CN116361542A

CN116361542A - Product recommendation method, device, computer equipment and storage medium

Info

Publication number: CN116361542A
Application number: CN202310066751.2A
Authority: CN
Inventors: 苏晨晟; 张彬; 刘映楷; 帅翡芍
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2023-01-13
Filing date: 2023-01-13
Publication date: 2023-06-30

Abstract

The application relates to a product recommendation method, a product recommendation device, computer equipment, a storage medium and a computer program product, and relates to the technical field of big data. The method comprises the following steps: predicting the resource loss information of the user through a pre-trained resource loss information prediction model, and confirming the user type of the user according to the resource loss information of the user; under the condition that the user type of the user is a resource loss user, acquiring index information of the user under a plurality of preset clustering indexes; inputting index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering to obtain target user types of the user; and recommending the target product associated with the target user category to the user. By adopting the method, the accuracy of product recommendation can be improved.

Description

Product recommendation method, device, computer equipment and storage medium

Technical Field

The present application relates to the field of big data technologies, and in particular, to a product recommendation method, apparatus, computer device, storage medium, and computer program product.

Background

With the development of big data technology, product recommendation technology is presented. In the existing product recommendation method, the users are generally classified by analyzing behavior data of the users by experts, and then corresponding products are recommended to the users based on classification results of the users.

However, the conventional product recommendation method is easily subjectively influenced by expert experience, so that the classification result of the user is inaccurate, and a proper product cannot be recommended to the user, thereby resulting in lower accuracy of product recommendation.

Disclosure of Invention

In view of the foregoing, it is desirable to provide a product recommendation method, apparatus, computer device, computer-readable storage medium, and computer program product that can improve accuracy of product recommendation.

In a first aspect, the present application provides a product recommendation method. The method comprises the following steps:

predicting the resource loss information of a user through a pre-trained resource loss information prediction model, and confirming the user type of the user according to the resource loss information of the user;

acquiring index information of the user under a plurality of preset clustering indexes under the condition that the user type of the user is a resource loss user;

inputting index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering to obtain a target user category of the user;

and recommending the target product associated with the target user category to the user.

In one embodiment, the predicting the resource loss information of the user through the pre-trained resource loss information prediction model, and determining the user type of the user according to the resource loss information of the user includes:

acquiring resource information of a user;

performing resource loss prediction processing on the resource information of the user through the pre-trained resource loss information prediction model to obtain the resource loss information of the user;

and under the condition that the resource loss information of the user meets a preset resource loss threshold, confirming the user type of the user as a resource loss user.

In one embodiment, the pre-trained resource loss information prediction model is trained by:

sample resource information of each prediction sample user is obtained;

for each prediction sample user, dividing the sample resource information into observation period resource information, interval period resource information and expression period resource information according to the time sequence of the sample resource information of the prediction sample user;

inputting the observation period resource information, the interval period resource information and the performance period resource information into a resource loss information prediction model to be trained to obtain performance period resource prediction information of the prediction sample user;

And training the resource loss information prediction model to be trained according to the difference between the expression period resource prediction information and the expression period resource information to obtain a trained resource loss information prediction model, and taking the trained resource loss information prediction model as the pre-trained resource loss information prediction model.

In one embodiment, the pre-trained cluster model is trained by:

acquiring sample index information of a clustered sample user under the plurality of preset cluster indexes;

clustering sample index information of the clustered sample users under the plurality of preset clustering indexes through a clustering model to be trained to obtain sample user categories of the clustered sample users;

confirming category difference information among the sample user categories;

and training the clustering model to be trained under the condition that the category difference information is smaller than a preset difference threshold value to obtain a training-completed clustering model which is used as the pre-training clustering model.

In one embodiment, the obtaining the index information of the user under a plurality of preset cluster indexes includes:

acquiring index information of the user under a plurality of basic characteristic indexes;

Confirming correlation information among the plurality of basic feature indexes;

fusing the basic characteristic indexes of which the corresponding correlation information meets a preset correlation threshold value to obtain a plurality of preset clustering indexes;

and aiming at each preset clustering index, carrying out fusion processing on index information of the user under the basic characteristic index corresponding to the preset clustering index to obtain the index information of the user under the preset clustering index.

