WO2019223379A1 - 一种产品推荐方法和装置 - Google Patents

一种产品推荐方法和装置 Download PDF

Info

Publication number
WO2019223379A1
WO2019223379A1 PCT/CN2019/076240 CN2019076240W WO2019223379A1 WO 2019223379 A1 WO2019223379 A1 WO 2019223379A1 CN 2019076240 W CN2019076240 W CN 2019076240W WO 2019223379 A1 WO2019223379 A1 WO 2019223379A1
Authority
WO
WIPO (PCT)
Prior art keywords
product
user
feature matrix
recommended
matrix
Prior art date
Application number
PCT/CN2019/076240
Other languages
English (en)
French (fr)
Inventor
张连彬
Original Assignee
阿里巴巴集团控股有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 阿里巴巴集团控股有限公司 filed Critical 阿里巴巴集团控股有限公司
Publication of WO2019223379A1 publication Critical patent/WO2019223379A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241Advertisements
    • G06Q30/0251Targeted advertisements
    • G06Q30/0255Targeted advertisements based on user history
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241Advertisements
    • G06Q30/0251Targeted advertisements
    • G06Q30/0269Targeted advertisements based on user profile or attribute
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06Asset management; Financial planning or analysis

Definitions

  • the present disclosure relates to the field of data processing technology, and in particular, to a product recommendation method and device.
  • Cold start and data sparseness are common problems in the product recommendation area.
  • Cold start means product recommendation in the absence of a large amount of user data support; sparse data means that the items that interact with the user are only the tip of the iceberg of the overall project, resulting in extremely sparse data in the user item rating matrix.
  • the user's behavior information is scarce, and there is not a large amount of user data for product recommendation, resulting in cold start problems;
  • the purchase behavior of users' financial wealth management products accounts for only a small part of the total registered users of the wealth management product platform, and the problem of data sparseness is also very prominent.
  • the most widely used personalized recommendation method is based on single-field collaborative filtering, that is, recommending to the target user the product that the user who is most similar to his interests and preferences, or the product that is most similar to the product that he once liked.
  • single-field collaborative filtering that is, recommending to the target user the product that the user who is most similar to his interests and preferences, or the product that is most similar to the product that he once liked.
  • how to give users a more satisfactory recommendation result under the condition of cold start and sparse data is an urgent problem to be solved.
  • one or more embodiments of the present specification provide a product recommendation method and device to improve the quality of product recommendation in the absence of data.
  • a product recommendation method is provided. The method is used to determine whether to recommend a product to be recommended to a target user. The method includes:
  • the multi-domain information includes: the target user's purchase data in the product area of the product to be recommended and purchase data in other product areas;
  • the user feature matrix comprising: a plurality of feature values quantified according to the multi-domain information
  • For one of the products to be recommended obtain the user feature matrix of multiple users who purchased the product to be recommended, and obtain the product to be recommended based on the feature values in the user feature matrix of the multiple users.
  • the corresponding product feature matrix
  • the user feature matrix and product feature matrix are respectively input into a pre-trained machine learning model to obtain a user preference vector and a product preference vector.
  • the user preference vector is used to represent a target user's preference in product purchase, and the product preference vector Used to indicate the characteristics of a user who purchases the product to be recommended;
  • the selection evaluation value is greater than a predetermined recommendation threshold, it is determined to recommend the product to be recommended to the target user.
  • a product recommendation device is provided.
  • the device is used to determine whether to recommend a product to be recommended to a target user.
  • the device includes:
  • An information acquisition module configured to acquire multi-domain information associated with the target user, where the multi-domain information includes: the target user's purchase data in the product field of the product to be recommended and purchase data in other product fields;
  • a user matrix construction module configured to construct a user feature matrix of the target user according to the multi-domain information, the user feature matrix including: a plurality of feature values quantified according to the multi-domain information;
  • a product matrix construction module is configured to obtain, for one product to be recommended, the user feature matrix of multiple users who purchase the product to be recommended, and based on the feature values in the user feature matrix of the multiple users To obtain a product feature matrix corresponding to the product to be recommended;
  • a model processing module configured to input the user feature matrix and the product feature matrix into a pre-trained machine learning model, respectively, to obtain a user preference vector and a product preference vector, and the user preference vector is used to represent a target user's preference on product purchase ,
  • the product preference vector is used to represent a characteristic of a user who purchases the product to be recommended;
  • An output processing module configured to obtain a selection evaluation value between the product to be recommended and the target user according to the user preference vector and the product preference vector, and the selection evaluation value is used to indicate that the target user purchases the target user Probability of product to be recommended;
  • the recommendation determining module is configured to determine to recommend the product to be recommended to the target user when the selection evaluation value is greater than a predetermined recommendation threshold.
  • a product recommendation device includes a memory, a processor, and computer instructions stored on the memory and executable on the processor. When the processor executes the instructions, the following steps are implemented:
  • the multi-domain information includes: the target user's purchase data in the product area of the product to be recommended and purchase data in other product areas;
  • the user feature matrix comprising: a plurality of feature values quantified according to the multi-domain information
  • For one of the products to be recommended obtain the user feature matrix of multiple users who purchased the product to be recommended, and obtain the product to be recommended based on the feature values in the user feature matrix of the multiple users.
  • the corresponding product feature matrix
  • the user feature matrix and product feature matrix are respectively input into a pre-trained machine learning model to obtain a user preference vector and a product preference vector.
  • the user preference vector is used to represent a target user's preference in product purchase, and the product preference vector Used to indicate the characteristics of a user who purchases the product to be recommended;
  • the selection evaluation value is greater than a predetermined recommendation threshold, it is determined to recommend the product to be recommended to the target user.
  • the product recommendation method and device of one or more embodiments of this specification help the user to select the appropriate feature by integrating user behavior data and basic information in multiple fields and intelligently sensing the preference characteristics of the user related to the product purchase by using a deep neural network.
  • Financial wealth management products effectively alleviate the problem of sparse transaction data and cold start in the industry, effectively improve the accuracy of personalized recommendation of financial wealth management products, and provide target users with more accurate recommendation services.
  • FIG. 1 is a process of model training provided by one or more embodiments of the present specification
  • FIG. 2 is an attribute interaction operation principle of a feature matrix provided by one or more embodiments of the present specification
  • FIG. 3 is a schematic diagram of neural network processing provided by one or more embodiments of the present specification.
  • FIG. 4 is a schematic structural diagram of a product recommendation device provided by one or more embodiments of the present specification.
  • One or more embodiments of this specification provide a product recommendation method when the data is sparse.
  • the description of this method takes the recommendation of financial wealth management products as an example, but it is understandable that the method can also be applied to other cold start features.
  • Product recommendation scenarios are possible.
  • the recommendation method incorporates user behavior data from different fields, and alleviates data sparseness and cold-start problems in the product to-be-recommended field by using behavior information in other fields outside the product-to-be-recommended field. Because users' purchase behaviors in other fields can also reflect the user's identity characteristics, environmental characteristics, lifestyle tastes and other information that help reflect the user's product purchase preferences, the recommendations in the field of recommended products can also be a good reference.
  • the recommendation method also uses a machine learning model.
  • a machine learning model For example, using a deep neural network as an example, the output of the deep neural network model is used to assist in product recommendation.
  • the deep neural network model can be model trained first, and the trained model can be used for product recommendation.
  • a target matrix for model training can be constructed based on the actual collected data of product purchase.
  • the actual collected data may include user purchase data of the product, for example, it may be the user ’s actual purchase record of financial wealth management products.
  • user A purchased fund J1, user B purchased stock G1 and fund J1, and user C purchased Fund J2, etc.
  • a target matrix can be constructed, as shown in Table 1 below, but it is not limited to this:
  • the target matrix may include a user's purchase selection value for a product, and the purchase selection value is used to indicate whether the user purchases a product.
  • the purchase selection value may include "1" or "0". When the value is 1, it indicates that the user has purchased the product; when the value is 0, it indicates that the user has not purchased the product.
  • the target matrix can be used as a training target for a deep neural network model. When the deviation between the output of the model under training and the training target becomes smaller and smaller, and the deviation reaches a predetermined threshold, the training of the model is ended, and the training The closed model is directly used for recommendation of subsequent financial wealth management products.
  • a user feature matrix for each user can be separately constructed, and product characteristics of each product can be separately constructed matrix.
  • the constructed user feature matrix and product feature matrix are input into a machine learning model to be trained, and a model output matrix is output.
  • the model output matrix includes each purchase selection value output through the machine learning model.
  • model training is described in detail in FIG. 1 as follows. This process describes how to construct the above-mentioned user feature matrix, product feature matrix, and how to input the matrix into the model to train the model.
  • step 100 multi-domain information associated with the target user is obtained, where the multi-domain information includes: target user's purchase data in the product area of the product to be recommended and purchase data in other product areas.
  • the target user is the user of the product to be recommended. For example, if the user wants to recommend the product to user A, but does not know which product to recommend to the user A, it is necessary to determine the recommendation to the user A through the recommendation method of this embodiment. Product, then the user A can be called the target user.
  • the target users here may be users in the target matrix, and these users have actually experienced actual purchase behavior.
  • the target user may be a user who has not yet purchased certain products for product recommendation.
  • the product area of the product to be recommended is the financial wealth management product.
  • the purchase data of the target user in the product area of the product to be recommended may include, for example, the transaction amount of the user purchasing a financial wealth management product.
  • the purchase data in other product areas can be the purchase of non-financial wealth management products, for example, it can be clothes, rice cookers, etc.
  • the purchase data in the other product field may be the purchase price of the products in the other field, for example, the clothes purchased are 200 yuan, and the rice cooker is 350 yuan. Regardless of the product data of the product to be recommended or the purchase data of other product fields, the data is generated by the target user's purchase.
  • multi-domain information is not limited to the aforementioned purchase data of different product domains, and may also include other information. There are several examples below, including but not limited to the following information:
  • user attribute information of the target user may be a user's gender, age, education, and the like.
  • the associated user of the target user may have a friend relationship, a transfer relationship, and the like with the target user.
  • a friend relationship as an example, it may be data of a financial product purchased by a friend of the target user.
  • friend A of user A has purchased a financial product and the transaction amount is 20,000.
  • target user's lending behavior data can be used for the lending behavior of the target user, the product of a certain category is loaned, and the amount of the loan.
  • a user feature matrix of the target user is constructed according to the multi-domain information, and the user feature matrix includes a plurality of feature values quantified according to the multi-domain information.
  • quantization may be performed based on the data collected in step 100 and converted into feature values.
  • Table 2 illustrates the form of a user feature matrix
  • the coarse-grained processing of the product can be performed first.
  • Coarse-grained processing is to transform the more detailed data in the data set into generalized and highly comprehensive data. If, for the purchase data of one product category, the number of products purchased under the product category reaches a coarse-grained processing condition, a plurality of products under the product category are subjected to coarse-grained processing.
  • the purchase data of other product areas include the purchase of clothing, rice cookers, and other relatively fine categories
  • target users have purchased Boaisi DFB-B 0.8L, Oaks AR-Y0801 , Lobe LBF-091BM and other small-capacity rice cookers are processed as 0-1L non-computer mini rice cookers, Midea MB-WHS30C96, Mijia Pressure IH, Panasonic SR-AE101-K and other household intelligent rice cookers and other rice cookers. Then, when the user feature matrix is constructed, these products are divided very finely.
  • small-capacity rice cookers such as Boaisi DFB-B 0.8L, Oaks AR-Y0801, and Lobe LBF-091BM are processed as 0-1L non-computer mini rice cookers, Midea MB-WHS30C96, Mijia Pressure IH, Panasonic SR-AE101-K And other household automatic intelligent rice cookers are processed into 3L-4L intelligent microcomputer rice cookers.
  • Whether to perform coarse-grained processing on the purchase data of a product category can set coarse-grained processing conditions.
  • the condition may be that the number of products purchased under the product category reaches a certain number threshold, for example, the number of products under the same product category reaches more than six.
  • the characteristic dimensions of user attribute information, social relationships, and financial and financial product purchase and loan behaviors in Table 2 because of its small characteristic dimensions and high information content, coarse-grained processing may not be required.
  • the transaction amount can be reasonably divided into multiple intervals according to the transaction amount of the purchased financial wealth management product, such as “ ⁇ P1”, “P1-P2”, “P2-P3” and other intervals in Table 2. If the amount of the user's purchase of the financial management product is within this range, it is marked as 1; otherwise, it is 0.
  • the purchase data of users who have an affiliation with the target user to purchase financial wealth management products Since the affiliated user may be multiple users, the user of all users who have established a social relationship with the target user may be first
  • the transaction amounts are averaged and marked based on the average amount. For example, if the average value is in the interval "P1-P2", the eigenvalue corresponding to the interval can be marked with 1.
  • the purchase data of other products are processed with coarse granularity.
  • all products in this category can be reasonably divided into each attribute interval of Table 2 using the price index, and the value of 0-1 of the user's overall purchase frequency is standardized as its attribute value, which reflects the target user's purchase of this category.
