A modeling method based on user behavior data and a user behavior prediction method

Abstract

The present invention discloses a modeling method and a user behavior prediction method based on user behavior data, which are applied to the field of artificial intelligence, and include: constructing a hybrid relationship network according to the historical behavior data of users, and determining a hybrid relationship matrix including implicit relationships and explicit relationships according to the hybrid relationship network; constructing a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data; obtaining a feature extraction model based on the hybrid relationship matrix and the multi-perspective data matrix, and obtaining user behavior features according to the feature extraction model; constructing a user behavior prediction model based on the prediction model and the user behavior features. The present invention obtains more significant and diverse-scale user behavior features based on the hybrid relationship matrix containing implicit relationships and explicit relationships, and the multi-perspective data matrix, which can characterize the differences in different perspectives and the consistency in the same perspective, making the constructed prediction model more reliable and accurate.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and particularly relates to a modeling method and a user behavior prediction method based on user behavior data. Background Art

[0002] The modeling of game user behavior mainly mines and explores big data through various statistical analysis methods. For example, the current time series behavior modeling method based on a relational network constructs a relational network by statistically analyzing the interaction relationships between users, and uses the relational network and user feature data as model inputs, which are then input into a time series model to predict future user behavior trends, thereby constructing a user behavior model. However, this method ignores the potential characteristics of users, resulting in inaccurate prediction of the constructed user behavior prediction model. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a modeling method and device based on user behavior data, a user behavior prediction method and device, an electronic device, and a storage medium, which solve the problem that the relational network and user feature data in the prior art cannot fully represent user behavior, resulting in an unreliable and inaccurate user behavior prediction model.

[0004] To solve the above technical problems, the present invention provides a modeling method based on user behavior data, including:

[0005] Constructing a mixed relational network according to the historical behavior data of users, and determining a mixed relational matrix including implicit relationships and explicit relationships according to the mixed relational network;

[0006] Constructing a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data;

[0007] Obtaining a feature extraction model based on the mixed relational matrix and the multi-perspective data matrix, and obtaining user behavior features according to the feature extraction model;

[0008] Constructing a user behavior prediction model based on a prediction model and the user behavior features.

[0009] Optionally, constructing a mixed relational network according to the historical behavior data of users includes:

[0010] Constructing a node set according to user data;

[0011] Constructing an edge set according to the behavior relationship data between users;

[0012] Taking the preference label data between users as an implicit relationship component;

[0013] Taking the user attention behavior data as an explicit relationship component;

[0014] Obtain a mixed relationship weight set according to the implicit relationship component and the explicit relationship component;

[0015] Construct the mixed relationship network from the node set, the edge set, and the mixed relationship weight set.

[0016] Optionally, use the preference label data between users as the implicit relationship component, including:

[0017] The expression formula of the implicit relationship component is: ;

[0018] represents the preference label relationship value of the new user; and represent the relationship values of short-term answer likes and long-term answer like labels respectively; , and are preset positive hyperparameters;

[0019] represents user and user 's preference label relationship value, and the calculation formula is: ;

[0020] represents user and user 's number of common preference labels; represents user 's number of preference labels.

[0021] Optionally, determine a mixed relationship matrix including implicit and explicit relationships according to the mixed relationship network, including:

[0022] Use the robust principal component analysis method to perform matrix decomposition on the initial mixed relationship matrix in the mixed relationship weight set, and use the augmented Lagrangian multiplier method, singular value decomposition, and soft threshold to solve the optimal solution of the optimization problem to obtain a low-rank relationship matrix and a sparse matrix;

[0023] Use the low-rank relationship matrix as the mixed relationship matrix.

[0024] Optionally, obtain a feature extraction model based on the mixed relationship matrix and the multi-perspective data matrix, and obtain user behavior features according to the feature extraction model, including:

[0025] Input the mixed relationship matrix and the multi - perspective data matrix into a multi - perspective auto - encoder, and train the multi - perspective auto - encoder based on a loss function to obtain the feature extraction model; the multi - perspective graph auto - encoder includes an encoder and a dual decoder;

[0026] Obtain the user behavior features based on the feature extraction model.

[0027] Optionally, constructing a multi - perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data includes:

[0028] Divide the user behavior data into m user behavior subsets;

[0029] Divide the economic behavior data into m economic behavior subsets;

[0030] Concatenate the user behavior subsets and the economic behavior subsets to obtain a data matrix representing time - domain features;

[0031] Use Fourier transform to obtain frequency - domain features from the historical behavior data, and obtain a data matrix representing frequency - domain features;

[0032] Obtain the multi - perspective data matrix based on the data matrix representing time - domain features and the data matrix representing frequency - domain features.

[0033] The present invention also provides a user behavior prediction method, including:

[0034] Obtain the user behavior data to be predicted;

[0035] Input the user behavior data into the user behavior prediction model obtained by the above - mentioned modeling method based on user behavior data, and predict the future behavior of the user. The user behavior prediction model includes a feature extraction model and a prediction model.

