User consumption behavior classification method, device, equipment and computer storage medium

Abstract

The present application discloses a method, device, equipment and computer storage medium for classifying user consumption behavior. The method for classifying user consumption behavior obtains user consumption behavior data to be classified; inputs the user consumption behavior data to be classified into a preset user consumption behavior classification model, and outputs a classification result; wherein the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine and a tree leaf node encoding. According to the embodiments of the present application, user consumption behavior can be classified more accurately.

Description

Technical Field

[0001] The present application belongs to the technical field of user consumption behavior classification, and in particular relates to a user consumption behavior classification method, device, electronic device and computer storage medium. Background Art

[0002] China Mobile user information is characterized by large amounts of data, and traditional methods such as random forests, decision trees, and support vector machines are often used to classify users. However, these methods have two problems: on the one hand, due to the lack of feature attributes and discrete features in mobile user data, data sparsity often occurs, making traditional classification algorithms unusable or causing poor prediction performance; on the other hand, in order to improve the classification effect, these traditional classification algorithms need to perform feature engineering on mobile user data, often requiring various feature combinations and feature crossover attempts to find the combination features suitable for the scenario. However, manual feature engineering is time-consuming and labor-intensive, often fails to achieve the expected results, and is costly.

[0003] Therefore, how to more accurately classify user consumption behavior is a technical problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0004] The embodiments of the present application provide a method, device, electronic device, and computer storage medium for classifying user consumption behavior, which can more accurately classify user consumption behavior.

[0005] In a first aspect, an embodiment of the present application provides a method for classifying user consumption behavior, including:

[0006] Obtain user consumption behavior data to be classified;

[0007] The user consumption behavior data to be classified is input into a preset user consumption behavior classification model, and the classification result is output; wherein, the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine, and tree leaf node encoding.

[0008] Optionally, before inputting the user consumption behavior data to be classified into a preset user consumption behavior classification model and outputting the classification result, the method further includes:

[0009] Obtain sample user consumption behavior data;

[0010] Arrange the sample user consumption behavior data according to a preset format to obtain a training set;

[0011] The training set is input into the fully connected neural network, factorization machine and tree leaf node encoding for model training to obtain the user consumption behavior classification model.

[0012] Optionally, the sample user consumption behavior data is arranged according to a preset format to obtain a training set, including:

[0013] Arrange the sample user consumption behavior data in a structured two-dimensional table format to obtain arranged sample user consumption behavior data;

[0014] The consumption behavior data of sample users after random sampling and arrangement are used to obtain the training set.

[0015] Optionally, the user consumption behavior data to be classified is input into a preset user consumption behavior classification model, and the classification results are output, including:

[0016] Using fully connected neural networks, factorization machines, and tree leaf node encoding, we determine the implicit high-order cross features, first-order and second-order explicit cross features, and explicit high-order cross features of the user consumption behavior data to be classified.

[0017] The classification results are output based on implicit high-order cross features, first-order and second-order explicit cross features, and explicit high-order cross features.

[0018] In a second aspect, an embodiment of the present application provides a user consumption behavior classification device, comprising:

[0019] The first acquisition module is used to obtain the user consumption behavior data to be classified;

[0020] The output module is used to input the user consumption behavior data to be classified into a preset user consumption behavior classification model and output the classification results; wherein, the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine and tree leaf node encoding.

[0021] Optionally, the device further comprises:

[0022] The second acquisition module is used to obtain consumption behavior data of sample users;

[0023] The arrangement module is used to arrange the sample user consumption behavior data according to a preset format to obtain a training set;

[0024] The model training module is used to input the training set into the fully connected neural network, factorization machine and tree leaf node encoding for model training to obtain the user consumption behavior classification model.

[0025] Optional, permutation module, including:

[0026] an arrangement unit, configured to arrange the sample user consumption behavior data in a structured two-dimensional table format to obtain arranged sample user consumption behavior data;

[0027] The random sampling unit is used to randomly sample the sample user consumption behavior data after arrangement to obtain a training set.

