Client portrait construction method and device, storage medium and electronic equipment

Abstract

The invention discloses a customer portrait construction method and device, a storage medium and electronic equipment, and relates to the field of financial science and technology. The method comprises the following steps: acquiring target customer data, wherein the target customer data comprises customer static information data, customer dynamic information data and article information data; word embedding processing is carried out on the target customer data to obtain a first word vector and a second word vector, the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the article information data; performing fusion processing on the first word vector and the second word vector through a target model to obtain a customer fusion feature vector; and determining a customer fusion feature tag according to the customer fusion feature vector, and constructing a customer portrait according to the customer fusion feature tag. According to the method and the device, the technical problem that the accuracy of the constructed customer portrait is relatively low due to the fact that discrete labels are adopted to represent customer portrait information in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of financial technology, and in particular, to a method, device, storage medium and electronic device for constructing a customer portrait. Background Art

[0002] At present, financial institutions usually recommend products to customers based on customer portraits. In the process of building customer portraits, they usually obtain multiple labels of customer portraits based on customer information, and regard the labels as discrete attribute values. They make targeted customer recommendations and analyses by measuring similarity. However, in actual situations, the label information in the portrait is correlated, and the discrete label representation method will miss a lot of information and cannot express the correlation between labels, resulting in low accuracy of the constructed customer portrait.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present invention provide a method, device, storage medium and electronic device for constructing a customer portrait, so as to at least solve the technical problem that discrete tags are used in the prior art to represent customer portrait information, resulting in low accuracy of the constructed customer portrait.

[0005] According to one aspect of an embodiment of the present invention, a method for constructing a customer portrait is provided, comprising: obtaining target customer data, wherein the target customer data includes customer static information data, customer dynamic information data and item information data, and the item information data is information data of items related to behavioral data in the customer dynamic information data; performing word embedding processing on the target customer data to obtain a first word vector and a second word vector, wherein the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the item information data; performing fusion processing on the first word vector and the second word vector through a target model to obtain a customer fusion feature vector, wherein the target model is obtained by training a recurrent neural network with sample data, and the customer fusion feature vector is used to characterize the fusion features of the customer static information data, the customer dynamic information data and the item information data; determining a customer fusion feature label based on the customer fusion feature vector, and constructing a customer portrait based on the customer fusion feature label.

[0006] Furthermore, obtaining target customer data includes: obtaining original customer data, wherein the original customer data includes customer basic data, customer behavior data, and item data of items related to the customer behavior data; encoding the customer basic data to obtain a customer static feature vector, and encoding the item data to obtain an item feature vector; extracting features from the customer behavior data through a pre-trained feature processing model to obtain customer dynamic features, and encoding the customer dynamic features to obtain a customer dynamic feature vector; generating target customer data based on the customer static feature vector, the customer dynamic feature vector, and the item feature vector.

[0007] Furthermore, word embedding processing is performed on the target customer data to obtain a first word vector and a second word vector, including: obtaining a word embedding matrix of a customer static feature vector, a word embedding matrix of a customer dynamic feature vector, and a word embedding matrix of an item feature vector; multiplying the customer static feature vector by the word embedding matrix of the customer static feature vector to obtain a first word vector; multiplying the item feature vector by the word embedding matrix of the item feature vector to obtain an item feature word vector after word embedding; processing the customer dynamic feature vector and the word embedding matrix of the customer dynamic feature vector through a pre-trained feature processing model to obtain a dynamic feature word vector after word embedding; and performing information fusion on the item feature word vector after word embedding and the dynamic feature word vector after word embedding to obtain a second word vector.

[0008] Furthermore, the first word vector and the second word vector are fused through the target model to obtain a customer fused feature vector, including: inputting the first word vector and the second word vector into the target model; performing vector concatenation and fusion on the first word vector and the second word vector through the target model, and outputting the customer fused feature vector.

[0009] Furthermore, determining a customer fusion feature label based on the customer fusion feature vector includes: obtaining label feature vectors of multiple preset portrait labels; performing similarity calculations on each label feature vector and the customer fusion feature vector to obtain multiple calculation results, wherein the calculation results are used to characterize the degree of similarity between the label feature and the customer fusion feature; determining a target portrait label from multiple preset portrait labels based on the multiple calculation results, and determining a customer fusion feature label based on the target portrait label.

[0010] Furthermore, the target model is generated through the following steps: obtaining sample data, and dividing the sample data into a training set and a test set, wherein the sample data is labeled customer data; constructing an initial model based on a recurrent neural network, and training the initial model according to the training set to obtain a trained model; optimizing the trained model according to the test set and preset evaluation criteria to obtain a target model.

[0011] Furthermore, after constructing a customer profile based on the customer fusion feature tags, the method also includes: obtaining item information of multiple items to be recommended; determining a target item to be recommended from the multiple items to be recommended based on the customer profile and the item information of the multiple items to be recommended, and pushing the item information of the target item to be recommended to the target customer corresponding to the customer profile.