[0036] The present invention also provides a modeling device based on user behavior data, including:

[0037] A mixed relationship matrix acquisition module, configured to construct a mixed relationship network according to the historical behavior data of the user, and determine a mixed relationship matrix including implicit relationships and explicit relationships according to the mixed relationship network;

[0038] A multi - perspective data matrix acquisition module, configured to construct a multi - perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data;

[0039] A user behavior feature acquisition module, configured to obtain a feature extraction model based on the mixed relationship matrix and the multi - perspective data matrix, and obtain user behavior features according to the feature extraction model;

[0040] A user behavior prediction model construction module, configured to construct a user behavior prediction model based on a prediction model and the user behavior characteristics.

[0041] The present invention also provides a user behavior prediction device, including:

[0042] A data acquisition module, configured to acquire user behavior data to be predicted;

[0043] A prediction module, configured to input the user behavior data into the user behavior prediction model obtained by the above-mentioned modeling method based on user behavior data, and predict the future behavior of the user. The user behavior prediction model includes a feature extraction model and a prediction model.

[0044] The present invention also provides an electronic device, including:

[0045] A memory, configured to store a computer program;

[0046] A processor, configured to implement the steps of the above-mentioned modeling method and / or user behavior prediction method based on user behavior data when executing the computer program.

[0047] The present invention also provides a storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, the steps of the above-mentioned modeling method based on user behavior data are implemented.

[0048] It can be seen that the present invention constructs a hybrid relationship network according to the historical behavior data of the user, and determines a hybrid relationship matrix including implicit relationships and explicit relationships according to the hybrid relationship network; constructs a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data; obtains a feature extraction model based on the hybrid relationship matrix and the multi-perspective data matrix, and obtains user behavior characteristics according to the feature extraction model; constructs a user behavior prediction model based on the prediction model and the user behavior characteristics. The present invention constructs a hybrid relationship network according to the explicit relationships in the behavior data, collocates the implicit relationships of the user behavior data, and forms a multi-perspective data matrix with the user behavior data and economic behavior data. Based on the hybrid relationship matrix including implicit relationships and explicit relationships, and the multi-perspective data matrix, user behavior characteristics are obtained. The user behavior characteristics obtained in this way are discriminative, can characterize the differences between different perspectives and the consistency of the same perspective, can strengthen the prediction effect of the prediction model, and make the constructed prediction model more reliable and accurate.

[0049] In addition, the present invention also provides a modeling device based on user behavior data, a user behavior prediction method and device, an electronic device, and a storage medium, which also have the above-mentioned beneficial effects. Description of the Drawings

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained according to the provided accompanying drawings.

[0051] Figure 1 It is a flowchart of a modeling method based on user behavior data provided by an embodiment of the present invention;

[0052] Figure 2 It is an example diagram of the process for obtaining a low-rank relationship matrix provided by an embodiment of the present invention;

[0053] Figure 3 It is an example diagram of the process for obtaining user behavior characteristics provided by an embodiment of the present invention;

[0054] Figure 4 It is an example diagram of the process for a modeling method based on user behavior data provided by an embodiment of the present invention;

[0055] Figure 5 It is a flowchart of a user behavior prediction method provided by an embodiment of the present invention;

[0056] Figure 6 It is a schematic structural diagram of a modeling device based on user behavior data provided by an embodiment of the present invention;

[0057] Figure 7 It is a schematic structural diagram of a user behavior prediction device provided by an embodiment of the present invention;

[0058] Figure 8 It is a schematic structural diagram of a modeling device based on user behavior data provided by an embodiment of the present invention. Specific embodiments

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0060] Please refer to Figure 1 , Figure 1 It is a flowchart of a modeling method based on user behavior data provided by an embodiment of the present invention. The method may include:

[0061] S101: Construct a hybrid relationship network based on the historical behavior data of users, and determine a hybrid relationship matrix containing implicit relationships and explicit relationships according to the hybrid relationship network.

[0062] The execution subject of this embodiment is a terminal. This embodiment does not limit the type of the terminal, as long as it can complete the operations of the modeling method based on user behavior data. This embodiment collects a large amount of historical behavior data of users related to the in-game system. It can be understood that the collected historical behavior data of a large number of users includes the historical behavior data corresponding to each user. For example, answering question behavior data, consumption behavior data, and behavior data related to the in-game system, etc. Further, this embodiment can also group users according to specified attributes (such as region, age, label field, etc.), and perform modeling separately for different groups of users, so as to obtain user behavior prediction models with various fixed attributes, and the behavior prediction will be more accurate and refined.

[0063] This embodiment constructs a hybrid relationship network with implicit relationships and explicit relationships based on historical behavior data, and obtains a hybrid relationship matrix with implicit relationships and explicit relationships based on this network. The hybrid relationship network is a network model that combines multiple relationship inductive biases, aiming to improve the performance of deep learning models by constructing and processing relationships of different types of nodes and edges.