[0028] Optional output modules include:

[0029] A determination unit, configured to respectively determine the implicit high-order cross features, the first-order and second-order explicit cross features, and the explicit high-order cross features of the user consumption behavior data to be classified using a fully connected neural network, a factorization machine, and tree leaf node encoding;

[0030] The output unit is used to output the classification results based on the implicit high-order cross features, the first-order and second-order explicit cross features, and the explicit high-order cross features.

[0031] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising: a processor and a memory storing computer program instructions;

[0032] When the processor executes the computer program instructions, the user consumption behavior classification method shown in the first aspect is implemented.

[0033] In a fourth aspect, an embodiment of the present application provides a computer storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the user consumption behavior classification method shown in the first aspect is implemented.

[0034] The user consumption behavior classification method, device, electronic device and computer storage medium of the embodiment of the present application can classify user consumption behavior more accurately. The user consumption behavior classification method obtains user consumption behavior data to be classified; inputs the user consumption behavior data to be classified into a preset user consumption behavior classification model, and outputs a classification result; wherein the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine and a tree leaf node encoding. It can be seen that this method is based on a fully connected neural network as the basic architecture, uses tree leaf node encoding to extract explicit high-order cross features of mobile user data, and integrates a deep factor decomposition machine model to form a hybrid network structure, which can automatically extract multiple different types of cross features of mobile user data: implicit high-order cross features of neural networks, first-order and second-order explicit cross features of factor decomposition machines, and explicit high-order cross features based on gradient boosting tree classification rules. These can help the model fully learn the low-order and high-order feature information implicit in the user data, improve the model's representation ability for nonlinear classification scenarios, and thus can more accurately classify user consumption behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0036] Figure 1 This is a flowchart of a method for classifying user consumption behavior provided by an embodiment of the present application;

[0037] Figure 2 This is a flowchart of a method for classifying user consumption behaviors provided by another embodiment of the present application;

[0038] Figure 3 This is a flowchart of a method for classifying user consumption behaviors provided by another embodiment of the present application;

[0039] Figure 4 This is a schematic diagram of the structure of a user consumption behavior classification model provided by an embodiment of the present application;

[0040] Figure 5 This is a schematic diagram of the structure of a user consumption behavior classification device provided by an embodiment of the present application;

[0041] Figure 6 This is a structural diagram of a user consumption behavior classification device provided by another embodiment of the present application;

[0042] Figure 7 This is a structural diagram of an electronic device provided in yet another embodiment of the present application. DETAILED DESCRIPTION

[0043] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0044] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

[0045] As can be seen from the background technology section, traditional methods cannot accurately classify user consumption behavior. To address the problems of the existing technology, the embodiments of the present application provide a method, device, electronic device, and computer storage medium for classifying user consumption behavior. The following first introduces the user consumption behavior classification method provided in the embodiments of the present application.

[0046] Figure 1 FIG. 1 shows a flow chart of a method for classifying user consumption behaviors provided by an embodiment of the present application. Figure 1 As shown, the user consumption behavior classification method includes:

[0047] S101. Obtain user consumption behavior data to be classified.

[0048] S102. Input the user consumption behavior data to be classified into a preset user consumption behavior classification model and output the classification result; wherein the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine and tree leaf node encoding.

[0049] In one embodiment, before inputting the user consumption behavior data to be classified into a preset user consumption behavior classification model and outputting the classification result, the method further includes:

[0050] Obtain sample user consumption behavior data;

[0051] Arrange the sample user consumption behavior data according to a preset format to obtain a training set;

[0052] The training set is input into the fully connected neural network, factorization machine and tree leaf node encoding for model training to obtain the user consumption behavior classification model.

[0053] In one embodiment, the sample user consumption behavior data is arranged according to a preset format to obtain a training set, including:

[0054] Arrange the sample user consumption behavior data in a structured two-dimensional table format to obtain arranged sample user consumption behavior data;

[0055] The consumption behavior data of sample users after random sampling and arrangement are used to obtain the training set.

