Complaint gang identification method and device, equipment and storage medium

Abstract

The invention discloses a complaint gang identification method, apparatus and device, and a storage medium. The method comprises the steps of obtaining to-be-identified user data; whether the to-be-recognized user data belongs to sample data of a target clustering model in the training process is judged, and the target clustering model is obtained through semi-supervised training based on sample data corresponding to all nodes in a knowledge graph and a collected unmarked sample data set; the knowledge graph is constructed based on users having association relationships and associated sample data thereof; if yes, taking a risk identification result associated with the sample data to which the to-be-identified user data belongs as a gang risk identification result corresponding to the to-be-identified user data; and if not, performing clustering identification on the to-be-identified user data based on a pre-constructed knowledge graph to obtain a gang risk identification result. According to the method, the clustering model is obtained by using the knowledge graph and semi-supervised training, and the accuracy and efficiency of gang identification prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a method, device, equipment and storage medium for identifying a complaint group. Background Art

[0002] At present, the way to identify gang complaints is to characterize the gang complaint community through a series of operation steps and other machine learning algorithms. The specific operation process is: data preprocessing, feature engineering, model construction, model training, model tuning and model evaluation. Machine learning algorithms require a large amount of labeled data for training. However, the correct labeling of data on gang complaint behavior requires the participation of professionals and institutions, which usually involves complex review and verification processes and is time-consuming. In addition, gang complaint behavior is highly dynamic. As time goes by and strategies change, new types of complaint behaviors continue to emerge. When encountering new data distributions or behavior patterns, traditional machine learning models often need to be retrained or a large amount of model tuning, which is not only time-consuming but also inefficient. Summary of the invention

[0003] Based on this, it is necessary to provide a complaint group identification method, device, equipment and storage medium for the above-mentioned technical problems, so as to solve at least one of the above-mentioned technical problems.

[0004] The present invention provides a method for identifying a complaint group, comprising:

[0005] Obtain the user data to be identified;

[0006] Determine whether the user data to be identified belongs to the sample data of the target clustering model in the training process, wherein the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, and the knowledge graph is constructed based on users with associated relationships and their associated sample data;

[0007] If yes, the risk identification result associated with the sample data to which the to-be-identified user data belongs is used as the group risk identification result corresponding to the to-be-identified user data, wherein the risk identification result is determined based on the result obtained by clustering each of the sample data based on the target clustering model;

[0008] If not, cluster identification is performed on the user data to be identified based on a pre-built knowledge graph to obtain a group risk identification result corresponding to the user data to be identified.

[0009] According to a complaint group identification method provided by the present invention, the knowledge graph is constructed based on the following steps:

[0010] Get multiple sample data;

[0011] Based on each of the sample data, determining the association relationship between each complaining user and other users, wherein the complaining user refers to a user who has made a complaint;

[0012] Generate triples based on users with association relationships and their associated sample data;

[0013] Based on each of the triples, the knowledge graph is constructed.

[0014] According to a complaint group identification method provided by the present invention, the target clustering model is trained based on the following steps:

[0015] Performing level division on each node in the knowledge graph to obtain a division result of each node;

[0016] Performing cluster analysis on each node in the knowledge graph to obtain a target clustering result, wherein the target clustering result is used to characterize the community to which each node belongs;

[0017] Based on the sample data corresponding to each of the nodes, the community to which they belong, and the division results, training the target clustering model;

[0018] Obtain an unlabeled sample data set, and input each unlabeled sample data in the unlabeled sample data set into the target clustering model to obtain a prediction result of each unlabeled sample data, wherein the unlabeled sample data set refers to a sample data set that is not labeled;

[0019] Determine a sample label of the unlabeled sample data based on the prediction result of each of the unlabeled sample data;

[0020] Based on the unlabeled sample data and the sample labels, the target clustering model is optimized and trained.

[0021] According to a complaint group identification method provided by the present invention, after optimizing and training the target clustering model based on the unlabeled sample data and the sample labels, the method further includes:

[0022] Determine the model clustering results corresponding to each sample data output by the target clustering model during the model training process;

[0023] For each sample data: based on the model clustering result corresponding to the sample data, the sample data is cross-analyzed with the sample data of each node in the knowledge graph to obtain the risk identification result corresponding to the sample data.