[0064] Further, the construction of the hybrid relationship network according to the historical behavior data of users may include the following steps:

[0065] Step 21: Construct a node set according to user data;

[0066] Step 22: Construct an edge set according to the behavior relationship data between users;

[0067] Step 23: Use the preference label data between users as the implicit relationship component;

[0068] Step 24: Use the user attention behavior data as the explicit relationship component;

[0069] Step 25: Obtain a hybrid relationship weight set according to the implicit relationship component and the explicit relationship component;

[0070] Step 26: Construct a hybrid relationship network from the node set, the edge set, and the hybrid relationship weight set.

[0071] Specifically, assume that there are different users in a certain group, and construct a hybrid relationship network , where represents the node set, that is, the user set. That is to say, the user data in step 21 is each user; Represents an edge set, where represents the relationship from user to user . Represents a set of mixed relationship weights, that is, the initial mixed relationship weight matrix. Among them, represents the mixed relationship weight from user to user . Since the mixed relationship network is directional, and may not be equal. For the mixed relationship weight , it is composed of the explicit relationship component and the implicit relationship component . For specific calculations, reference can be made to Figure 2 . Figure 2 is an example diagram of the process for obtaining a low-rank relationship matrix provided by an embodiment of the present invention. As can be seen from Figure 2 , the explicit relationship component directly uses user attention behavior data, and the implicit relationship component mainly uses preference label data between users. Specifically, assume that user follows user , then the explicit relationship component ; assume that user follows user , then the explicit relationship component ; assume that user and user mutually follow each other, then the explicit relationship component . The implicit relationship component mainly uses the preference labels between users, and specifically may include the following steps:

[0072] The expression formula of the implicit relationship component is: ;

[0073] represents the preference label relationship value of the new user. Since is only selected once to represent the general preference of the user, and it is assumed to be a stable component that does not change with time; and represent the relationship values of short-term answer liking and long-term answer liking labels respectively; , and are preset positive hyperparameters;

[0074] represents user and user 's preference label relationship value, and the calculation formula is: ;

[0075] represents user and the number of tags preferred by the user ; Indicates the number of preferred tags of the user ; Indicates the set of preferred tags of the user Specifically, it is composed of the preferred tags of new users and the liked tags of answers

[0076] The user to the user The mixed relationship weight The formula expression is as follows .

[0077] At present, when modeling time series behavior based on the relationship network, only the explicit relationship features of users are basically utilized, and the utilization rate of implicit relationship features is not high, which indirectly weakens the prediction effect of the behavior prediction model. In this embodiment, when constructing the mixed relationship network, the explicit relationship component of the user's attention behavior and the implicit relationship component of the preferred tags between users are fully explored, further enhancing the mixed relationship network

[0078] Furthermore, determining the mixed relationship matrix including implicit and explicit relationships according to the mixed relationship network specifically may include the following steps

[0079] Step 31: Use the robust principal component analysis method to perform matrix decomposition on the initial mixed relationship matrix in the mixed relationship weight set, and use the augmented Lagrangian multiplier method, singular value decomposition, and soft threshold to solve the optimal solution of the optimization problem to obtain the low-rank relationship matrix and the sparse matrix

[0080] Step 32: Use the low-rank relationship matrix as the mixed relationship matrix

[0081] Specifically, this embodiment further considers the influence of outliers on the mixed relationship network, and uses a noise suppression algorithm to weaken the damage of outliers to the mixed relationship network, and a more significant relationship network can be obtained, and thus a more representative mixed relationship matrix can be obtained, as Figure 2 shown. Assume that the mixed relationship matrix has a good data structure, that is, it is low-rank, and only a small part is contaminated by noise or outliers, that is, it is sparse. Therefore, the calculated mixed relationship network is decomposed by robust principal component analysis, and the following problem is specifically optimized

[0082] ;

[0083] ;

[0084] Indicates the initial mixed relationship matrix represents a low-rank relational matrix; represents a sparse matrix; represents the nuclear norm of the low-rank relational matrix; represents the norm of the sparse matrix; is a regularization parameter used to balance the weights of the low-rank component and the sparse component. To solve the above optimization problem, it is necessary to construct a Lagrangian function with a penalty term , and use the augmented Lagrangian multiplier method, combined with the singular value decomposition and the soft threshold method to solve the optimal low-rank relational matrix and the sparse matrix . Among them, Y is the Lagrangian multiplier corresponding to the equality constraint A - L - S = 0, and u is the coefficient of the penalty function used to control the influence of the equality constraint on the objective function. Substitute the obtained low-rank mixing matrix to replace the above calculated mixed relational network , and input it to the next-stage feature extraction model, which not only ensures the global data structure but also can suppress the damage of some abnormal noises to the network.

[0085] S102: Construct a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data.

[0086] This embodiment does not limit the specific construction process of the multi-perspective data matrix, nor does it limit the features in the multi-perspective data matrix. As long as it satisfies the construction of the multi-perspective data matrix based on the user behavior data and economic behavior data.