[0056] In one embodiment, the user consumption behavior data to be classified is input into a preset user consumption behavior classification model, and the classification results are output, including:

[0057] Using fully connected neural networks, factorization machines, and tree leaf node encoding, we determine the implicit high-order cross features, first-order and second-order explicit cross features, and explicit high-order cross features of the user consumption behavior data to be classified.

[0058] The classification results are output based on implicit high-order cross features, first-order and second-order explicit cross features, and explicit high-order cross features.

[0059] This user consumption behavior classification method obtains user consumption behavior data to be classified; inputs the user consumption behavior data to be classified into a preset user consumption behavior classification model, and outputs a classification result. The user consumption behavior classification model is trained using a training set and comprises a fully connected neural network, a factorization machine, and tree leaf node encoding. This method, based on a fully connected neural network architecture, utilizes tree leaf node encoding to extract explicit high-order cross features from mobile user data. This method then integrates a deep factorization machine model to form a hybrid network structure. This hybrid network structure automatically extracts multiple types of cross features from mobile user data: implicit high-order cross features from the neural network, explicit first- and second-order cross features from the factorization machine, and explicit high-order cross features based on gradient boosting tree classification rules. These features help the model fully learn the low- and high-order feature information implicit in the user data, improving the model's representation capabilities for nonlinear classification scenarios and enabling more accurate classification of user consumption behavior. Furthermore, by introducing an embedding vector approach, discrete high-dimensional sparse features are mapped into continuous, dense vectors in a low-dimensional space, reducing model computational overhead.

[0060] The above content is specifically described below using a scenario embodiment.

[0061] This embodiment provides a user consumption behavior classification model based on a deep factor decomposition machine. It uses the leaf node encoding information of the gradient enhanced decision tree and the deep factor decomposition machine model to learn the low-order and high-order feature information behind the user's original features from these three aspects: user basic information, call usage and traffic usage (detailed data fields are shown in Table 1), so as to classify users into high-value customers, low-value customers, traffic-type high-value customers and traffic-type low-value customers.

[0062] Table 1

[0063]

[0064]

[0065] An embodiment of the present invention discloses a user consumption classification model based on a deep factor decomposition machine, comprising: extracting explicit high-order cross features using leaf node encoding information of a gradient-enhanced decision tree; fusing a deep factor decomposition machine model to form a hybrid network structure to automatically extract multiple different types of cross features; using an embedding vector method to map discrete high-dimensional sparse features into continuous, dense and loud features in a low-dimensional space; and finally predicting user consumption categories through a prediction module.

[0066] From the above embodiments of the present invention, it can be seen that the present invention deeply mines the user features in the user category classification task, uses the leaf node encoding information of the gradient boosting decision tree to extract the explicit high-order cross features of the mobile user data, and integrates the deep factor decomposition machine model to form a hybrid network structure, which can automatically extract various types of cross features of the mobile user data: the implicit high-order cross features of the neural network, the first-order and second-order explicit cross features of the factor decomposition machine, and the explicit high-order cross features based on the gradient boosting tree classification rules. This can help the model fully learn the low-order and high-order feature information implicit in the data, enhance the model's ability to represent nonlinear classification scenarios, and thus improve the accuracy of the classification model.

[0067] like Figure 2 As shown, another user consumption behavior classification method provided by an embodiment of the present application includes:

[0068] S201. Establish a deep factorization machine prediction model based on tree leaf node encoding.

[0069] Taking the original features of the user as input, the leaf nodes in the deep factor decomposition machine prediction model based on the tree leaf node encoding are the leaf nodes of the gradient boosting decision tree as the mined user features.

[0070] S202: Obtain a training set based on user data, and use the training set to perform prediction training on the deep factorization machine prediction model to obtain a trained prediction model.

[0071] S203: Use the trained prediction model to perform prediction and obtain output data.

[0072] This method uses the user's original features as input and builds a deep factorization machine prediction model based on tree leaf node encoding. The tree leaf nodes in the deep factorization machine prediction model based on tree leaf node encoding are the leaf nodes of the gradient boosting decision tree, which are used to mine user features. A training set is obtained based on user data, and the deep factorization machine network prediction model is trained with the training set to predict user categories, obtaining a trained prediction model. The user category is predicted based on the trained prediction model. This effectively improves the prediction performance of the DeepFM network and increases the accuracy of predicting user consumption types.