  • the frequency of products in the price range is
  • the target user purchases small-capacity products such as PALSIS DFB-B 0.8L, Oaks AR-Y0801, and Lobe LBF-091BM in the purchase behavior of other products under the category of 3L-4L intelligent microcomputer rice cookers.
  • the rice cooker is handled as a 0-1L non-computer mini rice cooker, Midea MB-WHS30C96, Mijia Pressure IH, Panasonic SR-AE101-K and other household automatic intelligent rice cookers, that is, a variety of products have been purchased under the same category.
  • the quantification of this lending behavior is similar to the quantification of financial wealth management products. It also divides the loan amount into multiple intervals reasonably. If the amount of the user's loan for this category of products falls within this interval, it is marked as 1; otherwise, it is 0.
  • a quantized value corresponds to 18 to 25 years old
  • a quantized value corresponds to 26 to 35 years old.
  • variable factors For example, for categorical variables, such as gender and education, you can encode the variable factors and label them. For example, a bachelor degree may correspond to a quantified value, and a graduate degree may correspond to a quantified value.
  • step 104 for a plurality of products, the user feature matrix of multiple users purchasing the product is obtained, and based on the feature values in the user feature matrix of the multiple users, a product feature matrix of the product is obtained. .
  • the product in this step is a financial product.
  • a product feature matrix can be constructed, a product feature matrix can correspond to a product, and the product can be each product in the target matrix.
  • the construction of the product feature matrix can be based on the user feature matrix.
  • the feature values may be weighted and averaged.
  • each user who purchases the product has a feature value corresponding to the age.
  • the feature values of multiple users may be weighted and averaged to obtain a comprehensive feature value corresponding to the age.
  • each of the multiple users has a feature value corresponding to category 1, and the feature values of multiple users can be weighted and averaged. A comprehensive characteristic value corresponding to category 1 is obtained.
  • each feature value in Table 2 corresponds to a different feature value position
  • the feature value position corresponding to x1 in Table 2 is [the row corresponds to the "P1-P2" interval, and the column corresponds to "Category 1"]
  • the eigenvalue position corresponding to the eigenvalue x2 is [the row corresponds to the "P2-P3" interval, and the column corresponds to the "category 1"].
  • the feature values corresponding to the same feature value position in the user feature matrix of multiple users may be weighted and averaged to obtain the feature values corresponding to the feature value position in the product feature matrix.
  • the feature values of multiple users can be weighted and averaged to finally obtain a product feature matrix that can reflect the overall characteristics of the user who purchased the product.
  • the setting of the weight when the feature value is weighted average may be determined according to the actual business situation. For example, if you think that a user's characteristic value is more important in reflecting the overall characteristics of the user, set its weight higher.
  • step 106 attribute interaction operations are respectively performed on the user feature matrix and the product feature matrix.
  • the attribute interaction operations of the user feature matrix and the product feature matrix can be performed.
  • the attribute interaction operation is to establish the correlation between the attributes that are not directly related in the matrix.
  • the constructed feature matrix is randomly sorted by the attribute column as a unit to generate multiple new feature matrices, and then the multiple new feature matrices are spliced to generate the attribute interaction.
  • this attribute interaction operation can be an optional operation. After the attribute interaction operation is performed, the potential association between different features can be more effectively discovered, so that it will be more effective in the subsequent use of machine learning models to perceive user preferences. accurate.
  • each feature such as feature 1, feature 2, and feature 3 corresponds to a different feature row.
  • feature 1 may be "Product 1 in the purchase behavior of financial wealth management products” in Table 1
  • feature 15 may be "Lending category 1 in lending behavior” in Table 1, that is, different characteristics Corresponds to different columns.
  • FIG. 2 it is equivalent to randomly moving between different columns in Table 1, sorting randomly by the unit of the column, and then stitching.
  • step 108 the user feature matrix and product feature matrix after interaction are respectively input into a machine learning model to obtain a user preference vector and a product preference vector.
  • the deep neural network includes two parallel neural networks, one of which is an intelligent perceptron for user behavior preferences, and the other is an intelligent perceptron for overall feature preferences of users who purchase the product, as shown in FIG. 3.
  • the attribute matrix after the attribute interaction and stitching is used as the input of the parallel neural network.
  • the user feature matrix after the attribute interaction is input into one neural network
  • the product feature matrix after the attribute interaction is input into another neural network.
  • the neural network can obtain the user preference vector and the product preference vector, respectively.
  • the user preference vector can be used to indicate a user's preference in product purchase, which is equivalent to indicating what kind of product a user likes to purchase.
  • the product preference vector can be used to indicate the characteristics of a user who purchases a product corresponding to the product feature matrix, which is equivalent to indicating that for a product, users with characteristics are more inclined to purchase the product.
  • a model output matrix is obtained according to a user preference vector and a product preference vector output by the model, and the model output matrix includes each purchase selection value output through a machine learning model.
  • a user feature matrix corresponding to a user is input to a neural network model to obtain a user preference vector; a product feature matrix corresponding to a product is input to a neural network model to obtain a product preference vector.
  • a purchase choice value can be obtained according to the user preference vector and the product preference vector.
  • an inner product of the user preference vector and the product preference vector may be obtained to obtain a purchase selection value, where the selection value represents a probability that the user purchases the product.
  • a user feature matrix For each user in the target matrix, a user feature matrix can be constructed, and for each product, a corresponding product feature matrix can be constructed separately. According to the method described above, one user's purchase selection value for one product can be obtained.
  • These purchase selection values can constitute a model output matrix, that is, each purchase selection value included in the model output matrix is a value output by the neural network model.
  • the target matrix includes the user's purchase selection value of the product, which is obtained based on the actual collected data. It is the actual purchase behavior of the user.
  • the target matrix is the mutual selection matrix between the user and the product.
  • the target matrix can be used as the training target of the neural network model. As the model is continuously optimized, the output of the neural network model will be closer to the actual occurrence value.
  • step 112 when the deviation between the model output matrix and the target matrix reaches a predetermined threshold, the model training ends.
  • the deviation between the model output matrix and the target matrix may be ended.
  • the deviation reaching the predetermined threshold may be that the deviation is smaller than or equal to the predetermined threshold, that is, the deviation between the two is sufficiently small.
  • the deviation can be measured using root mean square error RMSE (Root Mean Error) or mean absolute error MAE (Mean Absolute Deviation).
  • the products to be recommended include: product C1, product C2, and product C3. Then, which financial wealth management product to recommend to user Y will have a higher success rate.
  • the recommended method for this example executes.
  • a user feature matrix of user Y may be constructed first, and a product feature matrix of multiple products such as product C1, product C2, and product C3 may be separately constructed.
  • the user feature matrix of the user Y and the product feature matrix of the product C1 are respectively input into a parallel neural network to obtain a user preference vector and a product preference vector.
  • a user Y's selection evaluation value for the product C1 is obtained, and the selection evaluation value is used to indicate the probability that the target user purchases the evaluation product.
  • the selection evaluation value is calculated in the same way as the purchase selection value mentioned above, except that the two names are used to distinguish.
  • the purchase selection value is the value calculated during model training.
  • the selection evaluation value is calculated when the model is used after training. The value is used as the basis for recommending products to users.
  • the above products to be recommended C1, C2, and C3 may be referred to as evaluation products, that is, whether these products are to be recommended to the user Y.
  • a selection evaluation value can be obtained between the product feature matrix of each product and the user feature matrix of user Y.
  • a recommendation threshold may be set, and when the selection evaluation value is greater than a predetermined recommendation threshold, it is determined to recommend the evaluation product to the target user. For example, suppose the selection evaluation value of product C1 and user Y is 0.6, the selection evaluation value of product C2 and user Y is 0.8, the selection evaluation value of product C3 and user Y is 0.2, and assuming that the recommendation threshold is 0.55, then It is determined that product C1 and product C2 are recommended to user Y, and product C3 is not recommended.
  • the personalized recommendation method for financial wealth management products in this example helps users to choose suitable financial wealth management products by integrating user behavior data and basic information in multiple fields, and using deep neural networks to intelligently sense the preference characteristics of users related to product purchases. , Effectively alleviate the problem of sparse transaction data and cold start facing the industry, effectively improve the accuracy of personalized recommendations for financial wealth management products, provide more accurate recommendation services for target users, and become a powerful force to promote positive interaction between sales platforms and users Measures.
  • the device may be used to determine whether to recommend a product to be recommended to a target user.
  • the device may include: an information acquisition module 41, a user matrix construction module 42, a product matrix construction module 43, a model processing module 44, and an output. Processing module 45 and recommendation determination module 46.
  • An information acquisition module 41 is configured to acquire multi-domain information associated with the target user, where the multi-domain information includes: the target user's purchase data in the product field of the product to be recommended and purchase data in other product fields;
  • a user matrix construction module 42 is configured to construct a user feature matrix of the target user according to the multi-domain information, where the user feature matrix includes: a plurality of feature values quantified according to the multi-domain information;
  • a product matrix construction module 43 is configured to obtain, for one product to be recommended, the user feature matrix of multiple users who purchase the product to be recommended, and based on the features in the user feature matrix of the multiple users Value to obtain a product feature matrix corresponding to the product to be recommended;
  • a model processing module 44 is configured to input the user feature matrix and the product feature matrix into a pre-trained machine learning model, respectively, to obtain a user preference vector and a product preference vector, and the user preference vector is used to represent a target user's product purchase Preference, the product preference vector is used to represent the characteristics of a user who purchases the product to be recommended;
  • An output processing module 45 is configured to obtain a selection evaluation value between the product to be recommended and the target user according to the user preference vector and the product preference vector, and the selection evaluation value is used to indicate that the target user purchases Describe the probability of recommending a product;
  • the recommendation determining module 46 is configured to determine to recommend the product to be recommended to the target user when the selection evaluation value is greater than a predetermined recommendation threshold.
  • the user matrix construction module 42 is further configured to: if, for the purchase data of one product category, the number of products purchased under the product category reaches a coarse-grained processing condition, a plurality of products under the product category are processed. The product is coarse-grained.
  • the product matrix construction module 43 is specifically configured to perform a weighted average of the feature values corresponding to the same feature value position in the user feature matrix of the multiple users to obtain the product feature matrix corresponding to the feature value.
  • the characteristic value of the position is specifically configured to perform a weighted average of the feature values corresponding to the same feature value position in the user feature matrix of the multiple users to obtain the product feature matrix corresponding to the feature value. The characteristic value of the position.
  • the model processing module 44 is further configured to perform attribute interaction operations on the user feature matrix and the product feature matrix before the user feature matrix and the product feature matrix are respectively input into a pre-trained machine learning model; The interactive user feature matrix and product feature matrix are input into the machine learning model.
  • the device or module explained in the foregoing embodiments may be specifically implemented by a computer chip or entity, or may be implemented by a product having a certain function.
  • a typical implementation device is a computer, and the specific form of the computer may be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email sending and receiving device, and a game control Desk, tablet computer, wearable device, or a combination of any of these devices.
  • each step may be implemented in the form of software, hardware, or a combination thereof.
  • those skilled in the art may implement it in the form of software code, and may be computer-executable capable of implementing the logical functions corresponding to the steps. instruction.
  • the executable instructions may be stored in a memory and executed by a processor in the device.
  • the device may include a processor, a memory, and computer instructions stored on the memory and executable on the processor.
  • the processor executes the instructions to implement the following steps:
  • the multi-domain information includes: the target user's purchase data in the product area of the product to be recommended and purchase data in other product areas;
  • the user feature matrix comprising: a plurality of feature values quantified according to the multi-domain information
  • For one of the products to be recommended obtain the user feature matrix of multiple users who purchased the product to be recommended, and obtain the product to be recommended based on the feature values in the user feature matrix of the multiple users.
  • the corresponding product feature matrix
  • the user feature matrix and product feature matrix are respectively input into a pre-trained machine learning model to obtain a user preference vector and a product preference vector.
  • the user preference vector is used to represent a target user's preference in product purchase, and the product preference vector Used to indicate the characteristics of a user who purchases the product to be recommended;
  • the selection evaluation value is greater than a predetermined recommendation threshold, it is determined to recommend the product to be recommended to the target user.
  • one or more embodiments of the present specification may be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the present specification may adopt a computer program implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code therein. The form of the product.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • One or more embodiments of the specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules.
  • program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types.
  • One or more embodiments of the present specification may also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected through a communication network.
  • program modules may be located in local and remote computer storage media, including storage devices.