[0012] According to another aspect of an embodiment of the present invention, there is also provided a device for constructing a customer portrait, including: an acquisition module, used to acquire target customer data, wherein the target customer data includes customer static information data, customer dynamic information data and item information data, and the item information data is information data of items related to the behavioral data in the customer dynamic information data; a first processing module, used to perform word embedding processing on the target customer data to obtain a first word vector and a second word vector, wherein the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the item information data; a second processing module, used to perform fusion processing on the first word vector and the second word vector through a target model to obtain a customer fusion feature vector, wherein the target model is obtained by training a recurrent neural network with sample data, and the customer fusion feature vector is used to characterize the fusion features of the customer static information data, the customer dynamic information data and the item information data; a construction module, used to determine a customer fusion feature label based on the customer fusion feature vector, and construct a customer portrait based on the customer fusion feature label.

[0013] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned method for constructing a customer profile when running.

[0014] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, which includes one or more processors; a memory for storing one or more programs, so that when the one or more programs are executed by the one or more processors, the one or more processors are implemented to run the programs, wherein the programs are configured to execute the above-mentioned method for constructing customer portraits when running.

[0015] In an embodiment of the present invention, a method of comprehensively considering customer behavior and information about products related to the behavior is adopted. First, target customer data is acquired, and then word embedding processing is performed on the target customer data to obtain a first word vector and a second word vector. Then, the first word vector and the second word vector are fused through a target model to obtain a customer fusion feature vector, and then a customer fusion feature label is determined based on the customer fusion feature vector, and a customer profile is constructed based on the customer fusion feature label. Among them, the target customer data includes customer static information data, customer dynamic information data, and item information data. The item information data is information data of items related to the behavior data in the customer dynamic information data. The first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the item information data. The target model is obtained by training a recurrent neural network with sample data, and the customer fusion feature vector is used to characterize the fusion characteristics of the customer static information data, the customer dynamic information data, and the item information data.

[0016] In the above process, by acquiring the target customer data, a data basis is provided for the subsequent construction of the customer portrait, and the customer behavior and the information of the products related to the behavior are comprehensively considered, which can improve the accuracy and comprehensiveness of the construction of the customer portrait; by performing word embedding processing on the target customer data, the first word vector and the second word vector are obtained, and the use of word embedding technology is realized. By using continuous vectors to represent independent labels of different dimensions, the correlation between labels can be better captured, and the problem that discrete coding cannot express the correlation between labels is solved, and customer characteristics can be divided more accurately; the first word vector and the second word vector are fused through the target model to obtain the customer fusion feature vector, and the fusion of the basic characteristics of the customer, the behavior characteristics of the customer and the characteristics of the item by the deep neural network is realized, so as to better determine the customer's preference for the item; the customer fusion feature label can be determined based on the customer fusion feature vector, and the customer portrait can be constructed based on the customer fusion feature label. The label representation is richer and more accurate, which improves the accuracy of building the customer portrait.

[0017] It can be seen that through the technical solution of the present invention, the purpose of better constructing customer portraits and achieving accurate recommendations is achieved, thereby achieving the technical effect of improving the accuracy of constructing customer portraits, and further solving the technical problem in the prior art of using discrete labels to represent customer portrait information, resulting in low accuracy of the constructed customer portraits. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 is a flowchart of an optional method for constructing a customer portrait according to an embodiment of the present invention;

[0020] Figure 2 is a schematic diagram of an optional process of building a customer portrait according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of an optional device for constructing a customer portrait according to an embodiment of the present invention;

[0022] Figure 4 is a schematic diagram of an optional electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set between the system and the relevant user or organization. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0026] Example 1

[0027] According to an embodiment of the present invention, an embodiment of a method for constructing a customer portrait is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] Figure 1 is a flowchart of an optional method for constructing a customer portrait according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0029] Step S101, obtaining target customer data, wherein the target customer data includes customer static information data, customer dynamic information data and item information data, and the item information data is information data of items related to the behavior data in the customer dynamic information data.

[0030] Optionally, the target customer data may be obtained through an application system, a processor, an electronic device, and other devices, for example, the target customer data may be obtained through a customer portrait construction system. Optionally, the customer static information may be basic customer information such as gender, age, and occupation, and the customer static information data may be a static feature vector representing the customer static information. The customer dynamic information may be customer behavior information such as browsing records, click records, purchase records, and rating records, and the customer dynamic information data may be a dynamic feature vector representing the customer dynamic information. The item information may be product information of financial products related to the customer's browsing, clicking, purchasing, and rating behaviors, and the item information data may be an item feature vector representing the item information.

[0031] Step S102, perform word embedding processing on the target customer data to obtain a first word vector and a second word vector, wherein the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and item information data.