[0024] According to a complaint group identification method provided by the present invention, the level division of each node in the knowledge graph to obtain the division result of each node includes:

[0025] According to each of the sample data, the nodes in the knowledge graph are divided into complaint nodes and non-complaint nodes;

[0026] For any complaint node: if there is at least one edge between the complaint node and other complaint nodes, the complaint node is regarded as a primary user node; if there is no edge between the complaint node and other complaint nodes, the complaint node is regarded as a secondary user node;

[0027] The non-complaint nodes are divided into three levels of user nodes.

[0028] According to a complaint group identification method provided by the present invention, clustering analysis is performed on each node in the knowledge graph to obtain a target clustering result, including:

[0029] Performing cluster analysis on each node in the knowledge graph to obtain the community to which each node belongs;

[0030] Generate a fully connected graph based on each node of the same community;

[0031] Determine the weight between any two nodes in the fully connected graph;

[0032] Based on the weights between all nodes, a probability transfer matrix is ​​constructed;

[0033] Based on the probability transition moment, determining a community label corresponding to each node;

[0034] For any community: based on each weight in the fully connected graph and the community label corresponding to each node, each node in the community is screened and filtered to obtain the target clustering result.

[0035] According to a complaint group identification method provided by the present invention, clustering and identifying the user data to be identified based on a pre-built knowledge graph to obtain a group risk identification result corresponding to the user data to be identified includes:

[0036] Based on the user data to be identified, updating the knowledge graph;

[0037] Cluster and grade each node in the updated knowledge graph;

[0038] According to the clustering results and the division results, the user data to be identified and the sample data of each node in the updated knowledge graph are cross-analyzed to obtain the gang risk identification result.

[0039] The present invention also provides a complaint group identification device, comprising:

[0040] An acquisition module, used to acquire user data to be identified;

[0041] A judgment module, used to judge whether the user data to be identified belongs to the sample data of the target clustering model in the training process, wherein the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, and the knowledge graph is constructed based on users with associated relationships and their associated sample data;

[0042] a determination module, for, if yes, using the risk identification result associated with the sample data to which the to-be-identified user data belongs as the group risk identification result corresponding to the to-be-identified user data, wherein the risk identification result is determined based on the result obtained by clustering each of the sample data based on the target clustering model;

[0043] The identification module is used to cluster and identify the user data to be identified based on a pre-built knowledge graph, if not, to obtain the group risk identification result corresponding to the user data to be identified.

[0044] The present invention also provides a computer device, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the above-mentioned complaint group identification method when executing the computer-readable instructions.

[0045] The present invention also provides one or more readable storage media storing computer-readable instructions, which implement the above-mentioned complaint group identification method when executed by a processor.

[0046] The above-mentioned complaint group identification method, device, equipment and storage medium include: obtaining user data to be identified; judging whether the user data to be identified belongs to the sample data of the target clustering model in the training process, wherein the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, and the knowledge graph is constructed based on users with associated relationships and their associated sample data; if so, the risk identification result associated with the sample data to which the user data to be identified belongs is used as the group risk identification result corresponding to the user data to be identified, wherein the risk identification result is determined based on the result of clustering each sample data by the target clustering model; if not, clustering and identifying the user data to be identified based on the pre-constructed knowledge graph to obtain the group risk identification result corresponding to the user data to be identified. The present invention integrates multidimensional sample data sources to construct a knowledge graph, and can intuitively determine the association between different nodes and different variables based on the knowledge graph, thereby improving the accuracy and reliability of group identification prediction. In addition, by using limited data labels and a large amount of unlabeled data to optimize the training of clustering models, there is no need to manually label large amounts of data, thereby improving the efficiency and accuracy of gang identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0048] Figure 1 It is a flow chart of a method for identifying a complaint group in one embodiment of the present invention;

[0049] Figure 2 is a schematic diagram of a knowledge graph provided by an embodiment of the present invention;

[0050] Figure 3 is a schematic diagram of performing cluster analysis on each sample data provided by an embodiment of the present invention;

[0051] Figure 4 is a schematic diagram of the structure of a complaint group identification device in one embodiment of the present invention;

[0052] Figure 5 is a schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The terms used in one or more embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of the present invention. The singular forms of "a", "said" and "the" used in one or more embodiments of the present invention are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of the present invention refers to and includes any or all possible combinations of one or more associated listed items.