Abstract

一种产品推荐方法和装置,其中,所述方法用于确定是否将待推荐产品推荐给目标用户,该方法包括:获取目标用户关联的多领域信息,该信息包括:目标用户在待推荐产品的产品领域的购买数据和其他产品领域的购买数据;根据多领域信息构建目标用户的用户特征矩阵;对于一个待推荐产品,获取购买待推荐产品的多个用户的用户特征矩阵,并基于矩阵中的特征值,得到待推荐产品的产品特征矩阵;分别将用户特征矩阵和产品特征矩阵输入机器学习模型,得到用户偏好向量和产品偏好向量;根据用户偏好向量和产品偏好向量得到待推荐产品和目标用户之间的选择评估值;在选择评估值大于预定的推荐阈值时,则确定将待推荐产品推荐给目标用户。

Description

一种产品推荐方法和装置 技术领域
本公开涉及数据处理技术领域,特别涉及一种产品推荐方法和装置。
背景技术
在产品推荐领域,冷启动和数据稀疏是常见的问题。冷启动即在缺少大量用户数据支撑的情况下进行产品推荐;数据稀疏即与用户产生交互关系的项目仅为总体项目的冰山一角,导致了用户项目评分矩阵的数据极端稀疏。例如,在金融理财产品的推荐上,由于金融理财行业本身所具有的交易数额大、频次低等属性,用户的行为信息稀少,并没有大量的用户数据用以做产品推荐,产生冷启动问题;并且,用户的金融理财产品的购买行为仅占理财产品平台的总注册用户的一小部分,数据稀疏问题也十分突出。
迄今为止,应用最为广泛的个性化推荐方法是基于单领域的协同过滤,即给目标用户推荐与他兴趣偏好最为相似的用户喜欢的产品,或与他曾经喜欢过的产品最为相似的产品。但是,如何在冷启动和数据稀疏的情况下,给用户一个较为满意的推荐结果,是一个亟待解决的问题。
发明内容
有鉴于此,本说明书一个或多个实施例提供一种产品推荐方法和装置,以提高数据缺少的情况下的产品推荐质量。
具体地,本说明书一个或多个实施例是通过如下技术方案实现的:
第一方面,提供一种产品推荐方法,所述方法用于确定是否将待推荐产品推荐给目标用户,所述方法包括:
获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的 产品特征矩阵;
分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
第二方面,提供一种产品推荐装置,所述装置用于确定是否将待推荐产品推荐给目标用户,所述装置包括:
信息获取模块,用于获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
用户矩阵构建模块,用于根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
产品矩阵构建模块,用于对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的产品特征矩阵;
模型处理模块,用于分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
输出处理模块,用于根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
推荐确定模块,用于在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
第三方面,提供一种产品推荐设备,所述设备包括存储器、处理器,以及存储在存储器上并可在处理器上运行的计算机指令,所述处理器执行指令时实现以下步骤:
获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述 待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的产品特征矩阵;
分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
本说明书一个或多个实施例的产品推荐方法和装置,通过融合多个领域的用户行为数据与基本信息,并利用深度神经网络智能化感知用户与产品购买相关的偏好特征,帮助用户挑选合适的金融理财产品,有效缓解该行业所面临的交易数据稀疏与冷启动问题,有效提高了金融理财产品个性化推荐的准确度,为目标用户提供更准确的推荐服务。
附图说明
为了更清楚地说明本说明书一个或多个实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本说明书一个或多个实施例中记载的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1为本说明书一个或多个实施例提供的模型训练的过程;
图2为本说明书一个或多个实施例提供的特征矩阵的属性交互操作原理;
图3为本说明书一个或多个实施例提供的神经网络处理示意图;
图4为本说明书一个或多个实施例提供的一种产品推荐装置的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本说明书一个或多个实施例中的技术方案,下面将结合本说明书一个或多个实施例中的附图,对本说明书一个或多个实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本说明书一部分实施例,而不是全部的实施例。基于本说明书一个或多个实施例,本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例,都应当属于本公开保护的范围。
本说明书一个或多个实施例提供了一种数据稀疏时的产品推荐方法,该方法的描述以金融理财产品的推荐为例,但是可以理解的是,该方法同样可以适用于其他具有冷启动特点的产品推荐场景。
其中,该推荐方法融合了来自不同领域的用户行为数据,借助待推荐的产品领域之外的其他领域的行为信息,缓解待推荐产品领域的数据稀疏和冷启动问题。因为用户在其他领域的购买行为也可以反应用户的身份特征、环境特征、生活品味等有助于反应用户产品购买偏好的信息,对待推荐产品领域的推荐也具有很好的借鉴作用。
此外,该推荐方法也使用了机器学习模型,例如,以深度神经网络为例,利用该深度神经网络模型的输出结果来辅助进行产品推荐。当然,深度神经网络模型可以先进行模型训练,并利用训练完成的模型进行产品推荐的使用。
模型的训练:
首先,训练用于金融理财产品推荐的深度神经网络模型。
可以根据产品购买的实际采集数据,构建模型训练的目标矩阵。该实际采集数据中可以包括用户对产品的购买数据,比如可以是用户对金融理财产品的实际购买记录,例如,用户A购买了基金J1,用户B购买了股票G1和基金J1,用户C购买了基金J2,等。根据上述的实际采集数据,可以构建目标矩阵,如下表1示例一种目标矩阵,但不局限于此:
表1 目标矩阵
  产品1(基金J1) 产品2(股票G1) 产品3(基金J2)
用户A 1 0 0
用户B 1 1 0
用户C 0 0 1
上述的表1中,目标矩阵可以包括用户对产品的购买选择值,所述购买选择值用于表示用户是否购买产品。示例性的,购买选择值可以包括“1”或者“0”,当数值是1时,表示用户购买了该产品;当数值是0时,表示用户未购买该产品。该目标矩阵可以作为深度神经网络模型的训练目标,当训练中的模型的输出结果与该训练目标之间的偏差越来越小,并且偏差达到预定阈值时,才结束模型的训练,并将训练结束的模型直接用于后续金融理财产品的推荐。
接着,对于目标矩阵中的各个用户(例如,用户A、用户B)和各个产品(例如,产品1、产品2),可以分别构建每个用户的用户特征矩阵,并分别构建各个产品的产品特征矩阵。并将构建的用户特征矩阵和产品特征矩阵输入待训练的机器学习模型,输出模型输出矩阵,该模型输出矩阵包括经过所述机器学习模型输出的各个购买选择值。在所述模型输出矩阵和目标矩阵的偏差达到预定阈值时,模型训练结束。
如下通过图1详细的描述模型训练的过程,该过程中描述了如何构建上述的用户特征矩阵、产品特征矩阵,以及如何将矩阵输入模型以训练模型的过程。
在步骤100中,获取目标用户关联的多领域信息,所述多领域信息包括:目标用户在待推荐产品的产品领域的购买数据和其他产品领域的购买数据。
本步骤中,目标用户是待推荐产品的用户,比如,想要向用户A推荐产品,但是尚不知道向该用户A推荐哪个产品,需要通过本实施例的推荐方法来确定要向用户A推荐的产品,那么该用户A可以称为目标用户。
需要说明的是,在模型训练中,这里的目标用户可以是目标矩阵中的用户,这些用户其实已经发生了实际的购买行为。而在后续的模型训练结束后的模型使用中,目标用户可以是待进行产品推荐的尚未对某些产品购买的用户。
以金融理财产品为例,待推荐产品的产品领域即金融理财产品,目标用户在该待推荐产品的产品领域的购买数据,例如可以包括:用户购买某个金融理财产品的交易金额。而其他产品领域的购买数据可以是非金融理财产品的购买,例如,可以是购买衣服,购买电饭煲等。所述的其他产品领域的购买数据可以是购买该其他领域产品的购买价格,比如,购买的衣服是200元,购买的电饭煲是350元。而不论是待推荐产品的产品领域或者其他产品领域的购买数据,都是由目标用户来进行购买而产生的数据。