[0087] Further, the construction of the multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data may specifically include the following steps:

[0088] Step 41: Divide the user behavior data into m user behavior subsets;

[0089] Step 42: Divide the economic behavior data into m economic behavior subsets;

[0090] Step 43: Concatenate the user behavior subsets and the economic behavior subsets to obtain data matrices representing time-domain features;

[0091] Step 44: Use Fourier transform to obtain frequency-domain features from the historical behavior data to obtain data matrices representing frequency-domain features;

[0092] Step 45: Obtain the multi-perspective data matrix based on the data matrices representing time-domain features and the data matrices representing frequency-domain features.

[0093] Specifically, two major categories of data, namely answering behavior data (including answering duration, answering accuracy rate, etc.) and economic behavior data (number of times of purchasing items, recharge amount, etc.), are filtered out from a large amount of user behavior data. Each behavior category is divided into non-overlapping and equally sized behavior subsets. The data matrix of each perspective respectively extracts one behavior subset from the answering behavior category and the economic behavior category for feature splicing. Therefore, data matrices representing time-domain features are obtained. Fourier transform is used to mine frequency-domain information from time-series data (i.e., historical behavior data). Assume a time-series data matrix of length , and the data matrix representing time-domain features obtained through Fourier transform. The data matrix representing time-domain features is spliced into the data matrix representing frequency-domain features to obtain a multi-perspective data matrix containing time-domain features and frequency-domain features.

[0094] S103: Obtain a feature extraction model based on the hybrid relationship matrix and the multi-perspective data matrix, and obtain user behavior features according to the feature extraction model.

[0095] In this embodiment, based on the hybrid relationship matrix obtained in step S101 and the multi-perspective data matrix obtained in step S102, deeper-level user behavior features are further obtained. This embodiment does not limit the specific feature extraction model.

[0096] Furthermore, obtaining user behavior features based on the hybrid relationship matrix and the multi-perspective data matrix specifically may include the following steps:

[0097] Step 51: Input the hybrid relationship matrix and the multi-perspective data matrix into a multi-perspective autoencoder, and train the multi-perspective autoencoder based on the loss function to obtain a feature extraction model; the multi-perspective graph autoencoder includes an encoder and a dual decoder;

[0098] Step 52: Obtain user behavior features based on the feature extraction model.

[0099] Specifically, the feature extraction model in this embodiment is a multi-perspective autoencoder composed of an encoder based on a graph convolutional neural network and a dual decoder. Assume the th multi-perspective data matrix is , and the low-rank relationship matrix (i.e., the hybrid relationship matrix) is , then the output of the th graph convolutional neural network layer is expressed as , where , is the identity matrix to ensure that multiplying it with the feature matrix does not ignore the node's own features, is ​The degree matrix, i.e., , represents a non-linear activation function, represents the element in the i-th row and j-th column of the low-rank relationship matrix, is the weight matrix of the -th layer, .

[0100] In this embodiment, considering that the decoder of the original graph autoencoder reconstructs the adjacency matrix from the encoded features in an inner product form, which may lose the feature information of the data itself and cause the feature extraction model to ignore learning the intrinsic characteristics of the data. Therefore, in the dual decoder of this embodiment, one decoder reconstructs the behavioral features of user i to obtain the reconstructed multi-view data matrix , and the other decoder reconstructs the low-rank relationship matrix in an inner product form to obtain the reconstructed mixed relationship matrix , which is expressed by the formula , where represents a non-linear activation function. The loss function of the final multi-view graph autoencoder is:

[0101] ;

[0102] ;

[0103] is the loss of the reconstructed multi-view data matrix, is the loss of the reconstructed mixed relationship matrix, is the consistency loss, is the regularization term, is the i-th multi-view data matrix, is the i-th reconstructed multi-view data matrix, is the mixed relationship matrix, is the reconstructed mixed relationship matrix, and represent the encoder outputs, which are the behavioral features of user i and user j respectively, m represents the number of subsets, is the weight, represents the Frobenius norm of the matrix, which is a matrix norm used to measure the magnitude of matrix elements; represents the penalty term coefficient to prevent the model from overfitting.

[0104] During the training process of the multi-view autoencoder, the optimal parameters are obtained by optimizing the loss function. Since the multi-view graph encoder shares network parameters, after the training is completed, user behavioral feature representations Perform splicing to obtain the final user behavior characteristics 。 Figure 3 This is a flowchart example of a user behavior feature acquisition method provided by an embodiment of the present invention.

[0105] S104: Construct a user behavior prediction model based on the prediction model and user behavior characteristics.

[0106] This embodiment does not limit the prediction model. For example, the prediction model in this embodiment can be a long short-term memory network with an attention mechanism. By adding an attention layer between the long short-term memory network and the fully connected layer, it is possible to pay more attention to relevant main information in the prediction and ensure the accuracy of the prediction. Input the output of the feature extraction model, that is, the user behavior characteristics, into the long short-term memory network with an attention mechanism to predict the future user behavior trend. Further, when the number of new users accumulates to a certain extent, the hybrid relationship network can be reconstructed, and the feature extraction model and the prediction model can be retrained, that is, the previously constructed hybrid relationship network, feature extraction model, and prediction model are adjusted to perform optimization and update operations to achieve more accurate prediction of the future economic trends of each group.