[0055] In one embodiment, specifically, Figure 1 As shown, Figure 1 : is a flow chart of a method for identifying a complaint group in an embodiment of the present invention. The present invention provides a method for identifying a complaint group, comprising the following steps:

[0056] Step S11, obtaining user data to be identified;

[0057] It should be noted that the user data to be identified includes the contact numbers of friends in the target user’s voice communication circle, the number of friend users, contact numbers, online time, age, number of user overdue payments, account status, number of complaint calls made, and use of complaint-related applications.

[0058] Step S12, determining whether the user data to be identified belongs to sample data of the target clustering model during training;

[0059] It should be noted that the target clustering model is trained using a part of labeled data (supervised learning) and a large amount of unlabeled data (unsupervised learning). In this embodiment, the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, thereby increasing the generalization ability of the model. The specific training process of the target clustering model is specifically described in the following embodiments and will not be repeated here.

[0060] It should be noted that the knowledge graph is constructed based on users with associated relationships and their associated sample data. The construction process of the knowledge graph is specifically described in the following embodiments and will not be repeated here.

[0061] It should be noted that the sample data includes the contact numbers of friends in the user's voice communication circle, the number of friend users, contact numbers, online time, age, account status, number of complaint calls, use of complaint-related applications, time when the complaint occurred, the contact number of the other end of the call, the number of outgoing calls, and the use of complaint-related apps.

[0062] Step S13: If yes, the risk identification result associated with the sample data to which the user data to be identified belongs is used as the group risk identification result corresponding to the user data to be identified;

[0063] Specifically, it is determined whether the user data to be identified belongs to the sample data of the target clustering model during the training process. If it belongs to the sample data of the target clustering model during the training process, since the sample data has been fully calculated in the model training, the risk identification result associated with the sample data to which the user data to be identified belongs is directly used as the group risk identification result corresponding to the user data to be identified, without further processing.

[0064] In one embodiment, the process of determining the risk identification result is as follows:

[0065] Determine the model clustering results corresponding to each sample data output by the target clustering model during the model training process; for each sample data: based on the model clustering results corresponding to the sample data, cross-analyze the sample data with the sample data of each node in the knowledge graph to obtain the risk identification result corresponding to the sample data.

[0066] Specifically, during the model training process, the target clustering model identifies and obtains the model clustering results of each sample data participating in the model training. For each sample data: the sample data and other sample data of each node in the knowledge graph are cross-analyzed to analyze the attribute similarity (for example, cosine similarity) between the sample data and the first-level user node, the second-level user node, and the third-level user node, as well as whether there is a call communication behavior, thereby determining the division level corresponding to the user data to be identified based on the attribute similarity and whether there is a call communication behavior, and then using the division level as the risk identification result corresponding to the sample data. In other embodiments, the risk identification result corresponding to the sample data is determined in combination with the division level and the clustering result. The process of determining the division level is specifically described in the following embodiments and will not be repeated here.

[0067] Step S14: if not, cluster identification is performed on the user data to be identified based on a pre-built knowledge graph to obtain a group risk identification result corresponding to the user data to be identified.

[0068] Specifically, if it is not sample data of the target clustering model during training, it proves that the user data to be identified is new user data. The user data to be identified is added to the knowledge graph to judge the user behavior and attributes of the user data to be identified, for example, the frequency of its complaint behavior, the characteristics of the complaint object, the call mode, the network activity, etc. are analyzed, so as to infer the possible group complaints of the data and output the corresponding group risk identification results.

[0069] In one embodiment, cluster identification is performed on the user data to be identified based on a pre-built knowledge graph to obtain a group risk identification result corresponding to the user data to be identified, including:

[0070] Based on the user data to be identified, the knowledge graph is updated; each node in the updated knowledge graph is clustered and graded; according to the clustering results and the division results, the user data to be identified and the sample data of each node in the updated knowledge graph are cross-analyzed to obtain the gang risk identification result.