[0032] Optionally, the target customer data may be processed by word embedding to obtain a first word vector and a second word vector. For example, discrete labels may be mapped to a continuous vector space using word embedding technology, so that labels with similar semantics or relevance are closer in the vector space, and the correlation between labels is better captured, making the representation of labels richer and more accurate. Optionally, the first word vector may be obtained by word embedding the customer static information data, and the second word vector may be obtained by word embedding the customer dynamic information data and the item information data, respectively, and then fusing them.

[0033] Step S103, the first word vector and the second word vector are fused through the target model to obtain a customer fusion feature vector, wherein the target model is obtained by training a recurrent neural network with sample data, and the customer fusion feature vector is used to characterize the fusion features of customer static information data, customer dynamic information data and item information data.

[0034] Optionally, the target model is a pre-trained recurrent neural network (RNN) model, which takes the customer's behavior data as input, and uses the recurrent neural network to learn the representation of the customer's behavior. After the forward propagation process of the neural network, the hidden layer representation of the customer's behavior can be obtained. Optionally, the first word vector and the second word vector are fused through the target model to obtain a customer fusion feature vector, which realizes the fusion of customer static features, dynamic features and item features, and can comprehensively express the customer's characteristics and behavior preferences.

[0035] Step S104, determining the customer fusion feature label based on the customer fusion feature vector, and constructing the customer portrait based on the customer fusion feature label.

[0036] Optionally, a customer fusion feature label can be determined based on the customer fusion feature vector, and a customer portrait can be constructed based on the customer fusion feature label. For example, the features are mapped to the dimensions of the customer portrait through the fully connected layer or softmax layer of the model, so that the customer fusion feature label is determined, so that the customer portrait can be constructed based on the customer fusion feature label. Optionally, the customer portrait may include dimensions such as the customer's interest preferences, product preferences, and values.

[0037] Based on the scheme defined in the above steps S101 to S104, it can be known that in an embodiment of the present invention, a method of comprehensively considering customer behavior and information about products related to the behavior is adopted, firstly, target customer data is obtained, then word embedding processing is performed on the target customer data to obtain a first word vector and a second word vector, then the first word vector and the second word vector are fused through a target model to obtain a customer fusion feature vector, then the customer fusion feature label is determined based on the customer fusion feature vector, and a customer profile is constructed based on the customer fusion feature label. Among them, the target customer data includes customer static information data, customer dynamic information data, and item information data, the item information data is information data of items related to the behavior data in the customer dynamic information data, the first word vector corresponds to the customer static information data, the second word vector corresponds to the customer dynamic information data and the item information data, the target model is obtained by training a recurrent neural network with sample data, and the customer fusion feature vector is used to characterize the fusion characteristics of customer static information data, customer dynamic information data, and item information data.

[0038] It is easy to notice that in the above process, by acquiring the target customer data, a data basis is provided for the subsequent construction of the customer portrait, and the customer behavior and the information of the products related to the behavior are comprehensively considered, which can improve the accuracy and comprehensiveness of the construction of the customer portrait; by performing word embedding processing on the target customer data, the first word vector and the second word vector are obtained, and the word embedding technology is adopted. By representing independent labels of different dimensions through continuous vectors, the correlation between labels can be better captured, and the problem that discrete coding cannot express the correlation between labels is solved, and customer characteristics can be divided more accurately; the first word vector and the second word vector are fused through the target model to obtain the customer fusion feature vector, and the fusion of the basic characteristics of the customer, the behavior characteristics of the customer and the characteristics of the item by the deep neural network is realized, so as to better determine the customer's preference for the item; the customer fusion feature label can be determined based on the customer fusion feature vector, and the customer portrait can be constructed based on the customer fusion feature label. The label representation is richer and more accurate, which improves the accuracy of building the customer portrait.

[0039] It can be seen that through the technical solution of the present invention, the purpose of better constructing customer portraits and achieving accurate recommendations is achieved, thereby achieving the technical effect of improving the accuracy of constructing customer portraits, and further solving the technical problem in the prior art of using discrete labels to represent customer portrait information, resulting in low accuracy of the constructed customer portraits.

[0040] In an optional embodiment, obtaining target customer data includes: obtaining original customer data, wherein the original customer data includes customer basic data, customer behavior data, and item data of items related to the customer behavior data; encoding the customer basic data to obtain a customer static feature vector, and encoding the item data to obtain an item feature vector; extracting features from the customer behavior data using a pre-trained feature processing model to obtain customer dynamic features, and encoding the customer dynamic features to obtain a customer dynamic feature vector; and generating target customer data based on the customer static feature vector, the customer dynamic feature vector, and the item feature vector.

[0041] Optionally, in the process of acquiring target customer data, the original customer data is first acquired, for example, the customer's basic data, behavior data, and item information data related to the behavior. The behavior data may include the customer's click records, purchase records, rating records, etc., and the item information data may include the item's attributes, labels, descriptions, etc.