[0107] Applying the modeling method based on user behavior data provided by the embodiments of the present invention, a hybrid relationship network is constructed according to the historical behavior data of users, and a hybrid relationship matrix including implicit relationships and explicit relationships is determined according to the hybrid relationship network; a multi-perspective data matrix is constructed based on the user behavior data and economic behavior data in the historical behavior data; a feature extraction model is obtained based on the hybrid relationship matrix and the multi-perspective data matrix, and user behavior features are obtained according to the feature extraction model; a user behavior prediction model is constructed based on the prediction model and the user behavior features. According to the explicit relationships in the behavior data, this method constructs a hybrid relationship network by combining with the implicit relationships of the user behavior data, and forms a multi-perspective data matrix with the user behavior data and the economic behavior data. Based on the hybrid relationship matrix including implicit relationships and explicit relationships, and the multi-perspective data matrix, user behavior features are obtained. The user behavior features obtained in this way are discriminative, can characterize the differences between different perspectives and the consistency of the same perspective, can strengthen the prediction effect of the prediction model, and make the constructed prediction model more reliable and accurate. Moreover, by combining the explicit relationship component of the user attention behavior data with the implicit relationship component of the preference label data between users to construct a hybrid relationship network, and using robust principal component analysis to obtain a low-rank relationship matrix, not only the connection and structural characteristics of the relationship network are strengthened, but also the damage of outliers to the relationship network is suppressed. Moreover, based on the combination of the answering behavior and the consumption behavior to form multi-perspective time-domain features for Fourier transform to obtain multi-perspective frequency-domain features, and using a dual graph autoencoder for feature extraction, the user behavior features obtained in this way can retain the intuitive time-domain features and also mine the discriminative frequency-domain features; moreover, using a long short-term memory network with an attention mechanism as the user behavior prediction model can pay more attention to relevant main information in the prediction and ensure the accuracy of the prediction; moreover, training the user behavior prediction model based on the user behavior features obtained by the above method can strengthen the behavior prediction effect of the behavior user behavior prediction model and make the user behavior prediction model more reliable.

[0108] For the convenience of understanding the present invention, please specifically refer to Figure 4 , Figure 4 which is a flowchart example of a modeling method based on user behavior data provided by the embodiments of the present invention, and specifically may include:

[0109] Obtain user behavior data, cluster the user behavior data based on user attributes to obtain each group. Feature extraction mainly includes inputting the low-rank relationship matrix and the multi-view data matrix into the multi-view graph autoencoder for feature extraction to mine user behavior features with more hierarchical representations. Among them, the low-rank relationship matrix is obtained by performing robust principal component analysis on the mixed relationship matrix, and the multi-view data matrix includes both time-domain features and frequency-domain features. Based on the user behavior features obtained from feature extraction, perform user behavior prediction, predict the future behavior trend of users in this group, and then combine the relevant behavior analysis of the game economic system to obtain the economic trend of the overall users in the corresponding group.

[0110] Please refer to Figure 5 , Figure 5 is a flowchart of a user behavior prediction method provided by an embodiment of the present invention. The method may include:

[0111] Step 201: Obtain the user behavior data to be predicted;

[0112] Step 202: Input the user behavior data into the user behavior prediction model obtained by the above modeling method based on user behavior data to predict the future behavior of the user. The user behavior prediction model includes a feature extraction model and a prediction model.

[0113] Input the user behavior data to be predicted into the user behavior prediction model obtained by using the above modeling method based on user behavior data to predict the future behavior of the user. In actual application, the user behavior data of a group can be input into the corresponding feature extraction model of this group to obtain the user behavior features of this group, and input the user behavior features into the prediction model to predict the future behavior trend of users in this group. Then, by combining the relevant behavior analysis of the game economic system, the economic trend of the overall users in the corresponding group can be obtained.

[0114] Applying the user behavior prediction method provided by the embodiment of the present invention, by inputting the obtained user behavior data to be predicted into the user behavior prediction model obtained by the above modeling method based on user behavior data, the future behavior of the user is predicted. The feature extraction model in the behavior prediction model of this method can extract more representative and discriminative user behavior features, making the prediction effect of the behavior prediction model better and the prediction result more reliable.

[0115] Next, the modeling device based on user behavior data provided by the embodiment of the present invention will be introduced. The modeling device based on user behavior data described below can be mutually corresponding and referred to with the modeling method based on user behavior data described above.