[0071] Specifically, based on the user data to be identified, new entities and attribute relationships are extracted, and based on the new entities and attribute relationships, the original knowledge graph is updated, and then each node in the updated knowledge graph is clustered to obtain a clustering result. According to the clustering result, the complaint group to which the user data to be identified may belong can be determined. Optionally, the user data to be identified may belong to a known group community or may not belong to a known group community. In one embodiment, in order to more accurately identify the group risk, the user data to be identified and the sample data of each node in the updated knowledge graph are cross-analyzed to analyze the attribute similarity (for example, cosine similarity) between the user data to be identified and the first-level user node, the second-level user node, and the third-level user node, and whether there is a call communication behavior, so as to determine the division level corresponding to the user data to be identified according to the attribute similarity and whether there is a call communication behavior, and then use the division level as the group risk identification result. In other embodiments, the group risk identification result is determined in combination with the division level and the clustering result. For example, the specific process of the division level is shown in Table 1:

[0072]

[0073] Among them, the highest grayscale setting for group complaint samples is level 7, and the first-level user node, that is, the classification level of the group user is level 7. If the user data to be identified has a social circle behavior with the first-level user node and is similar to the first-level user node attributes, the classification level of the user data to be identified is determined to be level 6; if the user data to be identified has a social circle behavior with the first-level user node and is similar to the second-level user node attributes, or is similar to the first-level user node attributes and has a social circle call behavior with the second-level user node, the classification level of the user data to be identified is determined to be level 5; if the user data to be identified has a social circle behavior with the third-level user node and is similar to the first-level user node attributes, or has a social circle behavior with the second-level user node and is similar to the second-level user node attributes, or has a social circle behavior with the first-level user node If the user data to be identified has a social circle behavior with the third-level user node and is similar to the attributes of the second-level user node, the division level of the user data to be identified is determined to be level 4; if the user data to be identified has a social circle behavior with the third-level user node and is similar to the attributes of the second-level user node, or has a social circle behavior with the second-level user node and is similar to the attributes of the third-level user node, the division level of the user data to be identified is determined to be level 3; if the user data to be identified has a social circle behavior with the third-level user node and is similar to the attributes of the third-level user node, the division level of the user data to be identified is determined to be level 2; if the user data to be identified has no social circle behavior with all sample data and has different attributes, the division level of the user data to be identified is determined to be level 1.

[0074] The embodiment of the present invention, through the above scheme, includes: obtaining the user data to be identified; judging whether the user data to be identified belongs to the sample data of the target clustering model in the training process, wherein the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, and the knowledge graph is constructed based on users with associated relationships and their associated sample data; if so, the risk identification result associated with the sample data to which the user data to be identified belongs is used as the group risk identification result corresponding to the user data to be identified, wherein the risk identification result is determined based on the result obtained by clustering each sample data by the target clustering model; if not, clustering and identifying the user data to be identified based on the pre-constructed knowledge graph to obtain the group risk identification result corresponding to the user data to be identified. The embodiment of the present invention integrates multi-dimensional sample data sources to construct a knowledge graph, and can intuitively determine the association between different nodes and different variables based on the knowledge graph, thereby improving the accuracy and reliability of group identification prediction. In addition, by using limited data labels and a large amount of unlabeled data to optimize the training of clustering models, there is no need to manually label large amounts of data, thereby improving the efficiency and accuracy of gang identification.

[0075] In one embodiment of the present invention, the knowledge graph is constructed based on the following steps:

[0076] Step S21, obtaining a plurality of sample data;

[0077] Step S22, based on each of the sample data, determining the association relationship between each complaining user and other users, wherein the complaining user refers to a user who has made a complaint;

[0078] Step S23, generating triples based on users with associated relationships and their associated sample data;

[0079] Step S24: construct the knowledge graph based on each of the triples.