[0073] like Figure 3 As shown, another method for classifying user consumption behavior provided by an embodiment of the present application includes:

[0074] S301. Establish a deep factorization machine prediction model.

[0075] Specifically, the user's original features are used as input to establish a deep factorization machine prediction model based on tree leaf node encoding, which includes a fully connected neural network part, a factorization machine part and a tree leaf node encoding part.

[0076] The fully connected neural network part and the factorization machine part share the embedding vector, and the embedding vector is updated simultaneously during backpropagation.

[0077] The fully connected neural network part is mainly composed of three layers of fully connected layers. For each layer, tanh is used as the activation function. Specifically, the network parameters are the connection weights W of each layer. l FCN and bias b l FCN , where W l FCN and b l FCN is known and remains fixed after modeling is completed. The input data is passed through the embedding layer to obtain the embedding vectors corresponding to different feature attributes. In the forward propagation process, the embedding vector is used as the input of the input layer, and the output of each neuron in each layer after the activation function σ(·) is calculated. Finally, the output vector of the last layer L of the fully connected neural network part can be obtained.

[0078]

[0079] The factorization machine part consists of a second-order cross layer, which shares the feature embedding vector with the fully connected neural network part, and combines each embedding vector into an inner product to obtain a second-order combined feature vector. Specifically, the network parameters are the weights w of the linear combination FM and bias b FM Among them, w FM and b FM is known and is fixed after the model is completed. In the forward propagation process, the embedding vector is used as the input of the input layer to calculate the output of the second-order cross layer.

[0080]

[0081] in, is the shared embedding vector, and It can be calculated in advance at each iteration and persisted to the memory variable. It is called directly during the process to speed up the calculation.

[0082] The leaf node encoding component consists of a gradient boosted decision tree and an embedding layer. In each iteration, the gradient boosted decision tree fits and constructs a new CART tree based on the negative gradient of the loss function. The number of iterations determines the number of trees. Furthermore, each tree forms leaf nodes after a split. Each sample falls on a leaf node in each tree. The location of the sample on each leaf node is recorded, and the encoded information based on the leaf nodes is extracted as more representative features of the mined user.

[0083] Use the gradient boosting decision tree for pre-training to obtain the GBDT pre-training model T M (x):

[0084]

[0085] Among them, T M-1 (x) is the gradient boosting decision tree generated in the (M-1)th round of iteration, g Mk is the optimal value of the kth leaf node of the Mth tree, K is the number of leaf nodes in the current round of GBDT, I(x∈C Mk ) is the leaf node region to which the sample x belongs. Perform supervised learning on the encoding representation of the sample features and obtain the encoding feature δ of the leaf node of each sample x :

[0086] Tree Leaves Encoding: x→δ x

[0087] Then, the encoded feature δ x As the input of the deep factorization machine embedding layer, the embedding vector of each tree is obtained, and the obtained embedding vectors are concatenated as the output vector of the encoding part of the tree sub-node

[0088]

[0089] Among them, e m Represents the output vector produced by the m-th tree.

[0090] Compared with the existing deep factor decomposition machine, which can only extract explicit second-order cross features and implicit high-order cross features, the gradient-enhanced decision tree leaf node encoding is introduced through pre-training and input into the model in the form of entity embedding. It can automatically learn and extract explicit high-order cross features of user attributes, enriching the types of cross feature learning of the deep factor decomposition machine, thereby further improving the prediction accuracy of the model.

[0091] S302: Obtain a training set based on the input user feature data.

[0092] For example, the user's raw data can be arranged in a structured, two-dimensional table, with each row representing a sample and each column representing the sample's characteristic attributes. This data can then be shuffled and rearranged to create a training set. The samples are real data, for example, user data from a consumption type classification scenario, including detailed features such as basic user information, phone bill usage, and data usage, as input data for the training set.