In one embodiment, before recommending the target product associated with the target user category to the user, the method further comprises:

confirming each user category of the pre-trained cluster model;

acquiring characteristic information of each user category; the characteristic information of the user category is obtained according to sample index information of the associated sample user of the user category under the plurality of preset clustering indexes; the associated sample users of the user categories are the clustered sample users used for training the pre-trained cluster model, and the corresponding sample user categories are the clustered sample users of the user categories;

aiming at each user category, according to the characteristic information of the user category, confirming the association relationship between the user category and the product;

The recommending the target product associated with the target user category to the user comprises the following steps:

inquiring the association relation, and confirming a product associated with the target user category as a target product;

recommending the target product to the user.

In one embodiment, for each user category, according to the feature information of the user category, the determining the association relationship between the user category and the product includes:

predicting the product consumption behavior of the associated sample user of the user category according to the characteristic information of the user category aiming at each user category;

confirming a product associated with the product consumption behavior of the associated sample user;

and according to the user category and the product associated with the product consumption behavior of the association sample user, confirming the association relationship between the user category and the product.

In a second aspect, the present application further provides a product recommendation device. The device comprises:

the loss information prediction module is used for predicting the resource loss information of the user through a pre-trained resource loss information prediction model, and confirming the user type of the user according to the resource loss information of the user;

The index information acquisition module is used for acquiring index information of the user under a plurality of preset clustering indexes under the condition that the user type of the user is a resource loss user;

the user category clustering module is used for inputting index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering to obtain target user categories of the user;

and the target product recommending module is used for recommending the target product associated with the target user category to the user.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor which when executing the computer program performs the steps of:

In a fourth aspect, the present application also provides a computer-readable storage medium. The computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:

In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of:

The product recommending method, the device, the computer equipment, the storage medium and the computer program product are characterized in that firstly, the resource loss information of a user is predicted through a pre-trained resource loss information prediction model, and the user type of the user is confirmed according to the resource loss information of the user; then, under the condition that the user type of the user is a resource loss user, acquiring index information of the user under a plurality of preset clustering indexes; inputting index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering to obtain target user types of the user; and finally recommending the target product associated with the target user category to the user. Therefore, the user type can be accurately confirmed as the target guest group needing to be recommended by the product by the pre-trained resource loss information prediction model; and then confirming the user category of the user through index information of the user under a plurality of preset clustering indexes and a pre-trained clustering model, confirming a target product required to be recommended to the user according to the user category, and further completing product recommendation of the user. The product recommendation method based on the process can not be interfered by subjective influences, can realize accurate classification of the users and product recommendation for the users, improves the accuracy of product recommendation, and can develop personalized services for the users and effectively save the resource loss users.

Drawings

FIG. 1 is a flow chart of a product recommendation method according to an embodiment;

FIG. 2 is a flowchart illustrating a training step of a pre-trained resource outage information prediction model according to one embodiment;

FIG. 3 is a flow diagram of training steps for a pre-trained cluster model in one embodiment;

FIG. 4 is a schematic diagram of clustering results of clustering sample users in one embodiment;

FIG. 5 is a flowchart of a product recommendation method according to another embodiment;

FIG. 6 is a block diagram of a product recommendation device in one embodiment;

fig. 7 is an internal structural diagram of a computer device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.

In an exemplary embodiment, as shown in fig. 1, a product recommendation method is provided, and the method is applied to a server for illustration in this embodiment; it will be appreciated that the method may also be applied to a terminal, and may also be applied to a system comprising a server and a terminal, and implemented by interaction between the server and the terminal. The server can be realized by an independent server or a server cluster formed by a plurality of servers; the terminal can be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things equipment and portable wearable equipment, and the internet of things equipment can be smart speakers, smart televisions, smart air conditioners, smart vehicle-mounted equipment and the like. In this embodiment, the portable wearable device may be a smart watch, a smart bracelet, a headset, etc., the method includes the following steps:

Step S102, predicting the resource loss information of the user through a pre-trained resource loss information prediction model, and confirming the user type of the user according to the resource loss information of the user.

The resource loss information prediction model is trained in advance according to the resource loss information of the predicted sample user. The resource loss information is used for representing whether the user continues to select the service of the enterprise in a period of the future, and taking a product recommendation scene of a bank as an example, the resource loss information of the user can be the resource loss rate of the user; the resources of the user include the user's assets, the user's investment in financial products, the user's investment in investment products, etc.

Specifically, the server predicts the resource information such as the assets of the user, the investment of the user on financial products and the like through a pre-trained resource loss information prediction model to obtain the resource loss rate of the user so as to represent the resource loss information; and then, according to the resource loss information, confirming the user type of the user.