[0080] Specifically, we obtain operator data, and based on the operator data, we collect a large amount of data related to complaint risks, including but not limited to: the number of complaint calls, the number of visits to complaint applications, the number of users in the voice communication circle, and other related data. We screen the data samples and account periods collected from multiple aspects where complaints have occurred, and we need to select the fields required to build the knowledge graph, such as: online time, age, number of user overdue payments, account status, number of complaint calls, and traffic used for complaint-related applications. Furthermore, we select samples with normal user account status based on account status. In order to ensure the accuracy of data quality, we remove outliers and duplicates from the screened samples. For example, we use mean interpolation to handle missing values ​​in the samples.

[0081] In addition, it should be noted that when constructing a knowledge graph, entity information needs to be extracted from a large amount of sample data. In the scenario of identifying group complaints, data related to the group and the complaint are mainly extracted, such as the contact number where the complaint occurred, the time when the complaint occurred, the number of times the complaint occurred, the number of times the complaint was made, the number of times the complaint was made, the contact number of the other end call, the number of calls made, and the use of complaint apps. In combination with the application scenario of this solution, the complaint group, that is, the connection between users who made complaints, is found through operator data. Therefore, the entity in this embodiment is each user who has made a complaint, and the attribute is the various data of user behavior. By using the calling call and the other end contact number, a user relationship with the calling call as the chain can be established, wherein the calling call refers to the contact number where the complaint occurred, and the other end contact number refers to the number that is connected to the contact number where the complaint occurred. Thus, based on each of the sample data, the association relationship between each complaining user and other users is determined. Furthermore, based on the users with associated relationships and their associated sample data, triples are generated. For example, let the user be user, user behavior has n dimensions, denoted as {B1, B2, …, Bn}, and the peer user be opp_user, then the triple is<user,B,opp_user> In a specific embodiment, the number of contacts between the complaining user and the peer user can be combined to construct a triplet as follows: <user, number of calls, peer user>.

[0082] Further, based on each of the triples, the knowledge graph is constructed, referring to Figure 2 , Figure 2 Schematic diagram of a knowledge graph provided by an embodiment of the present invention. In one embodiment, in order to improve the accuracy and completeness of the knowledge graph, the TransD algorithm is used to map the entities and relationships of each triple to the same dimension. The TransD algorithm is an existing mature algorithm and will not be described in detail here.

[0083] The embodiment of the present invention uses the above scheme to extract entities that need to be identified from sample data and map them to nodes of the knowledge graph, so that the potential correlation behaviors between gangs can be more effectively explored in combination with the knowledge graph, thereby improving the accuracy of gang identification.

[0084] In one embodiment of the present invention, the target clustering model is trained based on the following steps:

[0085] Step S31, classifying each node in the knowledge graph to obtain a classification result for each node;

[0086] Specifically, after the knowledge graph is established, the nodes in the knowledge graph are divided into complaint nodes and non-complaint nodes based on the complaint behavior and the related behavior of voice calls. Furthermore, for any complaint node: if there is at least one edge between the complaint node and other complaint nodes, the complaint node is regarded as a first-level user node; if there is no edge between the complaint node and other complaint nodes, the complaint node is regarded as a second-level user node; the non-complaint node is divided into a third-level user node.

[0087] Step S32, performing cluster analysis on each node in the knowledge graph to obtain a target clustering result, wherein the target clustering result is used to characterize the community to which each node belongs;

[0088] In one embodiment, the local density of the sample node i is calculated, for example, the local density of the sample node i is calculated using a truncated kernel. If there is a sample node i, then for any sample node j where i≠j, the Euclidean distance between two sample nodes is calculated, and the neighborhood truncation distance of the sample node i is calculated, and then the local density of the sample node i is calculated based on the Euclidean distance and the neighborhood truncation distance, and the calculation formula is as follows:

[0089]

[0090]

[0091] Among them, d ij is the Euclidean distance between sample nodes i and j, d c is the neighborhood cutoff distance of sample node i, ρ i is the local density of sample node i. In addition, before calculating each sample node i, it is necessary to sort the local density of each data point and calculate the minimum distance between each sample node i and other points with higher density.