[0093] S303: Use the training set as input data for the fully connected neural network part, the factorization machine part, and the gradient boosting decision tree part, obtain the embedding vectors of the fully connected neural network part and the factorization machine part, and obtain the leaf node position information of the gradient boosting decision tree part as the mined user features.

[0094] Exemplarily, the update formula of the splicing vector includes:

[0095]

[0096] in, is the output vector of the fully connected neural network part, is the output vector of the factorization machine part, It is the output vector of the tree subnode part.

[0097] S304: Predict the user category based on the trained prediction model.

[0098] Specifically, it may include: using the selected user data as prediction input data of the data prediction formula, and obtaining output data according to the data prediction formula. The data prediction formula may be:

[0099]

[0100] in, is the output weight matrix formula, is the output layer bias, and σ(·) is the sigmoid activation function.

[0101] like Figure 4 As shown, a structural diagram of a user consumption behavior classification model provided by an embodiment of the present application is shown. Figure 4 401: Fully connected neural network part; 402: The fully connected neural network part and the factorization machine part share the embedding vector; 403: Factorization machine part; 404: Pre-training using gradient boosting decision tree; 405: Tree leaf node encoding part; 406: Splicing and predicting the output vectors of each part.

[0102] Figure 5 This is a schematic diagram of the structure of a user consumption behavior classification device provided by an embodiment of the present application. Figure 5 As shown, the user consumption behavior classification device includes:

[0103] The first acquisition module 501 is used to acquire the user consumption behavior data to be classified;

[0104] Output module 502 is used to input the user consumption behavior data to be classified into a preset user consumption behavior classification model and output the classification results; wherein, the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine and tree leaf node encoding.

[0105] In one embodiment, the apparatus further comprises:

[0106] The second acquisition module is used to obtain consumption behavior data of sample users;

[0107] The arrangement module is used to arrange the sample user consumption behavior data according to a preset format to obtain a training set;

[0108] The model training module is used to input the training set into the fully connected neural network, factorization machine and tree leaf node encoding for model training to obtain the user consumption behavior classification model.

[0109] In one embodiment, the arrangement module includes:

[0110] an arrangement unit, configured to arrange the sample user consumption behavior data in a structured two-dimensional table format to obtain arranged sample user consumption behavior data;

[0111] The random sampling unit is used to randomly sample the sample user consumption behavior data after arrangement to obtain a training set.

[0112] In one embodiment, the output module 502 includes:

[0113] A determination unit, configured to respectively determine the implicit high-order cross features, the first-order and second-order explicit cross features, and the explicit high-order cross features of the user consumption behavior data to be classified using a fully connected neural network, a factorization machine, and tree leaf node encoding;

[0114] The output unit is used to output the classification results based on the implicit high-order cross features, the first-order and second-order explicit cross features, and the explicit high-order cross features.

[0115] Figure 5 Each module / unit in the device shown has the function of realizing Figure 1 The functions of each step in the embodiment can achieve the corresponding technical effects, which will not be described in detail here for the sake of brevity.

[0116] Figure 6 This is a structural diagram of a user consumption behavior classification device provided by another embodiment of the present application. Figure 6 As shown, the user consumption behavior classification device includes a modeling module 601, a training module 602 and a prediction module 603. The internal structure and connection relationship of the device are further described below in conjunction with its working principle.

[0117] A modeling module 601 is used to establish a deep factorization machine prediction model based on tree leaf node encoding, where the tree leaf nodes of the prediction model are leaf nodes of a gradient boosting decision tree;

[0118] A training module 602 is configured to obtain a training set based on user data, and use the training set to perform user category prediction training on a deep factorization machine prediction model based on tree leaf node encoding to obtain a trained prediction model;

[0119] The prediction module 603 is used to make predictions based on the trained prediction model and classify users into high-value users, low-value users, traffic-type high-value users and traffic-type low-value users.

[0120] Figure 7 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown.

[0121] The electronic device may include a processor 701 and a memory 702 storing computer program instructions.