[0116] Specifically, please refer to Figure 6, Figure 6 The following is a schematic structural diagram of a modeling device based on user behavior data provided by an embodiment of the present invention, which may include:

[0117] A hybrid relationship matrix acquisition module 100, configured to construct a hybrid relationship network according to historical behavior data of users, and determine a hybrid relationship matrix including implicit relationships and explicit relationships according to the hybrid relationship network;

[0118] A multi-perspective data matrix acquisition module 200, configured to construct a multi-perspective data matrix based on user behavior data and economic behavior data in the historical behavior data;

[0119] A user behavior feature acquisition module 300, configured to obtain a feature extraction model based on the hybrid relationship matrix and the multi-perspective data matrix, and obtain user behavior features according to the feature extraction model;

[0120] A user behavior prediction model construction module 400, configured to construct a user behavior prediction model based on a prediction model and the user behavior features.

[0121] Based on the above embodiment, the hybrid relationship matrix acquisition module 100 may include:

[0122] A node set construction unit, configured to construct a node set according to user data;

[0123] An edge set construction unit, configured to construct an edge set according to behavior relationship data between users;

[0124] An implicit relationship component unit, configured to use preference label data between users as an implicit relationship component;

[0125] An explicit relationship component unit, configured to use user attention behavior data as an explicit relationship component;

[0126] A hybrid relationship weight set construction unit, configured to obtain a hybrid relationship weight set according to the implicit relationship component and the explicit relationship component;

[0127] A hybrid relationship network construction unit, configured to construct the hybrid relationship network from the node set, the edge set, and the hybrid relationship weight set.

[0128] Based on the above embodiment, the implicit relationship component unit may include:

[0129] The expression formula of the implicit relationship component is: ;

[0130] represents the preference label relationship value of a new user; and respectively represent the relationship value of short-term answer likes and the relationship value of long-term answer like tags; , and are preset positive hyperparameters;

[0131] represents user and user 's preference tag relationship value, and the calculation formula is: ;

[0132] represents user and user 's common preference tag count; represents user 's preference tag count.

[0133] Based on the above embodiments, the hybrid relationship matrix acquisition module 100 may include:

[0134] A decomposition unit for performing matrix decomposition on the initial hybrid relationship matrix in the hybrid relationship weight set by using the robust principal component analysis method, and using the augmented Lagrangian multiplier method, singular value decomposition, and soft threshold to solve the optimal solution of the optimization problem, obtaining a low-rank relationship matrix and a sparse matrix;

[0135] A hybrid relationship matrix determination unit for using the low-rank relationship matrix as the hybrid relationship matrix.

[0136] Based on any of the above embodiments, the user behavior feature acquisition module 300 may include:

[0137] A training unit for inputting the hybrid relationship matrix and the multi-view data matrix into a multi-view autoencoder, and training the multi-view autoencoder based on a loss function to obtain the feature extraction model; the multi-view graph autoencoder includes an encoder and a dual decoder;

[0138] A user behavior feature acquisition unit for obtaining the user behavior features based on the feature extraction model.

[0139] Based on the above embodiments, the multi-view data matrix acquisition module 200 may include:

[0140] A user behavior subset division unit for dividing the user behavior data into m user behavior subsets;

[0141] An economic behavior subset division unit for dividing the economic behavior data into m economic behavior subsets;

[0142] A splicing unit for splicing the user behavior subsets and the economic behavior subsets to obtain A data matrix representing time-domain features;

[0143] A frequency-domain extraction unit, configured to obtain frequency-domain features from the historical behavior data by using Fourier transform, and obtain a data matrix representing the frequency-domain features;

[0144] A multi-view data matrix acquisition unit, configured to obtain the multi-view data matrix based on the data matrix representing the time-domain features and the data matrix representing the frequency-domain features.

[0145] It should be noted that the order of the modules and units in the above modeling device based on user behavior data can be changed before and after without affecting the logic.

[0146] The modeling device based on user behavior data provided by the embodiment of the present invention is applied, through the hybrid relationship matrix acquisition module 100, used to build a hybrid relationship network according to the user's historical behavior data, and determine the hybrid relationship matrix containing implicit relationships and explicit relationships according to the hybrid relationship network; the multi-perspective data matrix acquisition module 200, used to build a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data; the user behavior feature acquisition module 300, used to obtain a feature extraction model based on the hybrid relationship matrix and the multi-perspective data matrix, and obtain user behavior features according to the feature extraction model; the user behavior prediction model construction module 400, used to construct a user behavior prediction model based on the prediction model and the user behavior features. This device constructs a hybrid relationship network based on the explicit relationship in the behavior data and the implicit relationship of the user behavior data, and forms a multi-perspective data matrix with the user behavior data and the economic behavior data, and obtains user behavior features based on the hybrid relationship matrix containing implicit relationships and explicit relationships, and the multi-perspective data matrix. The user behavior features obtained in this way are discriminative, can characterize the differences between different perspectives and the consistency of the same perspective, can enhance the prediction effect of the prediction model, and make the constructed prediction model more reliable and accurate. In addition, the explicit relationship component of the user attention behavior data is combined with the implicit relationship component of the preference label data between users to construct a hybrid relationship network, and the low-rank relationship matrix is ​​obtained by robust principal component analysis, which not only strengthens the connection and structural characteristics of the relationship network, but also suppresses the damage of outliers to the relationship network. In addition, based on the combination of answering behavior and consumption behavior into multi-perspective time domain features, Fourier transform is performed to obtain multi-perspective frequency domain features, and dual graph autoencoders are used for feature extraction. The user behavior features obtained in this way can retain the intuitive features of the time domain and can also mine the discriminative features of the frequency domain; and the long short-term memory network with attention mechanism is used as the user behavior prediction model, which can pay more attention to the relevant main information in the prediction and ensure the accuracy of the prediction; and the user behavior prediction model is trained based on the user behavior features obtained by the above method, which can enhance the behavior prediction effect of the behavior user behavior prediction model and make the user behavior prediction model more reliable.

