to indicate the corresponding objects to be classified in the current comment scenario, and the edges between the nodes are used to indicate the preset association features between the nodes.

[0099] In some optional implementations, the classification module 602 is further configured to:

[0100] Based on the comments to be analyzed, determining the interactive link characteristics of the object to be classified, wherein the interactive link characteristics are used to indicate the operation link of the object to be classified to publish the comments to be analyzed;

[0101] Obtaining the comment content features of the comments to be analyzed and the object information features corresponding to the objects to be classified, and determining the target comments that hit the preset correlation features in the comments to be analyzed based on the interactive link features, the comment content features and the object information features;

[0102] Based on the preset association features corresponding to the target comments, the objects to be classified are associated to obtain a heterogeneous graph.

[0103] In some optional implementations, the classification module 602 is further configured to:

[0104] Based on the comment content features, target comments whose content and / or abnormal label similarity meet the first association condition are determined among the comments to be analyzed, and based on the interaction link features, target comments with the same comment source and / or the number of similar comments reaching a threshold are determined among the comments to be analyzed, and based on the object information features, target comments whose object display information similarity meets the second association condition are determined among the comments to be analyzed.

[0105] In some optional implementations, the classification module 602 is further configured to:

[0106] Based on the first graph classification model, the heterogeneous graphs are classified according to the comment scenarios to obtain homogeneous graphs corresponding to the various comment scenarios, wherein the comment scenarios include at least one of the following: the same interactive content comment scenarios, the same type of interactive content comment scenarios, and the object self-created content comment scenarios.

[0107] In some optional implementations, the analysis module 603 is further configured to:

[0108] Determine the input feature according to the content display feature, the interactive link feature and the object information feature corresponding to the object to be classified in the object collection, wherein the content display feature is used to indicate the content displayed to the object to be classified within a preset period of time;

[0109] The input features and the object collection are input into the second graph classification model to obtain abnormal objects.

[0110] In some optional implementations, the analysis module 603 is further configured to:

[0111] Obtain positive samples and negative samples corresponding to each input feature to obtain sample data;

[0112] The parameters of the second image classification model are adjusted based on the sample data.

[0113] In some optional embodiments, the device is also used for:

[0114] After obtaining the target collection, an abnormal identification is determined based on the abnormal features of the target objects in the target collection, and the target collection is marked according to the abnormal identification.

[0115] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0116] The object classification device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0117] The present disclosure also provides a computer device having the above Figure 6 The object classification device shown.

[0118] See also Figure 7 , Figure 7 is a schematic diagram of a computer device provided by an optional embodiment of the present disclosure, such as Figure 7 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 7 A processor 10 is taken as an example.

[0119] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0120] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0121] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0122] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0123] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0124] The embodiments of the present disclosure also provide a computer-readable storage medium. The above-mentioned method according to the embodiments of the present disclosure can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium and downloaded through a network, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0125] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0126] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0127] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0128] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0129] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0130] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for object classification, characterized in that: The method comprises: Get the comments to be analyzed posted by the object to be classified; Analyzing the comments to be analyzed based on preset correlation features, and classifying the objects to be classified according to the analysis results to obtain an object collection, wherein the preset correlation features are used to indicate abnormal correlations between the comments to be analyzed; An abnormal feature analysis is performed on each object to be classified in the object collection to obtain an abnormal object, and the object collection is adjusted according to the abnormal object to obtain a target collection.

2. The method according to claim 1, characterized in that The analyzing the comments to be analyzed based on the preset correlation features, and classifying the objects to be classified according to the analysis results to obtain an object collection, including: Analyze the comments to be analyzed based on preset correlation features to obtain comments to be analyzed that hit the preset correlation features, and construct a heterogeneous graph based on the objects to be classified corresponding to the comments to be analyzed, wherein the nodes in the heterogeneous graph are used to indicate the objects to be classified, and the edges between the nodes are used to indicate the preset correlation features between the nodes; Based on the first graph classification model, the heterogeneous graph is classified according to a preset comment scenario to obtain multiple homogeneous graphs, and the object collection is determined based on the homogeneous graphs, wherein the nodes in the homogeneous graph are used to indicate the corresponding objects to be classified in the current comment scenario, and the edges between the nodes are used to indicate the preset association features between the nodes.

3. The method according to claim 2, characterized in that The step of analyzing the comments to be analyzed based on the preset correlation features to obtain the comments to be analyzed that hit the preset correlation features, and constructing the heterogeneous graph based on the objects to be classified corresponding to the comments to be analyzed, includes: Based on the comments to be analyzed, determining the interactive link characteristics of the object to be classified, wherein the interactive link characteristics are used to indicate the operation link of the object to be classified to publish the comments to be analyzed; Obtaining the comment content features of the comments to be analyzed and the object information features corresponding to the objects to be classified, and determining the target comments that hit the preset association features in the comments to be analyzed based on the interactive link features, comment content features and object information features; Based on the preset association features corresponding to the target comments, the objects to be classified are associated to obtain the heterogeneous graph.

4. The method according to claim 3, characterized in that: The determining of the target comment that hits the preset association feature based on the interactive link feature, the display content feature, and the object information feature includes: Based on the comment content features, target comments whose content and / or abnormal label similarity meet the first association condition are determined in the comments to be analyzed, and based on the interactive link features, target comments with the same comment source and / or the number of similar comments reaching a threshold are determined in the comments to be analyzed, and based on the object information features, target comments whose object display information similarity meets the second association condition are determined in the comments to be analyzed.

5. The method according to claim 2, characterized in that: The step of classifying the heterogeneous graphs according to a preset comment scenario based on the first graph classification model to obtain a plurality of homogeneous graphs, and determining the object collection according to the homogeneous graphs includes: Based on the first graph classification model, the heterogeneous graph is classified according to the comment scenarios to obtain a homogeneous graph corresponding to each comment scenario, wherein the comment scenario includes at least one of the following: the same interactive content comment scenario, the same type of interactive content comment scenario, and the object self-created content comment scenario.

6. The method according to claim 1, characterized in that The abnormal feature analysis is performed on each object to be classified in the object collection to obtain an abnormal object, including: Determine the input feature according to the content display feature, the interactive link feature and the object information feature corresponding to the object to be classified in the object collection, wherein the content display feature is used to indicate the content displayed to the object to be classified within a preset time period; The input features and the object collection are input into a second graph classification model to obtain the abnormal object.

7. The method according to claim 6, characterized in that The method further comprises: Obtaining positive samples and negative samples corresponding to each of the input features to obtain sample data; The parameters of the second image classification model are adjusted based on the sample data.

8. The method according to claim 1, characterized in that: The method further comprises: After obtaining the target collection, an abnormality identifier is determined based on abnormal features of the target objects in the target collection, and the target collection is marked according to the abnormality identifier.

9. An object classification device, characterized in that: The device comprises: An acquisition module is used to obtain the comments to be analyzed posted by the object to be classified; A classification module, used to analyze the comments to be analyzed based on preset correlation features, and classify the objects to be classified according to the analysis results to obtain an object collection, wherein the preset correlation features are used to indicate abnormal correlations between the comments to be analyzed; The analysis module is used to perform abnormal feature analysis based on each object to be classified in the object collection to obtain abnormal objects, and adjust the object collection according to the abnormal objects to obtain a target collection.

10. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the object classification method according to any one of claims 1 to 8 by executing the computer instructions.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the object classification method according to any one of claims 1 to 8.

12. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the object classification method according to any one of claims 1 to 8.

Images