[0042] Optionally, after obtaining the original customer data, the original customer data is preprocessed, including data cleaning, feature extraction, and standardization. Optionally, the customer basic data is encoded to obtain the customer static feature vector, and the item data is encoded to obtain the item feature vector. Optionally, the features of different dimensions are discretized respectively, and converted into sparse vector representations using one-hot encoding, and a lookup table (also called an embedding matrix) is initialized for each feature dimension, in which each eigenvalue is associated with a vector. For example, the customer basic data is one-hot encoded to obtain the customer static feature vector, and the corresponding embedding matrix is ​​randomly initialized to obtain the word embedding matrix of the customer static feature vector; the item data is one-hot encoded to obtain the item feature vector, and the corresponding embedding matrix is ​​randomly initialized to obtain the word embedding matrix of the item feature vector.

[0043] Optionally, the customer behavior data is subjected to feature extraction through a pre-trained feature processing model to obtain customer dynamic features, and the customer dynamic features are encoded to obtain a customer dynamic feature vector. The pre-trained feature processing model may be a pre-trained deep learning model, which is used to extract features from the customer behavior data to obtain customer dynamic features, and one-hot encodes the customer dynamic features to obtain a customer dynamic feature vector, and at the same time, the corresponding embedding matrix is ​​randomly initialized to obtain a word embedding matrix of the customer dynamic feature vector.

[0044] Optionally, after the above-mentioned process of preprocessing the original customer data, target customer data may be generated according to the customer static feature vector, the customer dynamic feature vector and the item feature vector, that is, the target customer data is obtained.

[0045] It should be noted that in the above process, the customer's basic data, behavioral data and item information data related to the behavior are comprehensively considered, which can effectively solve the problems of insufficient dynamic information mining of customer behavior and polarization of customer preferences for items, thereby building a more accurate and comprehensive customer portrait and better understanding the customer's preferences and needs for different products.

[0046] In an optional embodiment, word embedding processing is performed on the target customer data to obtain a first word vector and a second word vector, including: obtaining a word embedding matrix of a customer static feature vector, a word embedding matrix of a customer dynamic feature vector, and a word embedding matrix of an item feature vector; multiplying the customer static feature vector by the word embedding matrix of the customer static feature vector to obtain a first word vector; multiplying the item feature vector by the word embedding matrix of the item feature vector to obtain an item feature word vector after word embedding; processing the customer dynamic feature vector and the word embedding matrix of the customer dynamic feature vector through a pre-trained feature processing model to obtain a dynamic feature word vector after word embedding; and performing information fusion on the item feature word vector after word embedding and the dynamic feature word vector after word embedding to obtain a second word vector.

[0047] Optionally, in the process of performing word embedding processing on the target customer data to obtain the first word vector and the second word vector, the first word vector and the second word vector can be calculated by obtaining the word embedding matrix of the customer's static feature vector, the word embedding matrix of the customer's dynamic feature vector, and the word embedding matrix of the item feature vector.

[0048] Optionally, the customer static feature vector is multiplied by the word embedding matrix of the customer static feature vector to calculate the first word vector; the item feature vector is multiplied by the word embedding matrix of the item feature vector to calculate the item feature word vector after word embedding. For example, after one-hot encoding the customer basic data, the encoded customer static feature vector is multiplied by the word embedding matrix of the customer static feature vector to obtain the corresponding static information embedding representation, that is, the first word vector; after one-hot encoding the item data, the encoded item feature vector is multiplied by the word embedding matrix of the item feature vector to obtain the corresponding item information embedding representation: item-embedding, that is, the item feature word vector after word embedding.

[0049] Optionally, the customer dynamic feature vector and the word embedding matrix of the customer dynamic feature vector are processed by a pre-trained feature processing model to obtain the dynamic feature word vector after word embedding, and then the item feature word vector after word embedding and the dynamic feature word vector after word embedding are information fused to obtain the second word vector. For example, the embedding of the customer behavior is first calculated, and then the customer behavior embedding and item-embedding are information fused to obtain the customer's dynamic behavior embedding. Specifically, after one-hot encoding the customer dynamic features, the encoded customer dynamic feature vector and the word embedding matrix of the customer dynamic feature vector are input into the pre-trained feature processing model (i.e., the deep learning model), and the embedding representation of the customer behavior is calculated by forward propagation, that is, the dynamic feature word vector after word embedding is obtained, and then the embedding of the customer behavior (i.e., the dynamic feature word vector after word embedding) and the corresponding item-embedding (i.e., the item feature word vector after word embedding) are information fused, for example, using splicing, weighting, and other operations to merge them into a comprehensive dynamic behavior embedding (i.e., the second word vector).

[0050] It should be noted that in the above process, by using word embedding technology to map discrete labels into a continuous vector space, labels with similar semantics or correlations are closer in the vector space, which can better capture the correlation between labels and accurately express the strong correlation between them, making the label representation richer and more accurate, solving the problem that discrete coding cannot express the correlation between labels and improving the accuracy of building customer portraits.