Step S104, under the condition that the user type of the user is a resource loss user, index information of the user under a plurality of preset clustering indexes is obtained.

The resource loss user refers to a user who can not select the service of the enterprise any more in a certain probability within a period of time. The preset clustering index is each index for clustering users, and in the application, the preset clustering index at least comprises a loss rate index, a user value index, a user contribution index and a channel preference index.

Specifically, the server predicts that the user can not select the service of the enterprise any more in a period of time according to the resource loss information of the user, confirms the user type of the user as the resource loss user, and obtains index information of the user under a plurality of preset cluster indexes such as a loss rate index, a user value index, a user contribution index, a channel preference index and the like.

And S106, inputting index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering to obtain the target user category of the user.

The clustering model is trained in advance, and is obtained through training of a clustering sample user. The target user category of the user is a clustering result obtained by the user according to a pre-trained clustering model. Taking a product recommendation scene of a bank as an example, users can be at least divided into a pre-support constraint type user, a potential periodic type user, a high-end necessary-keeping type user and a deposit dependent type user according to a pre-trained clustering model.

Specifically, the server acquires index information of the user under a plurality of preset clustering indexes including at least a churn rate index, a user value index, a user contribution index, a channel preference index and the like, and inputs the index information into a pre-trained clustering model for clustering processing to obtain a target user category of the user.

For example, the server performs clustering processing on index information of the user under a plurality of preset clustering indexes according to a pre-trained clustering model to obtain a clustering result of the user as a potential periodic user, and then the target user category of the user is the potential periodic user.

Step S108, recommending the target product associated with the target user category to the user.

And the target product associated with the target user category is obtained by confirming the association relationship between the user category and the product of the enterprise according to the pre-trained clustering model. In a product recommendation scenario of a bank, products of an enterprise include financial products, investment products, deposit products, pre-paid products, and the like.

Specifically, the server queries the association relationship between the user category and the product, confirms the target product associated with the target user category of the user, and recommends the target product to the user.

For example, if the user is a pre-support constraint type user, the main business of the user in the bank is a pre-support business, so that the corresponding target product is a pre-support product; the server can put relevant recommendation information of the pre-support products into the user, so that the use frequency of the user to the bank products is improved.

In the product recommendation method, a server predicts the resource loss information of a user through a pre-trained resource loss information prediction model, and confirms the user type of the user according to the resource loss information of the user; then, under the condition that the user type of the user is a resource loss user, acquiring index information of the user under a plurality of preset clustering indexes; inputting index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering to obtain target user types of the user; and finally recommending the target product associated with the target user category to the user. In this way, the server can accurately confirm the user type as the user of the resource loss user as a target guest group needing product recommendation through a pre-trained resource loss information prediction model; and then confirming the user category of the user through index information of the user under a plurality of preset clustering indexes and a pre-trained clustering model, confirming a target product required to be recommended to the user according to the user category, and further completing product recommendation of the user. The product recommendation method based on the process can not be interfered by subjective influences, can realize accurate classification of users and product recommendation for the users, improves the accuracy of product recommendation, and further can effectively save the users with lost resources.

In an exemplary embodiment, the step S102 predicts the resource loss information of the user through a pre-trained resource loss information prediction model, and confirms the user type of the user according to the resource loss information of the user, which specifically includes the following contents: acquiring resource information of a user; performing resource loss prediction processing on the resource information of the user through a pre-trained resource loss information prediction model to obtain the resource loss information of the user; and under the condition that the resource loss information of the user meets the preset resource loss threshold value, determining the user type of the user as the resource loss user.

In the product recommendation scene of the bank, the resource loss user is defined as a user with 50% or more reduction of the resource information of the expression period relative to the resource information of the last month of the observation period, so that the preset resource loss threshold is 50%. It can be understood that setting the preset resource loss threshold supports adjustment according to different service scenarios.

The resource information of the user may be classified into observation period resource information (1 month to 3 months), interval period resource information (4 months), and presentation period resource information (5 months to 7 months) according to the time sequence. Assuming that it is necessary to predict whether a user will experience a resource loss during the presentation period, it is necessary to compare the resource information of 5 months to 7 months with the resource information of 3 months during the observation period.