[0147] The user behavior prediction device provided by an embodiment of the present invention is introduced below. The user behavior prediction device described below and the user behavior prediction method described above can be referenced to each other.

[0148] Please refer to Figure 7 , Figure 7 A schematic diagram of the structure of a user behavior prediction device provided by an embodiment of the present invention may include:

[0149] A data acquisition module 500 is used to acquire user behavior data to be predicted;

[0150] A prediction module 600 is configured to input the user behavior data into a user behavior prediction model obtained by the above-mentioned modeling method based on user behavior data to predict the future behavior of the user. The user behavior prediction model includes a feature extraction model and a prediction model.

[0151] Applying the user behavior prediction device provided by the embodiment of the present invention, a data acquisition module 500 is configured to acquire user behavior data to be predicted; a prediction module 600 is configured to input the user behavior data into a user behavior prediction model obtained by the above-mentioned modeling method based on user behavior data to predict the future behavior of the user. The user behavior prediction model includes a feature extraction model and a prediction model. The feature extraction model in the behavior prediction model of this device can extract more representative and discriminative user behavior features, making the prediction effect of the behavior prediction model better and the prediction result more reliable.

[0152] The following introduces the electronic device provided by the embodiment of the present invention. The electronic device described below can be correspondingly referred to the above-mentioned modeling method based on user behavior data and / or user behavior prediction method.

[0153] Please refer to Figure 8 , Figure 8 which is a schematic structural diagram of an electronic device provided by the embodiment of the present invention, and may include:

[0154] A memory 10 is configured to store a computer program;

[0155] A processor 20 is configured to execute the computer program to implement the above-mentioned modeling method based on user behavior data.

[0156] The memory 10, the processor 20, and the communication interface 31 all complete communication with each other through a communication bus 32.

[0157] In the embodiment of the present invention, the memory 10 is used to store one or more programs. The program may include program codes, and the program codes include computer operation instructions. In the embodiment of the present invention, the memory 10 may store programs for implementing the following functions:

[0158] Construct a hybrid relationship network according to the historical behavior data of the user, and determine a hybrid relationship matrix including implicit relationships and explicit relationships according to the hybrid relationship network;

[0159] Construct a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data;

[0160] Obtain a feature extraction model based on the hybrid relationship matrix and the multi-perspective data matrix, and obtain user behavior features according to the feature extraction model;

[0161] A user behavior prediction model is constructed based on a prediction model and user behavior characteristics;

[0162] and / or;

[0163] Obtain user behavior data to be predicted;

[0164] Input the user behavior data into the user behavior prediction model obtained by the above-mentioned modeling method based on user behavior data, and predict the future behavior of the user. The user behavior prediction model includes a feature extraction model and a prediction model.

[0165] In a possible implementation, the memory 10 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function, etc.; the data storage area may store data created during use.

[0166] In addition, the memory 10 may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include NVRAM. The memory stores an operating system and operation instructions, executable modules or data structures, or subsets thereof, or extended sets thereof. Among them, the operation instructions may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and processing hardware-based tasks.

[0167] The processor 20 may be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic devices. The processor 20 may be a microprocessor or any conventional processor, etc. The processor 20 may call the program stored in the memory 10.

[0168] The communication interface 31 may be an interface of a communication module for connecting to other devices or systems.

[0169] Of course, it should be noted that Figure 7 The structure shown does not limit the modeling device based on user behavior data in the embodiments of the present invention. In practical applications, the modeling device based on user behavior data may include more or fewer components than Figure 7 shown, or combine some components.

[0170] Next, the storage medium provided by the embodiments of the present invention will be introduced. The storage medium described below can be mutually corresponding and referred to with the above-mentioned modeling method based on user behavior data.

[0171] The present invention also provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned modeling method based on user behavior data are implemented.

[0172] The storage medium may be a computer-readable storage medium, which may include: various media capable of storing program codes such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical discs.

[0173] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference may be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference may be made to the description in the method part.