[0051] In an optional embodiment, the first word vector and the second word vector are fused through a target model to obtain a customer fused feature vector, including: inputting the first word vector and the second word vector into the target model; performing vector concatenation and fusion on the first word vector and the second word vector through the target model, and outputting a customer fused feature vector.

[0052] Optionally, in the process of fusing the first word vector and the second word vector through the target model to obtain the customer fused feature vector, the first word vector and the second word vector are first input into the target model, and then the first word vector and the second word vector are vector-concatenated and fused through the target model to output the customer fused feature vector. For example, after calculating the customer's static information embedding (i.e., the first word vector) and the customer's dynamic behavior embedding (i.e., the second word vector), the customer's static information embedding and the customer's dynamic behavior embedding are input into a pre-trained RNN model, and the customer's static information embedding and the customer's dynamic behavior embedding (including the customer behavior embedding and the item-embedding) are vector-concatenated and fused through the model to output the customer fused feature vector to comprehensively express the customer's characteristics and behavioral preferences.

[0053] It should be noted that in the above process, the fusion of customer static characteristics, dynamic characteristics and item characteristics is realized, which can comprehensively express the customer's characteristics and behavioral preferences, thereby better dividing the preference polarity of items and improving the accuracy of building customer portraits.

[0054] In an optional embodiment, a customer fusion feature label is determined based on a customer fusion feature vector, including: obtaining label feature vectors of multiple preset portrait labels; performing similarity calculations on each label feature vector and the customer fusion feature vector to obtain multiple calculation results, wherein the calculation results are used to characterize the degree of similarity between the label feature and the customer fusion feature; based on the multiple calculation results, determining a target portrait label from multiple preset portrait labels, and determining a customer fusion feature label based on the target portrait label.

[0055] Optionally, in the process of determining the customer fusion feature label based on the customer fusion feature vector, first obtain the label feature vectors of multiple preset portrait labels, and then perform similarity calculations on each label feature vector and the customer fusion feature vector respectively to obtain multiple calculation results, wherein the calculation results are used to characterize the similarity between the label feature and the customer fusion feature. Specifically, the feature is mapped to the dimension of the customer portrait through the fully connected layer or softmax layer of the model, for example, the similarity between the label feature of the preset portrait label and the customer fusion feature is calculated respectively to obtain multiple similarity probability values.

[0056] Optionally, based on multiple calculation results, a target portrait tag can be determined from multiple preset portrait tags, and the customer fusion feature tag can be determined based on the target portrait tag. For example, based on multiple calculated similarity probability values, at least one preset portrait tag with a similarity probability value greater than a preset threshold can be selected from multiple preset portrait tags as the target portrait tag, and then the target portrait tag is used as the customer fusion feature tag.

[0057] It should be noted that, through the above process, the accurate determination of customer fusion feature labels is achieved, making the label representation richer and more accurate, and improving the accuracy of building customer portraits.

[0058] In an optional embodiment, a target model is generated by the following steps: obtaining sample data, and dividing the sample data into a training set and a test set, wherein the sample data is labeled customer data; constructing an initial model based on a recurrent neural network, and training the initial model based on the training set to obtain a trained model; optimizing the trained model based on the test set and preset evaluation criteria to obtain a target model.

[0059] Optionally, in the process of generating the target model, sample data is first obtained, and the sample data is divided into a training set and a test set, then an initial model is constructed based on a recurrent neural network, and the initial model is trained based on the training set to obtain a trained model, and then the trained model is optimized based on the test set and a preset evaluation standard to obtain a target model. For example, labeled customer behavior data and customer portraits are used as training samples, and the parameters of the model are optimized through a back propagation algorithm, and different loss functions (for example, cross entropy loss or mean square error loss) are used to measure the gap between the model's prediction results and the actual customer portrait, and the weights and parameters of the model are continuously adjusted through an iterative training process to improve the performance and accuracy of the model.

[0060] It should be noted that the target model is generated through the above process, which provides a data basis for the integration of customer static features, dynamic features and item features, thereby being able to comprehensively express customer characteristics and behavioral preferences and improve the accuracy of building customer portraits.

[0061] In an optional embodiment, after constructing a customer profile based on the customer fusion feature tags, item information of multiple items to be recommended is obtained; based on the customer profile and the item information of the multiple items to be recommended, a target item to be recommended is determined from the multiple items to be recommended, and the item information of the target item to be recommended is pushed to the target customer corresponding to the customer profile.

[0062] Optionally, after constructing a customer profile based on the customer fusion feature labels, the trained RNN model and the generated customer profile can be used for personalized recommendation activities. Optionally, first obtain the item information of multiple items to be recommended, and then determine the target item to be recommended from the multiple items to be recommended based on the customer profile and the item information of the multiple items to be recommended, and push the item information of the target item to be recommended to the target customer corresponding to the customer profile. For example, based on the customer profile, the financial products and loan plans that best suit the customer's interests and preferences can be recommended to the customer.