Specifically, the server acquires observation period resource information and interval period resource information of a user, predicts the observation period resource information and the interval period resource information of the user according to a pre-trained resource loss information prediction model to obtain a prediction result of the resource information of the user in a presentation period, and then confirms the resource loss information of the user in the presentation period by comparing the prediction result of the resource information based on the presentation period with the interval period resource information; if the resource loss information of the user in the expression period is more than 50%, the server confirms the user as the resource loss user.

For example, assuming an observation period of 1 month to 3 months, an interval period of 4 months, and a presentation period of 5 months to 7 months, the user's resource information is characterized by a lunar daily financial asset; the server firstly acquires the month and day uniform financial assets of the user in 1 month to 4 months, inputs data into a pre-trained resource loss information model, predicts the month and day uniform financial assets of the user in 5 months to 7 months through the pre-trained resource loss information model, and then confirms the reduction condition of the month and day uniform financial assets of the user in 5 months to 7 months relative to the month and day uniform financial assets of the user in 3 months as the resource loss information of the user in the expression period.

In this embodiment, the server predicts the resource loss condition of the user in the expression period through a pre-trained resource loss information prediction model and the resource information of the user in the observation period and the interval period, and screens out the user to be saved based on the prediction result of the resource loss condition, thereby determining the target guest group in the subsequent product recommendation.

In an exemplary embodiment, as shown in fig. 2, the pre-trained resource loss information prediction model in step S102 is trained in the following manner:

step S202, sample resource information of each prediction sample user is obtained.

In step S204, for each predicted sample user, the sample resource information is divided into observation period resource information, interval period resource information, and performance period resource information according to the time sequence of the sample resource information of the predicted sample user.

And S206, inputting the observation period resource information, the interval period resource information and the expression period resource information into a resource loss information prediction model to be trained to obtain expression period resource prediction information of a prediction sample user.

Step S208, training the resource loss information prediction model to be trained through the difference between the expression period resource prediction information and the expression period resource information to obtain a trained resource loss information prediction model, and taking the trained resource loss information prediction model as a pre-trained resource loss information prediction model.

Wherein the difference between the performance prediction information and the performance resource information is obtained according to a loss function between the performance prediction information and the performance resource information.

The resource loss information prediction model to be trained is a deep neural network consisting of an input layer, three hidden layers and an output layer.

Specifically, the server firstly acquires sample resource information of each predicted sample user, and divides the sample resource information into observation period resource information, interval period resource information and presentation period resource information according to a time sequence; then, predicting the observation period resource information and the interval period resource information through a resource loss information prediction model to be trained to obtain the expression period resource prediction information of a prediction sample user in the expression period; and then the server obtains a loss function between the performance period resource prediction information and the performance period resource information through the resource loss information prediction model to be trained, so as to determine the difference between the performance period resource prediction information and the performance period resource information, and under the condition that the difference between the performance period resource prediction information and the performance period resource information is larger than a preset resource information difference threshold value, the server adjusts parameters in the resource loss information prediction model to be trained, and retrains the resource loss information prediction model to be trained again based on the observation period resource information and the interval period resource information of the prediction sample user until the difference between the performance period resource prediction information and the performance period resource information is smaller than the preset resource information difference threshold value, so as to obtain the trained resource loss information prediction model, namely the pre-trained resource loss information prediction model.

In this embodiment, the server obtains the resource information prediction result of the predicted sample user in the performance period by predicting the resource information of the sample user in the observation period and the interval period, and trains the resource loss information prediction model to be trained according to the resource information prediction result of the predicted sample user in the performance period and the actual resource information of the predicted sample user in the performance period, thereby obtaining a model capable of accurately predicting the resource loss information, and providing an implementation basis for accurately predicting the resource loss information of the user in the follow-up process and determining the user to be recommended for the product recommendation process.

In an exemplary embodiment, as shown in fig. 3, in step S106, the pre-trained cluster model is trained by:

step S302, sample index information of a clustered sample user under a plurality of preset cluster indexes is obtained.

Step S304, clustering processing is carried out on sample index information of the clustered sample users under a plurality of preset clustering indexes through a clustering model to be trained, and sample user categories of the clustered sample users are obtained.

Step S306, confirming category difference information among the sample user categories.

Step S308, training the clustering model to be trained under the condition that the category difference information is smaller than a preset difference threshold value to obtain a training-completed clustering model which is used as a pre-training clustering model.

The clustering sample users are sample users of the predicted sample users, and the user types are resource loss users.