此外,多领域信息也不局限于上述的不同产品领域的购买数据,也可以包括其他的信息。如下示例几种,包括但不局限于下面的信息:
例如,所述目标用户的用户属性信息。该用户属性信息可以是用户的性别、年龄、学历等。
例如,所述目标用户的关联用户在待推荐产品的产品领域的购买数据。其中,目标用户的关联用户可以是与目标用户具有好友关系、转账关系等。以好友关系为例,可以是目标用户的好友发生过的购买金融理财产品的数据,比如,用户A的好友用户a购买过某个金融理财产品,且交易金额是2万。
例如,目标用户的借贷行为数据。该借贷行为数据可以目标用户发生的借贷行为,借贷了某个品类的产品,且借贷的金额是多少。
在步骤102中,根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值。
本步骤中,可以基于步骤100中采集到的数据进行量化,转化为特征值。
如下的表2示例一种用户特征矩阵的形式:
表2 用户特征矩阵
Figure PCTCN2019076240-appb-000001
如上表2,在进行特征值的量化之前,可以首先进行产品的粗粒度处理。粗粒度处理是将数据集中较为细化的数据转化为概括性、综合度较高的数据。若对于一个产品品类的购买数据,在所述产品品类下购买的产品数量达到粗粒度处理条件,则将所述产品品类下的多个产品进行粗粒度处理。举例来说,假设其他产品领域的购买数据包括了购买衣服、电饭煲等多个比较细的品类,并且,在电饭煲这一个品类上目标用户就购买了博爱思DFB-B 0.8L、奥克斯AR-Y0801、洛贝LBF-091BM等小容量电饭煲处理为0-1L 非电脑迷你电饭煲,美的MB-WHS30C96、米家压力IH、松下SR-AE101-K等家用全自动智能电饭煲等多种电饭煲。那么如果在用户特征矩阵构建时,将这些产品划分的很细,比如,表2中的其他产品购买行为中,包括产品1、产品2、产品3等很多个产品,如上述的博爱思DFB-B 0.8L、奥克斯AR-Y0801、洛贝LBF-091BM等多个产品,那么将造成很大的计算压力。因此,粗粒度处理可以将粒度水平较细的特征维度汇总到一个相对粗糙的粒度水平。
例如,博爱思DFB-B 0.8L、奥克斯AR-Y0801、洛贝LBF-091BM等小容量电饭煲处理为0-1L非电脑迷你电饭煲,美的MB-WHS30C96、米家压力IH、松下SR-AE101-K等家用全自动智能电饭煲处理为3L-4L智能微电脑电饭煲。而是否对一个产品品类的购买数据进行粗粒度处理,可以设置粗粒度处理条件。例如,该条件可以是在所述产品品类下购买的产品数量达到一定的数量阈值,比如,在同一个产品品类下的产品数量达到了6个以上。而对于表2中的用户属性信息、社交关系、金融理财产品购买与借贷行为等特征维度,由于其特征维度少、信息含量高,可以不用进行粗粒度处理。
如下分别说明如何进行各个维度的特征值量化,其中需要说明的是,如下的量化方法仅是示例,实际实施中并不局限于此,可以按照其他量化标准执行:
1)对于目标用户及与目标用户建立社交关系用户的金融理财产品购买行为:
例如,可以根据购买的金融理财产品的交易金额,将交易金额合理划分为多个区间,比如表2中的“<P1”、“P1-P2”、“P2-P3”等多个区间。若用户购买该金融理财产品的金额处于该区间内,则标记为1;否则为0。
其中,表2中的社交关系栏,与目标用户具有关联关系的用户购买金融理财产品的购买数据,由于关联关系的用户可能是多个用户,可以先将与目标用户建立社交关系的所有用户的交易金额进行平均,根据平均值的金额进行标记。比如,如果平均值处于区间“P1-P2”,则可以在对应该区间的特征值标记1。
2)对于其他产品的购买行为:
如上所述的,其他产品的购买数据进行了粗粒度处理,处在同一粗糙粒度水平的可以有多个产品,并且这些产品的价格上可以具有相对较大的差异。此时可以以价格这一指标将该品类下所有产品合理划分到表2的各个属性区间,并将用户购买频次总体0-1标准化后的值作为其属性值,反应目标用户购买该品类下该价格区间内产品的频繁程度。
举例来说:假设目标用户在其他产品的购买行为中,在3L-4L智能微电脑电饭煲这 一品类下,购买了博爱思DFB-B 0.8L、奥克斯AR-Y0801、洛贝LBF-091BM等小容量电饭煲处理为0-1L非电脑迷你电饭煲,美的MB-WHS30C96、米家压力IH、松下SR-AE101-K等家用全自动智能电饭煲,即同一品类下购买了多种产品。那么可以根据这些产品各自的购买价格,查看在“<P1”区间内购买的产品数量,并将该数量作为对应该区间的特征值。比如,在所述的“<P1”区间内购买了3个产品,则特征值是3;在所述的“P1-P2”区间内购买了1个产品,则对应该区间的特征值可以是1。
3)对于目标用户的借贷行为:
例如,该借贷行为的量化与金融理财产品的量化类似,同样是将借贷金额合理划分为多个区间,若用户借贷该品类产品的金额处于该区间内,则标记为1;否则为0。
4)对于用户的基本信息:
例如,对于数值型变量,如年龄,可以按照与交易金额相同的方法进行划分。示例性的,18岁~25岁对应一个量化值,26岁~35岁对应一个量化值。
例如,对于类别型变量,如性别、学历,则可以将变量因子编码后标注。示例性的,本科学历可以对应一个量化值,研究生学历可以对应一个量化值。
在步骤104中,对于多个产品,获取购买所述产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的特征值,得到所述产品的产品特征矩阵。
本步骤中的产品是金融理财产品。本步骤可以构建产品特征矩阵,一个产品特征矩阵可以对应一个产品,该产品可以是目标矩阵中的各个产品。其中,产品特征矩阵的构建可以基于用户特征矩阵。
例如,以一个产品为例,购买该金融理财产品的有多个用户,每一个用户都构建了表2所示的用户特征矩阵。那么可以基于多个用户分别对应的多个用户特征矩阵,将特征值进行加权平均。
比如,以基本信息中的年龄为例,购买该产品的每个用户都有一个对应年龄的特征值,可以将多个用户的特征值进行加权平均,得到一个年龄对应的综合特征值。
又比如,以表2中的其他产品购买行为中的品类1为例,多个用户中的每个用户都有一个对应该品类1的特征值,可以将多个用户的特征值进行加权平均,得到一个品类1对应的综合特征值。
还可以看到,表2中的各个特征值对应着不同的特征值位置,比如,表2中的x1 对应的特征值位置是[行对应“P1-P2”区间,列对应“品类1”],而特征值x2对应的特征值位置是[行对应“P2-P3”区间,列对应“品类1”]。在构建产品特征矩阵时,可以将多个用户的用户特征矩阵中对应同一特征值位置的特征值进行加权平均,得到产品特征矩阵中对应所述特征值位置的特征值。
即,表2中的各个列,都可以将多个用户的特征值进行加权平均,最终得到能够反应出购买该产品的用户整体特征的产品特征矩阵。
其中,特征值加权平均时的权重的设置,可以根据实际业务情况确定。比如,若认为某个用户的特征值在反应用户整体特征时更加重要一些,就将其权重设置的更高一些。
在步骤106中,分别对用户特征矩阵和产品特征矩阵进行属性交互操作。
本步骤中可以进行用户特征矩阵和产品特征矩阵的属性交互操作。属性交互操作是将矩阵中不直接相关的属性间建立相关关系,先将构建的特征矩阵以属性列为单位随机排序生成多个新的特征矩阵,再将多个新的特征矩阵拼接生成属性交互后的特征矩阵。需要说明的是,该属性交互操作可以是一个可选的操作,执行属性交互操作后,能够更有效的发现不同特征之间的潜在关联,从而在后续利用机器学习模型感知用户偏好时也会更加准确。
特征矩阵的属性交互操作的原理可以参见图2所示:
如图2所述,其中的特征1、特征2、特征3等各个特征对应着不同的特征列。以用户特征矩阵为例,特征1可以是表1中的“金融理财产品的购买行为中的产品1”,特征15可以是表1中的“借贷行为中的借贷品类1”,即不同的特征对应着不同列。根据图2所示,相当于将表1中的不同列之间进行了随机的移动,以列为单位进行随机排序,而后拼接。
在步骤108中,分别将交互后的用户特征矩阵和产品特征矩阵输入机器学习模型,得到用户偏好向量和产品偏好向量。
本步骤中,深度神经网络包含两个并行的神经网络,其中一个是用户行为偏好的智能感知器,另一个是购买该产品的用户总体特征偏好的智能感知器,如图3所示。将属性交互和拼接后的特征矩阵作为并行神经网络的输入,比如,属性交互后的用户特征矩阵输入一个神经网络,属性交互后的产品特征矩阵输入另一个神经网络。
经过神经网络的卷积层、池化层及全链接操作后,神经网络可以分别得到用户偏好向量和产品偏好向量。其中,所述用户偏好向量可以用于表示用户在产品购买上的偏好, 相当于表示一个用户喜欢购买什么样的产品。而所述产品偏好向量可以用于表示购买产品特征矩阵对应的产品的用户特点,即相当于表示对于一个产品来说,具有什么特点的用户更倾向于购买该产品。
在步骤110中,根据模型输出的用户偏好向量和产品偏好向量,得到模型输出矩阵,所述模型输出矩阵包括经过机器学习模型输出的各个购买选择值。