[0063] It should be noted that by building customer portraits, a reliable basis is provided for making personalized recommendations, thus enabling accurate recommendations.

[0064] Figure 2 is a schematic diagram of an optional process of building a customer portrait according to an embodiment of the present invention, such as Figure 2 As shown, firstly, the original customer data is obtained, for example, the basic data of the customer (for example, age, gender, etc.), behavior data and item information data related to the behavior are obtained, and word embedding processing is performed respectively to obtain static information embedding (i.e., user-information-embedding) representing the static characteristics of the customer, customer behavior embedding (i.e., action-embedding) representing the customer behavior characteristics and item-embedding representing the corresponding item information, and the action-embedding is fused with the item-embedding to obtain the user-action-embedding, and then the user-information-embedding is fused with the user-action-embedding to obtain the customer fusion feature vector (i.e., user-embedding), and then the correlation between the customer behavior and the item is calculated by the doc2vec algorithm, so as to realize the analysis of the multi-dimensional customer behavior characteristics of the customer based on the basic information of the customer, the purchase behavior attributes and the product preference attributes by using the deep learning algorithm, and to build a more accurate customer behavior portrait, so as to analyze the customer's preferences, habits, risk tolerance, etc. according to the portrait, and provide customers with more accurate personalized recommendation services.

[0065] It can be seen that through the technical solution of the present invention, the purpose of better constructing customer portraits and achieving accurate recommendations is achieved, thereby achieving the technical effect of improving the accuracy of constructing customer portraits, and further solving the technical problem in the prior art of using discrete labels to represent customer portrait information, resulting in low accuracy of the constructed customer portraits.

[0066] Example 2

[0067] According to an embodiment of the present invention, an embodiment of a device for constructing a customer portrait is provided, wherein: Figure 3 is a schematic diagram of an optional device for constructing a customer portrait according to an embodiment of the present invention, such as Figure 3 As shown, the device includes: an acquisition module 301, which is used to acquire target customer data, wherein the target customer data includes customer static information data, customer dynamic information data and item information data, and the item information data is information data of items related to the behavior data in the customer dynamic information data; a first processing module 302, which is used to perform word embedding processing on the target customer data to obtain a first word vector and a second word vector, wherein the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the item information data; a second processing module 303, which is used to perform fusion processing on the first word vector and the second word vector through a target model to obtain a customer fusion feature vector, wherein the target model is obtained by training a recurrent neural network with sample data, and the customer fusion feature vector is used to characterize the fusion features of the customer static information data, the customer dynamic information data and the item information data; a construction module 304, which is used to determine a customer fusion feature label based on the customer fusion feature vector, and to construct a customer portrait based on the customer fusion feature label.

[0068] It should be noted that the above-mentioned acquisition module 301, first processing module 302, second processing module 303 and construction module 304 correspond to steps S101 to S104 in the above-mentioned embodiment, and the examples and application scenarios implemented by the four modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned embodiment 1.

[0069] Optionally, the acquisition module includes: a first acquisition module, used to acquire original customer data, wherein the original customer data includes customer basic data, customer behavior data and item data of items related to the customer behavior data; a third processing module, used to encode the customer basic data to obtain the customer static feature vector, and encode the item data to obtain the item feature vector; a fourth processing module, used to extract features from the customer behavior data through a pre-trained feature processing model to obtain customer dynamic features, and encode the customer dynamic features to obtain a customer dynamic feature vector; a first generation module, used to generate target customer data based on the customer static feature vector, the customer dynamic feature vector and the item feature vector.

[0070] Optionally, the first processing module includes: a second acquisition module, used to acquire the word embedding matrix of the customer static feature vector, the word embedding matrix of the customer dynamic feature vector and the word embedding matrix of the item feature vector; a first calculation module, used to multiply the customer static feature vector by the word embedding matrix of the customer static feature vector to obtain a first word vector; a second calculation module, used to multiply the item feature vector by the word embedding matrix of the item feature vector to obtain the item feature word vector after word embedding; a fifth processing module, used to process the customer dynamic feature vector and the word embedding matrix of the customer dynamic feature vector through a pre-trained feature processing model to obtain the dynamic feature word vector after word embedding; a sixth processing module, used to perform information fusion on the item feature word vector after word embedding and the dynamic feature word vector after word embedding to obtain a second word vector.

[0071] Optionally, the second processing module includes: a seventh processing module, used to input the first word vector and the second word vector into a target model; an eighth processing module, used to perform vector concatenation and fusion on the first word vector and the second word vector through the target model, and output a customer fusion feature vector.

[0072] Optionally, the construction module includes: a third acquisition module, used to obtain label feature vectors of multiple preset portrait labels; a third calculation module, used to perform similarity calculation on each label feature vector and the customer fusion feature vector respectively, to obtain multiple calculation results, wherein the calculation results are used to characterize the degree of similarity between the label feature and the customer fusion feature; a determination module, used to determine the target portrait label from multiple preset portrait labels based on the multiple calculation results, and determine the customer fusion feature label based on the target portrait label.