The category difference information among the sample user categories is determined through a Caliski-Harabasz index; through the Calinski-Harabasz index, the clustering effect of the clustering model to be trained can be evaluated, and the higher the Calinski-Harabasz score is, the higher the degree of distinction between the user classes of each sample is, and the better the clustering effect is.

Specifically, the server acquires sample index information of each clustered sample user under a plurality of preset cluster indexes including at least a loss rate index, a user value index, a user contribution index, a channel preference index and the like, and performs clustering processing on each clustered sample user according to the sample index information of each clustered sample user under the plurality of preset cluster indexes through a clustering model to be trained, so as to obtain sample user types of each clustered sample user; the method comprises the steps that the number of sample user categories of a clustering model to be trained needs to be initialized; then, based on the Calinski-Harabasz index, confirming the ratio of the inter-cluster distance to the intra-cluster distance of each sample user category, namely category difference information among each sample user category, and taking the ratio as the Calinski-Harabasz score of the clustering model to be trained; under the condition that the Calinski-Harabasz score of the clustering model to be trained is smaller than a preset difference threshold, the distinction degree among all sample user categories is less obvious, and the number of sample user categories of the clustering model to be trained needs to be adjusted, so that the clustering model to be trained needs to be trained again based on the clustering sample users until the Calinski-Harabasz score of the clustering model to be trained is larger than the preset difference threshold, and the trained clustering model is obtained and used as a pre-trained clustering model.

In the training process of the clustering model, when the number of sample user categories is 4, the Calinski-Harabasz score is the largest, namely, when the number of clusters is 4, the clustering effect of the clustering model is the best; thus, the clustering model with the sample user category number of 4 is used as a pre-trained clustering model.

In this embodiment, the server obtains sample user categories of the clustered sample users through index information of the clustered sample users under a plurality of preset cluster indexes, and confirms the Calinski-Harabasz score of the clustering model to be trained based on each sample user category, so that the clustering model to be trained is trained, and further, a clustering model with high clustering effect distinction degree and accurate clustering result is obtained, a basis is provided for subsequently confirming the user category of the user to be recommended, and product recommendation is performed based on the user category, so that accuracy of product recommendation is improved.

In an exemplary embodiment, in step S104, index information of the user under a plurality of preset cluster indexes is obtained, which specifically includes the following contents: acquiring index information of a user under a plurality of basic characteristic indexes; confirming correlation information among a plurality of basic characteristic indexes; fusing the basic characteristic indexes of which the corresponding correlation information meets the preset correlation threshold value to obtain a plurality of preset clustering indexes; and aiming at each preset clustering index, carrying out fusion processing on index information of the user under the basic characteristic index corresponding to the preset clustering index to obtain the index information of the user under the preset clustering index.

The basic characteristic indexes at least comprise a loss rate index, a life cycle index, a product holding condition index, a income level index, a channel preference index and a contribution type index.

The correlation information among the plurality of basic characteristic indexes is used for representing the correlation among every two basic characteristic indexes; the correlation information may be determined by, for example, pearson correlation coefficient algorithm, or by determining the weight of each basic feature index by random gradient descent algorithm.

The loss rate index information is obtained according to the resource loss information of the user; the life cycle index information is obtained according to the age of the user; the product holding quantity index information is obtained according to the quantity of the held products of the user; the income level index information is obtained according to the income condition of the user in a period of time; the channel preference index xinxi is obtained according to the transaction proportion situation of the user in each channel (such as counter, online, self-service equipment and the like); the contribution type information is derived from the user's contribution type to the business, such as a pre-pay contribution, a deposit contribution, an investment contribution, and a credit card contribution.

Specifically, the server acquires index information of a user under a plurality of basic characteristic indexes including at least a churn rate index, a life cycle index, a product holding condition index, a income level index, a channel preference index, a contribution type index and the like; then the server confirms the correlation coefficient between each basic characteristic index, namely the correlation information; then fusing two basic characteristic indexes with the correlation coefficient larger than a preset correlation threshold, for example fusing a life cycle index and a income level index to obtain a user value index, fusing a product holding quantity index and a contribution type index to obtain a user contribution index, and obtaining a plurality of preset clustering indexes at least including a loss rate index, a user value index, a user contribution index and a channel preference index; finally, the server performs fusion processing on the index information of the user under the basic characteristic index corresponding to the preset clustering index aiming at the preset clustering index obtained through fusion to obtain index information of the user under the preset clustering index, for example, fusion of life cycle index information and income level index information to obtain user value index information, fusion of the holding quantity index information and contribution type index information of the product to obtain user contribution index information, and then the index information of the user under a plurality of preset clustering indexes can be obtained by combining the loss rate index information and channel preference index information of the user.