例如,将一个用户对应的用户特征矩阵输入神经网络模型,得到用户偏好向量;将一个产品对应的产品特征矩阵输入神经网络模型,得到产品偏好向量。可以根据该用户偏好向量和产品偏好向量,得到一个购买选择值。比如,可以将上述的用户偏好向量和产品偏好向量求取内积,得到购买选择值,该选择值表示上述的用户购买所述产品的概率。
对于目标矩阵中的各个用户都可以构建一个用户特征矩阵,对于各个产品都可以分别构建对应的产品特征矩阵。按照上述的方法,可以得到其中的一个用户对一个产品的购买选择值。这些购买选择值可以构成模型输出矩阵,即该模型输出矩阵中包括的各个购买选择值是神经网络模型输出的数值。
而目标矩阵中包括的用户对产品的购买选择值,是根据实际采集数据得到,是用户实际发生的购买行为,目标矩阵是真实发生的用户与产品的相互选择矩阵。可以将目标矩阵作为神经网络模型的训练目标,随着模型的不断优化,神经网络模型的输出结果与实际的发生数值将越接近。
在步骤112中,在所述模型输出矩阵和目标矩阵的偏差达到预定阈值时,模型训练结束。
例如,可以设定模型输出矩阵和目标矩阵的偏差达到预定阈值时,结束模型的训练。所述的偏差达到预定阈值可以是偏差值小于或等于预定的阈值,即两者之间的偏差足够小。其中,模型输出矩阵和目标矩阵的偏差的衡量可以有多种方法,例如,偏差衡量可以使用均方根误差RMSE(Root Mean Square Error)或平均绝对误差MAE(Mean Absolute Deviation)。模型训练结束后,按照训练好的神经网络模型在预测用户和产品之间的相互选择概率时,将会预测的与实际情况接近,有很大的预测成功概率。
对训练结束的模型的使用:
假设已经将并列的两个神经网络训练结束,如下以一个例子来说明如何使用训练好的模型来判断给用户推荐何种产品将具有更高的成功率。
例如,假设当前要向用户Y推荐金融理财产品,待推荐的产品包括:产品C1、产品C2、产品C3等多个产品,那么要向用户Y推荐哪个金融理财产品会成功率更高,可以按照本例子的推荐方法执行。
可以先构建用户Y的用户特征矩阵,并分别构建产品C1、产品C2、产品C3等多个产品的产品特征矩阵。接着,将用户Y的用户特征矩阵和产品C1的产品特征矩阵分别输入并行的神经网络,得到用户偏好向量和产品偏好向量。并基于这两个向量得到用户Y对产品C1的选择评估值,所述选择评估值用于表示目标用户购买评估产品的概率。该选择评估值与上述提到的购买选择值的计算方式相同,只是采用两个名称是为了区分,购买选择值是在模型训练时计算的数值,选择评估值是在模型训练完的使用时计算的数值,用于作为是否向用户推荐产品的依据。
上述待推荐的产品C1、产品C2、产品C3等多个产品可以称为评估产品,即评估这些产品是否要推荐给用户Y。每个产品的产品特征矩阵和用户Y的用户特征矩阵之间都可以分别得到一个选择评估值。可以设定一个推荐阈值,在所述选择评估值大于预定的推荐阈值时,则确定将所述评估产品推荐给所述目标用户。举例来说,假设产品C1和用户Y的选择评估值是0.6,产品C2和用户Y的选择评估值是0.8,产品C3和用户Y的选择评估值是0.2,并假设推荐阈值是0.55,那么可以确定向用户Y推荐产品C1和产品C2,不推荐产品C3。
本例子的金融理财产品的个性化推荐方法,通过融合多个领域的用户行为数据与基本信息,并利用深度神经网络智能化感知用户与产品购买相关的偏好特征,帮助用户挑选合适的金融理财产品,有效缓解该行业所面临的交易数据稀疏与冷启动问题,有效提高了金融理财产品个性化推荐的准确度,为目标用户提供更准确的推荐服务,成为促进销售平台与用户间良性互动的有力措施。
为了实现上述方法,本说明书至少一个实施例还提供了一种产品推荐装置。如图4所示,该装置可以用于确定是否将待推荐产品推荐给目标用户,该装置可以包括:信息获取模块41、用户矩阵构建模块42、产品矩阵构建模块43、模型处理模块44、输出处理模块45和推荐确定模块46。
信息获取模块41,用于获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
用户矩阵构建模块42,用于根据所述多领域信息,构建所述目标用户的用户特征矩 阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
产品矩阵构建模块43,用于对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的产品特征矩阵;
模型处理模块44,用于分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
输出处理模块45,用于根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
推荐确定模块46,用于在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
在一个例子中,用户矩阵构建模块42,还用于:若对于一个产品品类的购买数据,在所述产品品类下购买的产品数量达到粗粒度处理条件,则将所述产品品类下的多个产品进行粗粒度处理。
在一个例子中,产品矩阵构建模块43,具体用于对所述多个用户的用户特征矩阵中对应同一特征值位置的特征值,进行加权平均,得到所述产品特征矩阵中对应所述特征值位置的特征值。
在一个例子中,模型处理模块44,还用于在分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型之前,分别对所述用户特征矩阵和产品特征矩阵进行属性交互操作;将交互后的用户特征矩阵和产品特征矩阵,输入所述机器学习模型。
上述实施例阐明的装置或模块,具体可以由计算机芯片或实体实现,或者由具有某种功能的产品来实现。一种典型的实现设备为计算机,计算机的具体形式可以是个人计算机、膝上型计算机、蜂窝电话、相机电话、智能电话、个人数字助理、媒体播放器、导航设备、电子邮件收发设备、游戏控制台、平板计算机、可穿戴设备或者这些设备中的任意几种设备的组合。
为了描述的方便,描述以上装置时以功能分为各种模块分别描述。当然,在实施本说明书一个或多个实施例时可以把各模块的功能在同一个或多个软件和/或硬件中实现。
上述图中所示流程中的各个步骤,其执行顺序不限制于流程图中的顺序。此外,各个步骤的描述,可以实现为软件、硬件或者其结合的形式,例如,本领域技术人员可以将其实现为软件代码的形式,可以为能够实现所述步骤对应的逻辑功能的计算机可执行指令。当其以软件的方式实现时,所述的可执行指令可以存储在存储器中,并被设备中的处理器执行。
例如,对应于上述方法,本说明书一个或多个实施例同时提供一种产品推荐设备,该设备可以包括处理器、存储器、以及存储在存储器上并可在处理器上运行的计算机指令,所述处理器通过执行所述指令,用于实现如下步骤:
获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的产品特征矩阵;
分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
还需要说明的是,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白,本说明书一个或多个实施例可提供为方法、系统或计算机程序产品。因此,本说明书一个或多个实施例可采用完全硬件实施例、完全软件实 施例或结合软件和硬件方面的实施例的形式。而且,本说明书一个或多个实施例可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本说明书一个或多个实施例可以在由计算机执行的计算机可执行指令的一般上下文中描述,例如程序模块。一般地,程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中实践本说明书一个或多个实施例,在这些分布式计算环境中,由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中,程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
本说明书中的各个实施例均采用递进的方式描述,各个实施例之间相同相似的部分互相参见即可,每个实施例重点说明的都是与其他实施例的不同之处。尤其,对于数据处理设备实施例而言,由于其基本相似于方法实施例,所以描述的比较简单,相关之处参见方法实施例的部分说明即可。
上述对本说明书特定实施例进行了描述。其它实施例在所附权利要求书的范围内。在一些情况下,在权利要求书中记载的动作或步骤可以按照不同于实施例中的顺序来执行并且仍然可以实现期望的结果。另外,在附图中描绘的过程不一定要求示出的特定顺序或者连续顺序才能实现期望的结果。在某些实施方式中,多任务处理和并行处理也是可以的或者可能是有利的。
以上所述仅为本说明书一个或多个实施例的较佳实施例而已,并不用以限制本说明书一个或多个实施例,凡在本说明书一个或多个实施例的精神和原则之内,所做的任何修改、等同替换、改进等,均应包含在本说明书一个或多个实施例保护的范围之内。