[0073] Optionally, the device for constructing a customer portrait also includes the following modules, which are used to generate a target model through the following steps: a fourth acquisition module, used to acquire sample data and divide the sample data into a training set and a test set, wherein the sample data is labeled customer data; a training module, used to construct an initial model based on a recurrent neural network, and train the initial model according to the training set to obtain a trained model; an optimization module, used to optimize the trained model according to the test set and preset evaluation criteria to obtain a target model.

[0074] Optionally, the customer portrait construction device also includes: a fifth acquisition module, used to obtain item information of multiple items to be recommended after the customer portrait is constructed based on the customer fusion feature tags; a recommendation module, used to determine a target item to be recommended from multiple items to be recommended based on the customer portrait and the item information of the multiple items to be recommended, and push the item information of the target item to be recommended to the target customer corresponding to the customer portrait.

[0075] Example 3

[0076] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned method for constructing a customer profile when running.

[0077] Example 4

[0078] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, wherein: Figure 4 is a schematic diagram of an optional electronic device according to an embodiment of the present invention, such as Figure 4 As shown, the electronic device includes one or more processors; a memory for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors are used to run the programs, wherein the programs are set to execute the above-mentioned method for constructing customer portraits when running. When the processor executes the program, the following steps are implemented: obtaining target customer data, wherein the target customer data includes customer static information data, customer dynamic information data and item information data, and the item information data is information data of items related to the behavior data in the customer dynamic information data; performing word embedding processing on the target customer data to obtain a first word vector and a second word vector, wherein the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the item information data; fusing the first word vector and the second word vector through the target model to obtain a customer fusion feature vector, wherein the target model is obtained by training the recurrent neural network with sample data, and the customer fusion feature vector is used to characterize the fusion features of the customer static information data, the customer dynamic information data and the item information data; determining the customer fusion feature label based on the customer fusion feature vector, and constructing the customer portrait based on the customer fusion feature label.

[0079] Optionally, when the processor executes the program, it also implements the following steps: obtaining target customer data, including: obtaining original customer data, wherein the original customer data includes customer basic data, customer behavior data, and item data of items related to the customer behavior data; encoding the customer basic data to obtain a customer static feature vector, and encoding the item data to obtain an item feature vector; extracting features from the customer behavior data through a pre-trained feature processing model to obtain customer dynamic features, and encoding the customer dynamic features to obtain a customer dynamic feature vector; generating target customer data based on the customer static feature vector, the customer dynamic feature vector, and the item feature vector.

[0080] Optionally, when the processor executes the program, the following steps are also implemented: word embedding processing is performed on the target customer data to obtain a first word vector and a second word vector, including: obtaining a word embedding matrix of a customer static feature vector, a word embedding matrix of a customer dynamic feature vector, and a word embedding matrix of an item feature vector; multiplying the customer static feature vector by the word embedding matrix of the customer static feature vector to obtain a first word vector; multiplying the item feature vector by the word embedding matrix of the item feature vector to obtain an embedded item feature word vector; processing the customer dynamic feature vector and the word embedding matrix of the customer dynamic feature vector through a pre-trained feature processing model to obtain an embedded dynamic feature word vector; and performing information fusion on the embedded item feature word vector and the embedded dynamic feature word vector to obtain a second word vector.

[0081] Optionally, when the processor executes the program, the following steps are also implemented: the first word vector and the second word vector are fused through the target model to obtain a customer fused feature vector, including: inputting the first word vector and the second word vector into the target model; performing vector concatenation and fusion on the first word vector and the second word vector through the target model, and outputting a customer fused feature vector.

[0082] Optionally, when executing the program, the processor also implements the following steps: determining a customer fusion feature label based on a customer fusion feature vector, including: obtaining label feature vectors of multiple preset portrait labels; performing similarity calculations on each label feature vector and the customer fusion feature vector respectively to obtain multiple calculation results, wherein the calculation results are used to characterize the degree of similarity between the label feature and the customer fusion feature; determining a target portrait label from multiple preset portrait labels based on the multiple calculation results, and determining a customer fusion feature label based on the target portrait label.

[0083] Optionally, the processor also implements the following steps when executing the program: Generate a target model through the following steps: Obtain sample data, and divide the sample data into a training set and a test set, wherein the sample data is labeled customer data; Construct an initial model based on a recurrent neural network, and train the initial model based on the training set to obtain a trained model; Optimize the trained model based on the test set and preset evaluation criteria to obtain a target model.

[0084] Optionally, the processor also implements the following steps when executing the program: after constructing a customer profile based on the customer fusion feature tags, obtaining item information of multiple items to be recommended; determining a target item to be recommended from the multiple items to be recommended based on the customer profile and the item information of the multiple items to be recommended, and pushing the item information of the target item to be recommended to the target customer corresponding to the customer profile.