[0122] Specifically, the processor 701 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0123] The memory 702 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 702 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 702 may include removable or non-removable (or fixed) media. Where appropriate, the memory 702 may be inside or outside the electronic device. In certain embodiments, the memory 702 may be a non-volatile solid-state memory.

[0124] In one example, the memory 702 may be a read-only memory (ROM). In one example, the ROM may be a mask-programmable ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or a flash memory, or a combination of two or more of these.

[0125] The processor 701 implements any one of the user consumption behavior classification methods in the above embodiments by reading and executing computer program instructions stored in the memory 702 .

[0126] In one example, the electronic device may further include a communication interface 703 and a bus 710. Figure 7 As shown, the processor 701, the memory 702, and the communication interface 703 are connected via a bus 710 and communicate with each other.

[0127] The communication interface 703 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0128] Bus 710 includes hardware, software or both, and the parts of online data flow metering equipment are coupled to each other. For example, but not limitation, bus can include accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 710 can include one or more buses. Although the present application embodiment describes and shows specific bus, the application considers any suitable bus or interconnection.

[0129] In addition, the embodiments of the present application may be implemented by providing a computer storage medium having computer program instructions stored thereon; when the computer program instructions are executed by a processor, any one of the user consumption behavior classification methods in the above embodiments is implemented.

[0130] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0131] The functional modules shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in unit, a function card or the like. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium that can store or transmit information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0132] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0133] Aspects of the present application have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed via the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. This processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It is also understood that each box in the block diagram and / or the flowchart and the combination of the boxes in the block diagram and / or the flowchart can also be implemented by the dedicated hardware that performs the specified function or action, or can be implemented by the combination of dedicated hardware and computer instructions.

[0134] The above description is only a specific embodiment of the present application. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the scope of protection of the present application.

Claims

1. A method for classifying user consumption behavior, characterized in that: include: Obtain the user consumption behavior data to be classified; Input the user consumption behavior data to be classified into a preset user consumption behavior classification model, and output the classification result; wherein the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine, and a tree leaf node encoding, and the tree leaf node encoding includes a gradient boosting decision tree and an embedding layer; The user consumption behavior classification model uses the fully connected neural network as a basic architecture, uses the tree leaf node encoding to extract explicit high-order cross features of user consumption behavior data, integrates the factor decomposition machine to form a hybrid network structure, and extracts multiple different types of cross features of user consumption behavior data, wherein the cross features include implicit high-order cross features, first-order and second-order explicit cross features, and explicit high-order cross features; wherein the factor decomposition machine is composed of a second-order cross layer, and the fully connected neural network and the factor decomposition machine share an embedding vector, which is simultaneously updated in back propagation, and each embedding vector is combined in pairs to perform an inner product to obtain a second-order combined feature vector; the user consumption behavior data to be classified is input into a preset user consumption behavior classification model, and a classification result is output, including: Extracting implicit high-order cross features using the fully connected neural network; extracting first-order and second-order explicit cross-features using the factorization machine; Extracting explicit high-order cross-features using the tree leaf node encoding; Outputting the classification result based on the implicit high-order cross features, the first-order and second-order explicit cross features, and the explicit high-order cross features; The extracting of explicit high-order cross features by encoding the tree leaf nodes includes: Use the gradient boosted decision tree for pre-training to obtain the GBDT pre-training model T M (x): ； Among them, T M-1 (x) is the gradient boosted decision tree generated in the (M-1)th round of iteration, is the optimal value of the kth leaf node of the Mth tree, K is the number of leaf nodes in the current round of GBDT, is the leaf node to which the sample x belongs; Get the encoding features of the leaf nodes of each sample; Using the encoded features as the input of the embedding layer of the factor decomposition machine to obtain the embedding vector of each tree, and concatenating the obtained embedding vectors to obtain the output vector of the tree subnode encoding; The outputting the classification result based on the implicit high-order cross features, the first-order and second-order explicit cross features, and the explicit high-order cross features includes: The implicit high-order cross features, the first-order and second-order explicit cross features, and the displayed high-order cross features are concatenated to output the classification result.