[0174] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to their functions in the above description. Whether these functions are executed in the form of hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0175] Finally, it should also be noted that in this article, relationships such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0176] The above has introduced in detail a modeling method and device based on user behavior data, a user behavior prediction method and device, an electronic device, and a storage medium provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A modeling method based on user behavior data, characterized in that: include: Constructing a hybrid relationship network according to the historical behavior data of the user, and determining a hybrid relationship matrix including implicit relationships and explicit relationships according to the hybrid relationship network; Constructing a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data; Obtaining a feature extraction model based on the mixed relationship matrix and the multi-view data matrix, and obtaining user behavior features according to the feature extraction model; A user behavior prediction model is constructed based on the prediction model and the user behavior characteristics; Build a hybrid relationship network based on the user's historical behavior data, including: Build a node set based on user data; The edge set is constructed based on the behavioral relationship data between users; The preference label data between users is used as the implicit relationship component; Use user attention behavior data as explicit relationship components; Obtaining a mixed relationship weight set according to the implicit relationship component and the explicit relationship component; The hybrid relationship network is constructed by the node set, the edge set and the hybrid relationship weight set.

2. The modeling method based on user behavior data according to claim 1 is characterized in that: The preference label data between users is used as implicit relationship components, including: The expression formula of the implicit relationship component is: ; Represents the preference label relationship value of the new user; and They represent the short-term relationship value of answering likes and the long-term relationship value of answering likes labels respectively; , and is the preset positive hyperparameter; Indicates user With users The preference label relationship value is calculated as follows: ; Indicates user and users The number of commonly preferred tags; Indicates user The number of preference tags.

3. The modeling method based on user behavior data according to claim 1, characterized in that: Determining a hybrid relationship matrix including implicit relationships and explicit relationships according to the hybrid relationship network includes: The initial mixed relationship matrix in the mixed relationship weight set is decomposed by using a robust principal component analysis method, and the optimal solution of the optimization problem is solved by using an augmented Lagrange multiplier method, a singular value decomposition and a soft threshold to obtain a low-rank relationship matrix and a sparse matrix; The low-rank relationship matrix is ​​used as the mixed relationship matrix.

4. The modeling method based on user behavior data according to any one of claims 1 to 3, characterized in that: Obtaining a feature extraction model based on the mixed relationship matrix and the multi-view data matrix, and obtaining user behavior features according to the feature extraction model, including: Inputting the mixed relationship matrix and the multi-view data matrix into a multi-view autoencoder, and training the multi-view autoencoder based on a loss function to obtain the feature extraction model; the multi-view graph autoencoder includes an encoder and a dual decoder; The user behavior feature is obtained based on the feature extraction model.

5. The modeling method based on user behavior data according to claim 1, characterized in that: A multi-perspective data matrix is ​​constructed based on the user behavior data and economic behavior data in the historical behavior data, including: Dividing the user behavior data into m user behavior subsets; Dividing the economic behavior data into m economic behavior subsets; The user behavior subset and the economic behavior subset are spliced ​​together to obtain A data matrix representing the time domain characteristics; Using Fourier transform to obtain frequency domain features from the historical behavior data, and obtaining a data matrix representing the frequency domain features; The multi-view data matrix is ​​obtained based on the data matrix representing the time domain characteristics and the data matrix representing the frequency domain characteristics.

6. A user behavior prediction method, characterized in that: include: Obtain the user behavior data to be predicted; The user behavior data is input into a user behavior prediction model obtained by the modeling method based on user behavior data according to any one of claims 1 to 5 to predict the user's future behavior, wherein the user behavior prediction model includes a feature extraction model and a prediction model.

7. A modeling device based on user behavior data, characterized in that: include: A hybrid relationship matrix acquisition module, used to construct a hybrid relationship network according to the historical behavior data of the user, and determine a hybrid relationship matrix containing implicit relationships and explicit relationships according to the hybrid relationship network; A multi-perspective data matrix acquisition module, used to construct a multi-perspective data matrix based on the user behavior data and economic behavior data in the historical behavior data; A user behavior feature acquisition module, used to obtain a feature extraction model based on the mixed relationship matrix and the multi-view data matrix, and obtain user behavior features according to the feature extraction model; A user behavior prediction model construction module, used to construct a user behavior prediction model based on the prediction model and the user behavior characteristics; The mixed relationship matrix acquisition module includes: A node set construction unit, used to construct a node set according to user data; An edge set construction unit, used to construct an edge set according to the behavioral relationship data between users; An implicit relationship component unit, used to take the preference label data between users as implicit relationship components; An explicit relationship component unit, used to take the user attention behavior data as an explicit relationship component; A mixed relationship weight set construction unit, used to obtain a mixed relationship weight set according to the implicit relationship component and the explicit relationship component; A hybrid relationship network construction unit is used to construct the hybrid relationship network from the node set, the edge set and the hybrid relationship weight set.

8. A user behavior prediction device, characterized in that: include: A data acquisition module, used to acquire user behavior data to be predicted; A prediction module is used to input the user behavior data into a user behavior prediction model obtained by the modeling method based on user behavior data as described in any one of claims 1 to 5 to predict the user's future behavior, wherein the user behavior prediction model includes a feature extraction model and a prediction model.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the user behavior data-based modeling method and / or the user behavior prediction method as described in any one of claims 1 to 6 when executing the computer program.

Images