[0092] For the sample node with the highest local density, its relative distance is defined as:

[0093] δi=max(i≠j)(dij)

[0094] For other sample nodes, their relative distance is defined as:

[0095] δi=min(i≠j)(dij)

[0096] The sample node with the highest local density is set as the cluster center, and its relative distance is the maximum value. For other sample nodes, two indicators are used to determine the density peak: (1) the local density ρ is high (2) the relative distance δ is large. The density peak is found by using the decision value γ, which is defined as:

[0097] γi=ρi*δi

[0098] After finding the remaining density peaks using the decision value, for each sample node i, it is assigned to the closest density peak, and each sample node is clustered to obtain the clustering result, so as to determine the community to which each node belongs according to the clustering result.

[0099] In another embodiment, the nodes of any community in the clustering results obtained in the above embodiment may not all be the same gang or gang members. In order to improve the accuracy of clustering, in this embodiment, the above clustering results are filtered. More specifically: based on the nodes of the same community, a fully connected graph is generated; the weight between any two nodes in the fully connected graph is determined. Optionally, there are nodes i and j, and σ is the weight w ij The parameter d ij represents the Euclidean distance between node i and node j, then the edge weight calculation formula between node i and node j is:

[0100]

[0101] Furthermore, based on the weights between all nodes, a probability transfer matrix is ​​constructed;

[0102]

[0103] It should be noted that the larger the weight of the edge, the more similar the two nodes are, and the easier it is for the label to propagate. Based on the probability transfer moment, the label transfer algorithm is used to determine the community label corresponding to each node; further, for any community: based on the weights in the fully connected graph and the community label corresponding to each node, the nodes in the community are screened and filtered, for example, points with weights lower than the preset weight threshold and points whose community labels do not belong to the same community are screened to obtain the final target clustering result. By screening edges with too low weights and nodes with different labels, the noise belonging to different communities or no communities in the community can be effectively removed, thereby improving the accuracy of gang prediction.

[0104] Step S33, training the target clustering model based on the sample data corresponding to each of the nodes, the community to which they belong, and the division results;

[0105] Specifically, the first-level, second-level, and third-level user nodes that have completed clustering are recorded as label data. For example, the community to which the node belongs and the division results (first-level user node, second-level user node, third-level user node) are used as labels, and then based on the sample data and labels corresponding to each of the nodes, the target clustering model is trained.

[0106] Step S34, obtaining an unlabeled sample data set, and inputting each unlabeled sample data in the unlabeled sample data set into the target clustering model to obtain a prediction result of each unlabeled sample data;

[0107] Step S35, determining a sample label of the unlabeled sample data based on the prediction result of each of the unlabeled sample data;

[0108] Step S36: Optimize and train the target clustering model based on the unlabeled sample data and the sample labels.

[0109] It should be noted that, since the number of labeled samples may be limited, in order to improve the performance of the model, an unlabeled sample data set can be used to optimize the model, wherein the unlabeled sample data set refers to a sample data set that is not labeled. Specifically, an unlabeled sample data set is obtained, and each unlabeled sample data in the unlabeled sample data set is input into the target clustering model to obtain a prediction result for each unlabeled sample data, and the prediction result includes the confidence corresponding to the first-level user node, second-level user node, and third-level user node of which community the unlabeled sample node belongs to. For any sample data, when the maximum confidence in the prediction result is greater than the set confidence threshold, it proves that the sample data belongs to the community corresponding to the current knowledge graph, and then the sample label of the sample data can be set directly based on the prediction result of the sample data, and the sample data and the sample label are added to the labeled sample set to enhance the samples used to train the model. If the maximum confidence in the prediction result is not greater than the set confidence threshold, it proves that the unlabeled sample data does not belong to the community corresponding to the current knowledge graph, and then all unlabeled sample data with a maximum confidence not greater than the set threshold are added to the knowledge graph to perform hierarchical division and clustering on each node of the new knowledge graph, wherein the hierarchical division and clustering processing methods are similar to the above-mentioned steps S31 and S32, which will not be repeated here. Then, based on the results of the hierarchical division and the results of the clustering, the sample labels of the unlabeled sample data are determined, and the target clustering model is optimized and trained based on the unlabeled sample data and the sample labels. In this way, the target clustering model can be used to perform cluster analysis on each sample data, referring to Figure 3 , Figure 3 It is a schematic diagram of performing cluster analysis on various sample data provided by an embodiment of the present invention.