Claims (11)

  1. 一种产品推荐方法,所述方法用于确定是否将待推荐产品推荐给目标用户,所述方法包括:
    获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
    根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
    对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的产品特征矩阵;
    分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
    根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
    在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
  2. 根据权利要求1所述的方法,所述多领域信息,还包括如下至少一项:
    所述目标用户的关联用户在待推荐产品的产品领域的购买数据;
    所述目标用户的用户属性信息。
  3. 根据权利要求1所述的方法,所述构建目标用户的用户特征矩阵,包括:
    若对于一个产品品类的购买数据,在所述产品品类下购买的产品数量达到粗粒度处理条件,则将所述产品品类下的多个产品进行粗粒度处理。
  4. 根据权利要求1所述的方法,所述基于所述多个用户的用户特征矩阵中的特征值,得到所述待推荐产品对应的产品特征矩阵,包括:
    对所述多个用户的用户特征矩阵中对应同一特征值位置的特征值,进行加权平均,得到所述产品特征矩阵中对应所述特征值位置的特征值。
  5. 根据权利要求1所述的方法,在所述分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型之前,所述方法还包括:
    对所述机器学习模型进行训练,训练过程包括如下处理:
    根据产品购买的实际采集数据,构建模型训练的目标矩阵,所述实际采集数据包括 用户对产品的购买数据,所述目标矩阵包括:根据所述购买数据确定的用户对产品的购买选择值,所述购买选择值用于表示用户是否购买产品;
    对所述目标矩阵中的各个用户,分别构建每个用户的所述用户特征矩阵;
    对所述目标矩阵中的各个产品,分别构建各个产品的所述产品特征矩阵;
    将所述构建的用户特征矩阵和产品特征矩阵输入待训练的机器学习模型,并根据模型输出的用户偏好向量和产品偏好向量,得到模型输出矩阵,所述模型输出矩阵包括经过所述机器学习模型输出的各个购买选择值;
    在所述模型输出矩阵和目标矩阵的偏差达到预定阈值时,模型训练结束。
  6. 根据权利要求1所述的方法,所述分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型之前,所述方法还包括:
    分别对所述用户特征矩阵和产品特征矩阵进行属性交互操作;
    将交互后的用户特征矩阵和产品特征矩阵,输入所述机器学习模型。
  7. 一种产品推荐装置,所述装置用于确定是否将待推荐产品推荐给目标用户,所述装置包括:
    信息获取模块,用于获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
    用户矩阵构建模块,用于根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
    产品矩阵构建模块,用于对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的产品特征矩阵;
    模型处理模块,用于分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
    输出处理模块,用于根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
    推荐确定模块,用于在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
  8. 根据权利要求7所述的装置,
    所述用户矩阵构建模块,还用于:若对于一个产品品类的购买数据,在所述产品品 类下购买的产品数量达到粗粒度处理条件,则将所述产品品类下的多个产品进行粗粒度处理。
  9. 根据权利要求7所述的装置,
    所述产品矩阵构建模块,具体用于对所述多个用户的用户特征矩阵中对应同一特征值位置的特征值,进行加权平均,得到所述产品特征矩阵中对应所述特征值位置的特征值。
  10. 根据权利要求7所述的装置,
    所述模型处理模块,还用于在分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型之前,分别对所述用户特征矩阵和产品特征矩阵进行属性交互操作;将交互后的用户特征矩阵和产品特征矩阵,输入所述机器学习模型。
  11. 一种产品推荐设备,所述设备包括存储器、处理器,以及存储在存储器上并可在处理器上运行的计算机指令,所述处理器执行指令时实现以下步骤:
    获取所述目标用户关联的多领域信息,所述多领域信息包括:所述目标用户在所述待推荐产品的产品领域的购买数据和其他产品领域的购买数据;
    根据所述多领域信息,构建所述目标用户的用户特征矩阵,所述用户特征矩阵包括:根据所述多领域信息量化的多个特征值;
    对于一个所述待推荐产品,获取购买所述待推荐产品的多个用户的所述用户特征矩阵,并基于所述多个用户的用户特征矩阵中的所述特征值,得到所述待推荐产品对应的产品特征矩阵;
    分别将所述用户特征矩阵和产品特征矩阵输入预先训练的机器学习模型,得到用户偏好向量和产品偏好向量,所述用户偏好向量用于表示目标用户在产品购买上的偏好,所述产品偏好向量用于表示购买所述待推荐产品的用户特点;
    根据所述用户偏好向量和产品偏好向量,得到所述待推荐产品和所述目标用户之间的选择评估值,所述选择评估值用于表示所述目标用户购买所述待推荐产品的概率;
    在所述选择评估值大于预定的推荐阈值时,则确定将所述待推荐产品推荐给所述目标用户。
PCT/CN2019/076240 2018-05-22 2019-02-27 一种产品推荐方法和装置 WO2019223379A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810492957.0A CN108711075A (zh) 2018-05-22 2018-05-22 一种产品推荐方法和装置
CN201810492957.0 2018-05-22