[0085] The devices in this article can be servers, PCs, PADs, mobile phones, etc.

[0086] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0087] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0088] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0089] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0090] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0091] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0092] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a customer portrait, characterized in that: include: Acquire target customer data, wherein the target customer data includes customer static information data, customer dynamic information data, and item information data, wherein the item information data is information data of items related to the behavior data in the customer dynamic information data; Performing word embedding processing on the target customer data to obtain a first word vector and a second word vector, wherein the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the item information data; The first word vector and the second word vector are fused by a target model to obtain a customer fusion feature vector, wherein the target model is obtained by training a recurrent neural network with sample data, and the customer fusion feature vector is used to represent the fusion features of the customer static information data, the customer dynamic information data, and the item information data; A customer fusion feature label is determined based on the customer fusion feature vector, and a customer portrait is constructed based on the customer fusion feature label.

2. The method according to claim 1, characterized in that Obtain target customer data, including: Acquire original customer data, wherein the original customer data includes customer basic data, customer behavior data, and item data of items related to the customer behavior data; Encoding the customer basic data to obtain a customer static feature vector, and encoding the item data to obtain an item feature vector; Extracting features from the customer behavior data using a pre-trained feature processing model to obtain customer dynamic features, and encoding the customer dynamic features to obtain a customer dynamic feature vector; The target customer data is generated according to the customer static feature vector, the customer dynamic feature vector and the item feature vector.

3. The method according to claim 2, characterized in that The target customer data is subjected to word embedding processing to obtain a first word vector and a second word vector, including: Obtaining a word embedding matrix of the customer static feature vector, a word embedding matrix of the customer dynamic feature vector, and a word embedding matrix of the item feature vector; Multiplying the customer static feature vector by the word embedding matrix of the customer static feature vector to obtain the first word vector; Multiplying the item feature vector by the word embedding matrix of the item feature vector to obtain the item feature word vector after word embedding; Processing the customer dynamic feature vector and the word embedding matrix of the customer dynamic feature vector through the pre-trained feature processing model to obtain a dynamic feature word vector after word embedding; Information fusion is performed on the item feature word vector after the word embedding and the dynamic feature word vector after the word embedding to obtain the second word vector.

4. The method according to claim 1, characterized in that: The first word vector and the second word vector are fused by the target model to obtain a customer fused feature vector, including: Inputting the first word vector and the second word vector into the target model; The first word vector and the second word vector are concatenated and fused through the target model to output the customer fused feature vector.

5. The method according to claim 1, characterized in that Determining a customer fusion feature label according to the customer fusion feature vector includes: Obtain label feature vectors of multiple preset portrait labels; Performing similarity calculation on each of the label feature vectors and the customer fusion feature vector respectively to obtain multiple calculation results, wherein the calculation results are used to characterize the similarity between the label feature and the customer fusion feature; Based on the multiple calculation results, a target portrait label is determined from the multiple preset portrait labels, and based on the target portrait label, the customer fusion feature label is determined.

6. The method according to claim 1, characterized in that The target model is generated by following the steps below: Acquire the sample data, and divide the sample data into a training set and a test set, wherein the sample data is labeled customer data; Constructing an initial model based on the recurrent neural network, and training the initial model according to the training set to obtain a trained model; The trained model is optimized according to the test set and preset evaluation criteria to obtain the target model.

7. The method according to claim 1, characterized in that After constructing the customer portrait according to the customer fusion feature tag, the method further includes: Obtain item information of multiple items to be recommended; According to the customer portrait and the item information of the multiple items to be recommended, a target item to be recommended is determined from the multiple items to be recommended, and the item information of the target item to be recommended is pushed to the target customer corresponding to the customer portrait.

8. A device for constructing a customer portrait, characterized in that: include: An acquisition module, used to acquire target customer data, wherein the target customer data includes customer static information data, customer dynamic information data and item information data, wherein the item information data is information data of items related to the behavior data in the customer dynamic information data; A first processing module is used to perform word embedding processing on the target customer data to obtain a first word vector and a second word vector, wherein the first word vector corresponds to the customer static information data, and the second word vector corresponds to the customer dynamic information data and the item information data; A second processing module is used to fuse the first word vector and the second word vector through a target model to obtain a customer fusion feature vector, wherein the target model is obtained by training a recurrent neural network through sample data, and the customer fusion feature vector is used to represent the fusion features of the customer static information data, the customer dynamic information data, and the item information data; A construction module is used to determine a customer fusion feature label based on the customer fusion feature vector, and to construct a customer portrait based on the customer fusion feature label.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the method for constructing a customer portrait as described in any one of claims 1 to 7 when run.

10. An electronic device, characterized in that: The electronic device includes one or more processors; A memory for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to run the programs, wherein the program is configured to execute the method for constructing a customer profile as described in any one of claims 1 to 7 at runtime.

Images