[0110] The embodiment of the present invention uses the above scheme to train the model with a limited amount of labeled data, and then optimizes the model with a large amount of unlabeled data. As new samples continue to accumulate, the data set processed by the model will change, which may cause the model's clusters to evolve. In order to maintain the accuracy and adaptability of the model, the model parameters are adjusted by continuously inputting new samples, so that the model can continuously learn and adapt to new data distributions and sample characteristics, and can better capture the changing trends of complaint behaviors and gang relationships, thereby improving the model's prediction accuracy and response capabilities. And there is no need to manually label a large amount of data, saving time and cost.

[0111] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0112] In one embodiment, a complaint group identification device is provided, and the complaint group identification device corresponds one-to-one to the complaint group identification method in the above embodiment. Figure 4 As shown, Figure 4 : is a schematic diagram of a structure of a complaint group identification device in one embodiment of the present invention, the complaint group identification device comprises:

[0113] An acquisition module 41 is used to acquire user data to be identified;

[0114] A judgment module 42 is used to judge whether the user data to be identified belongs to the sample data of the target clustering model in the training process, wherein the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, and the knowledge graph is constructed based on users with associated relationships and their associated sample data;

[0115] A determination module 43, configured to, if yes, use the risk identification result associated with the sample data to which the to-be-identified user data belongs as the group risk identification result corresponding to the to-be-identified user data, wherein the risk identification result is determined based on the result obtained by clustering each of the sample data based on the target clustering model;

[0116] The identification module 44 is used to, if not, perform cluster identification on the user data to be identified based on a pre-built knowledge graph to obtain a group risk identification result corresponding to the user data to be identified.

[0117] For the specific limitations of the complaint group identification device, please refer to the limitations of the complaint group identification method above, which will not be repeated here. Each module in the above-mentioned complaint group identification device can be implemented in whole or in part through software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0118] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, Figure 5 : is a schematic diagram of a computer device in an embodiment of the present invention. The computer device includes a processor, a memory, a network interface and a database connected by a device bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a readable storage medium and an internal memory. The readable storage medium stores an operating device, a computer-readable instruction and a database. The internal memory provides an environment for the operation of the operating device and the computer-readable instructions in the readable storage medium. The database of the computer device is used to store data involved in the complaint group identification method. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer-readable instruction is executed by the processor, a complaint group identification method is implemented. The readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.

[0119] In one embodiment, a computer device is provided. The computer device may be a terminal device, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, and a network interface connected through a device bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a readable storage medium. The readable storage medium stores computer-readable instructions. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer-readable instructions are executed by the processor, a method for identifying a complaint group is implemented. The readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.

[0120] In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, and when the processor executes the computer-readable instructions, the steps of the complaint group identification method described above are implemented.

[0121] In one embodiment, a readable storage medium is provided, the readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the complaint gang identification method as described above are implemented. A person of ordinary skill in the art can understand that the implementation of all or part of the process in the above-mentioned embodiment method can be completed by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a non-volatile readable storage medium or a volatile readable storage medium. When the computer-readable instructions are executed, they may include the process of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0122] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0123] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for identifying a complaint group, characterized in that: include: Obtain the user data to be identified; Determine whether the user data to be identified belongs to the sample data of the target clustering model in the training process, wherein the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, and the knowledge graph is constructed based on users with associated relationships and their associated sample data; If yes, the risk identification result associated with the sample data to which the to-be-identified user data belongs is used as the group risk identification result corresponding to the to-be-identified user data, wherein the risk identification result is determined based on the result obtained by clustering each of the sample data based on the target clustering model; If not, cluster identification is performed on the user data to be identified based on a pre-built knowledge graph to obtain a group risk identification result corresponding to the user data to be identified.