Publications (1)

Publication Number Publication Date
WO2019223379A1 true WO2019223379A1 (zh) 2019-11-28

Family

ID=63869364

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/076240 WO2019223379A1 (zh) 2018-05-22 2019-02-27 一种产品推荐方法和装置

Country Status (3)

Country Link
CN (1) CN108711075A (zh)
TW (1) TWI740106B (zh)
WO (1) WO2019223379A1 (zh)

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110969491A (zh) * 2019-12-17 2020-04-07 深圳市梦网百科信息技术有限公司 一种基于网络路径的商品推送方法、系统及设备
CN111179041A (zh) * 2020-01-22 2020-05-19 中国铁道科学研究院集团有限公司电子计算技术研究所 乘意险产品推荐方法和装置
CN111210274A (zh) * 2020-01-06 2020-05-29 北京搜狐新媒体信息技术有限公司 一种广告推荐方法及系统
CN111241408A (zh) * 2020-01-21 2020-06-05 武汉轻工大学 推荐模型的构建系统及方法
CN111429232A (zh) * 2020-04-12 2020-07-17 中信银行股份有限公司 产品推荐方法、装置和电子设备及计算机可读存储介质
CN111814037A (zh) * 2019-12-31 2020-10-23 北京嘀嘀无限科技发展有限公司 数据处理方法、装置、存储介质及电子设备
CN111861635A (zh) * 2020-06-17 2020-10-30 北京邮电大学 一种针对商品分享时的好友推荐方法、装置及设备
CN112101980A (zh) * 2020-08-04 2020-12-18 北京思特奇信息技术股份有限公司 一种用户购买偏好分析的方法和系统
CN112115358A (zh) * 2020-09-14 2020-12-22 中国船舶重工集团公司第七0九研究所 一种利用知识图谱中多跳路径特征的个性化推荐方法
CN112200623A (zh) * 2020-09-27 2021-01-08 深圳市其乐游戏科技有限公司 产品推荐方法、装置、设备及存储介质
CN112632403A (zh) * 2020-12-24 2021-04-09 北京百度网讯科技有限公司 推荐模型的训练方法、推荐方法、装置、设备和介质
CN112988921A (zh) * 2019-12-13 2021-06-18 北京四维图新科技股份有限公司 地图信息变化的识别方法和装置
CN113222734A (zh) * 2021-05-21 2021-08-06 中国银行股份有限公司 银行金融资讯推荐系统及方法
CN113610580A (zh) * 2021-08-10 2021-11-05 平安科技(深圳)有限公司 产品推荐方法、装置、电子设备及可读存储介质
CN113763095A (zh) * 2020-11-27 2021-12-07 北京京东振世信息技术有限公司 信息推荐方法、装置和模型训练方法、装置
CN113781158A (zh) * 2021-08-23 2021-12-10 湖南大学 商品组合的推荐方法、装置及存储介质
CN114429384A (zh) * 2021-12-30 2022-05-03 杭州盟码科技有限公司 基于电商平台的产品智能推荐方法及系统
CN114885185A (zh) * 2022-04-28 2022-08-09 阿里巴巴(中国)有限公司 直播间推荐方法、内容推荐方法、终端及存储介质
CN115222461A (zh) * 2022-09-19 2022-10-21 杭州数立信息技术有限公司 一种智能营销精准推荐方法
WO2023051678A1 (zh) * 2021-09-29 2023-04-06 华为技术有限公司 一种推荐方法及相关装置
CN116738034A (zh) * 2022-10-10 2023-09-12 荣耀终端有限公司 一种信息推送方法及系统
CN116823382A (zh) * 2023-05-17 2023-09-29 南京邮电大学 一种基于大数据的产品推广方法

Families Citing this family (34)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108711075A (zh) * 2018-05-22 2018-10-26 阿里巴巴集团控股有限公司 一种产品推荐方法和装置
CN109711974A (zh) * 2018-11-20 2019-05-03 平安科技(深圳)有限公司 贷款产品自动匹配方法、装置、计算机设备和存储介质
CN109859061A (zh) * 2018-12-28 2019-06-07 阿里巴巴集团控股有限公司 一种关联用户的推荐方法和装置
CN110020910B (zh) * 2019-01-23 2022-05-24 创新先进技术有限公司 对象推荐方法和装置
CN110162714B (zh) * 2019-01-30 2023-11-14 腾讯科技(深圳)有限公司 内容推送方法、装置、计算设备和计算机可读存储介质
CN110059248B (zh) * 2019-03-21 2022-12-13 腾讯科技(深圳)有限公司 一种推荐方法、装置及服务器
CN111768218A (zh) * 2019-04-15 2020-10-13 北京沃东天骏信息技术有限公司 用于处理用户交互信息的方法和装置
CN110163662B (zh) * 2019-04-26 2024-04-05 创新先进技术有限公司 一种业务模型训练方法、装置及设备
CN110148057B (zh) * 2019-04-30 2021-09-14 德稻全球创新网络(北京)有限公司 一种企业内部结构性融资管理系统
CN110163723A (zh) * 2019-05-20 2019-08-23 深圳市和讯华谷信息技术有限公司 基于产品特征的推荐方法、装置、计算机设备及存储介质
CN110223137A (zh) * 2019-05-22 2019-09-10 平安科技(深圳)有限公司 产品组合推荐方法、装置、计算机设备和存储介质
CN110472145B (zh) * 2019-07-25 2022-11-29 维沃移动通信有限公司 一种内容推荐方法和电子设备
CN110619585A (zh) * 2019-08-16 2019-12-27 广州越秀金融科技有限公司 推荐数据的方法、装置、存储介质和处理器
CN110659410B (zh) * 2019-08-16 2024-04-02 平安科技(深圳)有限公司 智能推荐方法及装置、电子设备、计算机可读存储介质
CN110796512B (zh) * 2019-09-20 2023-09-05 平安科技(深圳)有限公司 产品信息推荐方法及装置、存储介质、计算机设备
CN110941758B (zh) * 2019-11-14 2022-09-16 支付宝(杭州)信息技术有限公司 推荐系统的合成特征生成方法和装置
CN111078997B (zh) * 2019-11-18 2023-08-22 中国银行股份有限公司 一种资讯推荐方法及装置
CN111461827B (zh) * 2020-03-31 2024-04-09 中国银行股份有限公司 产品评价信息的推送方法和装置
CN111598644B (zh) * 2020-04-01 2023-05-02 华瑞新智科技(北京)有限公司 一种物品推荐方法、装置以及介质
CN111652673B (zh) * 2020-05-09 2023-04-07 腾讯科技(深圳)有限公司 智能推荐方法、装置、服务器和存储介质
CN111598256B (zh) * 2020-05-18 2023-08-08 北京互金新融科技有限公司 目标客户的默搭购买行为的处理方法和装置
CN111626832B (zh) * 2020-06-05 2023-10-03 中国银行股份有限公司 产品推荐方法、装置及计算机设备
CN111538909A (zh) * 2020-06-22 2020-08-14 中国银行股份有限公司 一种信息推荐方法及装置
CN111914188A (zh) * 2020-08-19 2020-11-10 英华达(上海)科技有限公司 推荐目标用户的选择方法、系统、设备及存储介质
CN112132679A (zh) * 2020-09-30 2020-12-25 中国银行股份有限公司 产品推荐方法及装置
CN113763089A (zh) * 2020-10-27 2021-12-07 北京沃东天骏信息技术有限公司 物品推荐方法及装置、计算机可存储介质
CN112487292A (zh) * 2020-11-27 2021-03-12 杭州闪援车管家汽车服务有限公司 一种基于主动推荐的养车系统及方法
CN112561588A (zh) * 2020-12-18 2021-03-26 中国农业银行股份有限公司 一种产品推荐方法及装置
CN112529665A (zh) * 2020-12-18 2021-03-19 平安银行股份有限公司 基于组合模型的产品推荐方法、装置及计算机设备
CN112785376B (zh) * 2021-01-20 2022-08-19 电子科技大学 一种基于多任务学习的多领域推荐方法
CN113269610A (zh) * 2021-05-25 2021-08-17 索信达(北京)数据技术有限公司 银行产品的推荐方法、装置及存储介质
CN115880076A (zh) * 2022-04-20 2023-03-31 北京中关村科金技术有限公司 信托产品推荐方法、装置及存储介质
CN115496566B (zh) * 2022-11-16 2023-04-07 九州好礼(山东)电商科技有限公司 基于大数据的地区特产推荐方法及系统
CN116611896B (zh) * 2023-07-19 2023-10-24 山东省人工智能研究院 基于属性驱动解耦表征学习的多模态推荐方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103246672A (zh) * 2012-02-09 2013-08-14 中国科学技术大学 对用户进行个性化推荐的方法和装置
CN107330741A (zh) * 2017-07-07 2017-11-07 北京京东尚科信息技术有限公司 分品类电子券使用预测方法、装置及电子设备
CN107578270A (zh) * 2017-08-03 2018-01-12 中国银联股份有限公司 一种金融标签的构建方法、装置及计算设备
CN108711075A (zh) * 2018-05-22 2018-10-26 阿里巴巴集团控股有限公司 一种产品推荐方法和装置