2. The user consumption behavior classification method according to claim 1 is characterized in that: Before inputting the user consumption behavior data to be classified into a preset user consumption behavior classification model and outputting the classification result, the method further includes: Obtain sample user consumption behavior data; Arranging the sample user consumption behavior data according to a preset format to obtain the training set; The training set is input into the fully connected neural network, factor decomposition machine and tree leaf node encoding for model training to obtain the user consumption behavior classification model.

3. The user consumption behavior classification method according to claim 2 is characterized in that: The step of arranging the sample user consumption behavior data according to a preset format to obtain the training set includes: Arranging the sample user consumption behavior data in a structured two-dimensional table format to obtain arranged sample user consumption behavior data; The arranged sample user consumption behavior data is randomly sampled to obtain the training set.

4. A user consumption behavior classification device, characterized in that: include: The first acquisition module is used to acquire the user consumption behavior data to be classified; An output module is used to input the user consumption behavior data to be classified into a preset user consumption behavior classification model, and output a classification result; wherein the user consumption behavior classification model is a model obtained by model training using a training set, and the user consumption behavior classification model includes a fully connected neural network, a factor decomposition machine, and a tree leaf node encoding, and the tree leaf node encoding includes a gradient boosting decision tree and an embedding layer, wherein the factor decomposition machine is composed of a second-order cross layer, and the fully connected neural network and the factor decomposition machine share an embedding vector, and the embedding vector is simultaneously updated in back propagation, and each embedding vector is combined to perform an inner product to obtain a second-order combined feature vector; The user consumption behavior classification model uses the fully connected neural network as a basic architecture, uses the tree leaf node encoding to extract explicit high-order cross features of user consumption behavior data, integrates the factor decomposition mechanism to form a hybrid network structure, and extracts multiple different types of cross features of user consumption behavior data, wherein the cross features include implicit high-order cross features, first-order and second-order explicit cross features, and explicit high-order cross features; The output module comprises: A determination unit, configured to extract implicit high-order cross features using the fully connected neural network; extract first-order and second-order explicit cross features using the factorization machine; and extract explicit high-order cross features using the tree leaf node encoding; An output unit, configured to output the classification result based on the implicit high-order cross features, the first-order and second-order explicit cross features, and the explicit high-order cross features; The determining unit is specifically used for: Use the gradient boosted decision tree for pre-training to obtain the GBDT pre-training model T M (x): ； Among them, T M-1 (x) is the gradient boosted decision tree generated in the (M-1)th round of iteration, is the optimal value of the kth leaf node of the Mth tree, K is the number of leaf nodes in the current round of GBDT, is the leaf node to which the sample x belongs; Get the encoding features of the leaf nodes of each sample; Using the encoded features as the input of the embedding layer of the factor decomposition machine to obtain the embedding vector of each tree, and concatenating the obtained embedding vectors to obtain the output vector of the tree subnode encoding; The output unit is specifically used for: The implicit high-order cross features, the first-order and second-order explicit cross features, and the displayed high-order cross features are concatenated to output the classification result.

5. The user consumption behavior classification device according to claim 4, characterized in that: The device also includes: The second acquisition module is used to obtain consumption behavior data of sample users; An arrangement module, used to arrange the sample user consumption behavior data according to a preset form to obtain the training set; The model training module is used to input the training set into the fully connected neural network, factor decomposition machine and tree leaf node encoding for model training to obtain the user consumption behavior classification model.

6. The user consumption behavior classification device according to claim 5, characterized in that: The arrangement module comprises: An arranging unit, used to arrange the sample user consumption behavior data according to a structured two-dimensional table format to obtain arranged sample user consumption behavior data; The random sampling unit is used to randomly sample the arranged sample user consumption behavior data to obtain the training set.

7. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the user consumption behavior classification method as described in any one of claims 1-3 is implemented.

8. A computer storage medium, characterized in that: The computer storage medium stores computer program instructions, and when the computer program instructions are executed by the processor, the user consumption behavior classification method according to any one of claims 1 to 3 is implemented.

Images