2. The method for identifying complaint groups according to claim 1, characterized in that: The knowledge graph is constructed based on the following steps: Get multiple sample data; Based on each of the sample data, determining the association relationship between each complaining user and other users, wherein the complaining user refers to a user who has made a complaint; Generate triples based on users with association relationships and their associated sample data; Based on each of the triples, the knowledge graph is constructed.

3. The method for identifying complaint groups according to claim 1, characterized in that: The target clustering model is trained based on the following steps: Performing level division on each node in the knowledge graph to obtain a division result of each node; Performing cluster analysis on each node in the knowledge graph to obtain a target clustering result, wherein the target clustering result is used to characterize the community to which each node belongs; Based on the sample data corresponding to each of the nodes, the community to which they belong, and the division results, training the target clustering model; Obtain an unlabeled sample data set, and input each unlabeled sample data in the unlabeled sample data set into the target clustering model to obtain a prediction result of each unlabeled sample data, wherein the unlabeled sample data set refers to a sample data set that is not labeled; Determine a sample label of the unlabeled sample data based on the prediction result of each of the unlabeled sample data; Based on the unlabeled sample data and the sample labels, the target clustering model is optimized and trained.

4. The method for identifying complaint groups according to claim 3, characterized in that: After optimizing and training the target clustering model based on the unlabeled sample data and the sample labels, the method further includes: Determine the model clustering results corresponding to each sample data output by the target clustering model during the model training process; For each sample data: based on the model clustering result corresponding to the sample data, the sample data is cross-analyzed with the sample data of each node in the knowledge graph to obtain the risk identification result corresponding to the sample data.

5. The method for identifying complaint groups according to claim 3, characterized in that: The step of performing level division on each node in the knowledge graph to obtain the division result of each node includes: According to each of the sample data, the nodes in the knowledge graph are divided into complaint nodes and non-complaint nodes; For any complaint node: if there is at least one edge between the complaint node and other complaint nodes, the complaint node is regarded as a primary user node; if there is no edge between the complaint node and other complaint nodes, the complaint node is regarded as a secondary user node; The non-complaint nodes are divided into three levels of user nodes.

6. The method for identifying complaint groups according to claim 1, characterized in that: The clustering analysis is performed on each node in the knowledge graph to obtain a target clustering result, including: Performing cluster analysis on each node in the knowledge graph to obtain the community to which each node belongs; Generate a fully connected graph based on each node of the same community; Determine the weight between any two nodes in the fully connected graph; Based on the weights between all nodes, a probability transfer matrix is ​​constructed; Based on the probability transition moment, determining a community label corresponding to each node; For any community: based on each weight in the fully connected graph and the community label corresponding to each node, each node in the community is screened and filtered to obtain the target clustering result.

7. The method for identifying complaint groups according to claim 1, characterized in that: The clustering identification of the user data to be identified based on the pre-built knowledge graph to obtain the group risk identification result corresponding to the user data to be identified includes: Based on the user data to be identified, updating the knowledge graph; Cluster and grade each node in the updated knowledge graph; According to the clustering results and the division results, the user data to be identified and the sample data of each node in the updated knowledge graph are cross-analyzed to obtain the gang risk identification result.

8. A complaint group identification device, characterized in that: include: An acquisition module, used to acquire user data to be identified; A judgment module, used to judge whether the user data to be identified belongs to the sample data of the target clustering model in the training process, wherein the target clustering model is obtained by semi-supervised training based on the sample data corresponding to each node in the knowledge graph and the collected unlabeled sample data set, and the knowledge graph is constructed based on users with associated relationships and their associated sample data; a determination module, for, if yes, using the risk identification result associated with the sample data to which the to-be-identified user data belongs as the group risk identification result corresponding to the to-be-identified user data, wherein the risk identification result is determined based on the result obtained by clustering each of the sample data based on the target clustering model; The identification module is used to cluster and identify the user data to be identified based on a pre-built knowledge graph, if not, to obtain the group risk identification result corresponding to the user data to be identified.

9. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executed on the processor, characterized in that: When the processor executes the computer-readable instructions, it implements the complaint group identification method as described in any one of claims 1 to 7.

10. A readable storage medium having computer readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, the complaint group identification method as described in any one of claims 1 to 7 is implemented.

Images