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7308418B2 (en) * 2004-05-24 2007-12-11 Affinova, Inc. Determining design preferences of a group
CN101685458B (zh) * 2008-09-27 2012-09-19 华为技术有限公司 一种基于协同过滤的推荐方法和系统
TWI453684B (zh) * 2009-11-24 2014-09-21 Univ Nat Chiao Tung An Evaluation System and Method of Intelligent Mobile Service Commodity Application Information Retrieval Technology
TWI534732B (zh) * 2010-11-15 2016-05-21 Alibaba Group Holding Ltd Recommended information output method, system and server
US20150379609A1 (en) * 2014-06-30 2015-12-31 Kobo Incorporated Generating recommendations for unfamiliar users by utilizing social side information
CN106570008B (zh) * 2015-10-09 2020-03-27 阿里巴巴集团控股有限公司 推荐方法及装置
CN105184618A (zh) * 2015-10-20 2015-12-23 广州唯品会信息科技有限公司 新用户的商品个性化推荐方法及系统
CN106708883B (zh) * 2015-11-17 2020-09-29 阿里巴巴集团控股有限公司 推荐方法及装置
CN106384259A (zh) * 2016-09-08 2017-02-08 天津大学 一种融合社交信息的推荐系统解决方法
CN107273438B (zh) * 2017-05-24 2021-02-23 深圳大学 一种推荐方法、装置、设备及存储介质
CN107330115B (zh) * 2017-07-12 2020-04-28 广东工业大学 一种信息推荐方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103246672A (zh) * 2012-02-09 2013-08-14 中国科学技术大学 对用户进行个性化推荐的方法和装置
CN107330741A (zh) * 2017-07-07 2017-11-07 北京京东尚科信息技术有限公司 分品类电子券使用预测方法、装置及电子设备
CN107578270A (zh) * 2017-08-03 2018-01-12 中国银联股份有限公司 一种金融标签的构建方法、装置及计算设备
CN108711075A (zh) * 2018-05-22 2018-10-26 阿里巴巴集团控股有限公司 一种产品推荐方法和装置

Cited By (33)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112988921A (zh) * 2019-12-13 2021-06-18 北京四维图新科技股份有限公司 地图信息变化的识别方法和装置
CN110969491B (zh) * 2019-12-17 2023-08-29 深圳市梦网视讯有限公司 一种基于网络路径的商品推送方法、系统及设备
CN110969491A (zh) * 2019-12-17 2020-04-07 深圳市梦网百科信息技术有限公司 一种基于网络路径的商品推送方法、系统及设备
CN111814037B (zh) * 2019-12-31 2024-04-26 北京嘀嘀无限科技发展有限公司 数据处理方法、装置、存储介质及电子设备
CN111814037A (zh) * 2019-12-31 2020-10-23 北京嘀嘀无限科技发展有限公司 数据处理方法、装置、存储介质及电子设备
CN111210274A (zh) * 2020-01-06 2020-05-29 北京搜狐新媒体信息技术有限公司 一种广告推荐方法及系统
CN111241408A (zh) * 2020-01-21 2020-06-05 武汉轻工大学 推荐模型的构建系统及方法
CN111241408B (zh) * 2020-01-21 2023-05-30 武汉轻工大学 推荐模型的构建系统及方法
CN111179041A (zh) * 2020-01-22 2020-05-19 中国铁道科学研究院集团有限公司电子计算技术研究所 乘意险产品推荐方法和装置
CN111429232A (zh) * 2020-04-12 2020-07-17 中信银行股份有限公司 产品推荐方法、装置和电子设备及计算机可读存储介质
CN111861635A (zh) * 2020-06-17 2020-10-30 北京邮电大学 一种针对商品分享时的好友推荐方法、装置及设备
CN111861635B (zh) * 2020-06-17 2022-10-25 北京邮电大学 一种针对商品分享时的好友推荐方法、装置及设备
CN112101980A (zh) * 2020-08-04 2020-12-18 北京思特奇信息技术股份有限公司 一种用户购买偏好分析的方法和系统
CN112101980B (zh) * 2020-08-04 2024-04-02 北京思特奇信息技术股份有限公司 一种用户购买偏好分析的方法和系统
CN112115358B (zh) * 2020-09-14 2024-04-16 中国船舶重工集团公司第七0九研究所 一种利用知识图谱中多跳路径特征的个性化推荐方法
CN112115358A (zh) * 2020-09-14 2020-12-22 中国船舶重工集团公司第七0九研究所 一种利用知识图谱中多跳路径特征的个性化推荐方法
CN112200623A (zh) * 2020-09-27 2021-01-08 深圳市其乐游戏科技有限公司 产品推荐方法、装置、设备及存储介质
CN113763095A (zh) * 2020-11-27 2021-12-07 北京京东振世信息技术有限公司 信息推荐方法、装置和模型训练方法、装置
CN113763095B (zh) * 2020-11-27 2023-09-26 北京京东振世信息技术有限公司 信息推荐方法、装置和模型训练方法、装置
CN112632403B (zh) * 2020-12-24 2024-04-09 北京百度网讯科技有限公司 推荐模型的训练方法、推荐方法、装置、设备和介质
CN112632403A (zh) * 2020-12-24 2021-04-09 北京百度网讯科技有限公司 推荐模型的训练方法、推荐方法、装置、设备和介质
CN113222734A (zh) * 2021-05-21 2021-08-06 中国银行股份有限公司 银行金融资讯推荐系统及方法
CN113610580A (zh) * 2021-08-10 2021-11-05 平安科技(深圳)有限公司 产品推荐方法、装置、电子设备及可读存储介质
CN113610580B (zh) * 2021-08-10 2023-09-19 平安科技(深圳)有限公司 产品推荐方法、装置、电子设备及可读存储介质
CN113781158A (zh) * 2021-08-23 2021-12-10 湖南大学 商品组合的推荐方法、装置及存储介质
WO2023051678A1 (zh) * 2021-09-29 2023-04-06 华为技术有限公司 一种推荐方法及相关装置
CN114429384A (zh) * 2021-12-30 2022-05-03 杭州盟码科技有限公司 基于电商平台的产品智能推荐方法及系统
CN114429384B (zh) * 2021-12-30 2022-12-09 杭州盟码科技有限公司 基于电商平台的产品智能推荐方法及系统
CN114885185A (zh) * 2022-04-28 2022-08-09 阿里巴巴(中国)有限公司 直播间推荐方法、内容推荐方法、终端及存储介质
CN115222461A (zh) * 2022-09-19 2022-10-21 杭州数立信息技术有限公司 一种智能营销精准推荐方法
CN116738034A (zh) * 2022-10-10 2023-09-12 荣耀终端有限公司 一种信息推送方法及系统
CN116823382A (zh) * 2023-05-17 2023-09-29 南京邮电大学 一种基于大数据的产品推广方法
CN116823382B (zh) * 2023-05-17 2024-01-05 南京邮电大学 一种基于大数据的产品推广方法

Also Published As

Publication number Publication date
TW202004618A (zh) 2020-01-16
CN108711075A (zh) 2018-10-26
TWI740106B (zh) 2021-09-21

Similar Documents

Publication Publication Date Title
WO2019223379A1 (zh) 一种产品推荐方法和装置
US10354184B1 (en) Joint modeling of user behavior
US8706716B2 (en) Iterative and dynamic search of publicly available data based on augmentation of search terms and validation of data relevance
US20090276368A1 (en) Systems and methods for providing personalized recommendations of products and services based on explicit and implicit user data and feedback
US20130138555A1 (en) System and method of interpreting results based on publicly available data
US20130159288A1 (en) Information graph
WO2020238229A1 (zh) 交易特征生成模型的训练、交易特征的生成方法和装置
Alexandrova Connecting economic models to the real world: Game theory and the FCC spectrum auctions
US20210117889A1 (en) Co-operative resource pooling system
US11200593B2 (en) Predictive recommendation system using tiered feature data
CN110647683B (zh) 一种信息推荐方法、装置
US11182804B2 (en) Segment valuation in a digital medium environment
WO2020221022A1 (zh) 业务对象推荐方法
US20190087859A1 (en) Systems and methods for facilitating deals
CN108734587A (zh) 金融产品的推荐方法及终端设备
US20230093756A1 (en) Systems and methods for generating recommendations
WO2023000491A1 (zh) 一种应用推荐方法、装置、设备及计算机可读存储介质
CN112598472A (zh) 产品推荐方法、装置、系统、介质和程序产品
US10049372B2 (en) Search and rank organizations
CN111754287A (zh) 物品筛选方法、装置、设备和存储介质
CN107305662A (zh) 识别违规账户的方法及装置
WO2020150597A1 (en) Systems and methods for entity performance and risk scoring
CN111639989B (zh) 商品推荐方法和可读存储介质
CN110427565B (zh) 一种基于协同过滤的物品推荐方法、智能终端及存储介质
CN113792220A (zh) 一种目标对象推荐方法、装置、计算机设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19807689

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19807689

Country of ref document: EP

Kind code of ref document: A1