In this embodiment, the server fuses the multiple basic feature indexes and the index information of the user under the multiple basic feature indexes to obtain multiple preset clustering indexes for clustering the user and the index information of the user under the multiple preset clustering indexes, so that a large amount of and disordered user information can be divided and fused, the dimension of the information for determining the user category is reduced, and redundant information in the clustering process is reduced, thereby realizing accurate clustering of the user and further improving the accuracy of product recommendation.

In an exemplary embodiment, the step S108, before recommending the target product associated with the target user category to the user, specifically includes the following: confirming each user category of the pre-trained clustering model; acquiring characteristic information of each user category; and aiming at each user category, according to the characteristic information of the user category, confirming the association relationship between the user category and the product.

Step S108, recommending the target product associated with the target user category to the user, specifically including the following contents: inquiring the association relation, and confirming the product associated with the target user category as a target product; and recommending the target product to the user.

The characteristic information of the user category is obtained according to sample index information of the associated sample user of the user category under a plurality of preset clustering indexes; the associated sample users of the user categories are the clustered sample users used for training the pre-trained cluster model, and the corresponding sample user categories are the clustered sample users of the user categories. For example, as shown in fig. 4, which is a schematic diagram of a clustering result of clustering sample users, it is assumed that the clustering sample users are classified into 4 classes by 4 sample user classes (class a, class B, class C and class D) of a pre-trained clustering model; then, aiming at the class A sample user category, the corresponding characteristic information is obtained according to the sample index information of 6 clustered sample users in the area A of FIG. 4, wherein the clustered result is the clustered sample users of the class A in all clustered sample users; the 6 clustered sample users in region a of fig. 4 are associated sample users of class a sample user class.

Specifically, after obtaining a pre-trained clustering model, the server also needs to confirm the characteristic information of each user category according to the sample index information of the clustered sample users of which the sample user categories belong to each user category in the clustering result of the pre-trained clustering model; then according to the characteristic information of each user category, confirming the association relation between each user category and the product; and finally, inquiring the association relation between the user category and the product by the server, recommending the product recommendation associated with the target user category of the user to the user as a target product, and thus realizing accurate recommendation to the user.

For example, the user categories obtained by the server according to the pre-trained clustering model are respectively a pre-support constraint user, a potential periodic user, a high-end necessary-keeping user and a deposit dependent user; taking a pre-support constraint type user as an example, the server obtaining the characteristic information of the pre-support constraint type user according to the sample index information of the clustered sample users taking the sample user category as the pre-support constraint type user comprises: young people who have been working for some time have the lowest churn rate, have a higher number of held products, a higher revenue level, the highest contribution to the pre-pay, less to store consumption, and prefer on-line channels, often because the pre-pay amount is outstanding. Based on the characteristic information, the server can recommend the low-risk financial product, the pre-support product and other products as products related to the pre-support constraint type user to the user with the target user category of the pre-support constraint type.

Taking potential periodic users as an example, the server obtains feature information of the potential periodic users according to sample index information of clustered sample users taking sample user categories as potential periodic users, wherein the feature information comprises: young people who have just started working, have the highest churn rate, have a high number of products in hold, have a medium income level, have a high pre-pay contribution and prefer online channels, often use a three-party paymate, and receive new things faster. Based on the characteristic information, the server can recommend the innovative service products, the three-party payment binding activities and other products to the users with the target user category of potential periodic users as products associated with the potential periodic users.

Taking the high-end necessary-protection type user as an example, the server obtains the characteristic information of the high-end necessary-protection type user according to the sample index information of the clustered sample users with the sample user category as the high-end necessary-protection type user, and the method comprises the following steps: the system has the advantages of being stable in work, low in loss rate, and capable of enabling five-star level and above users to have the highest proportion, the most quantity of products to be held, the highest income level, the most credit card contribution, the lowest pre-payment contribution, preference of online channels and the most binding of three-party payment, and meanwhile, being the user with the most financial products. Based on the characteristic information, the server can recommend the investment product as the product associated with the high-end type-keeping user to the user of which the target user is the high-end type-keeping user, and meanwhile, the server can guide the user to pay and pay daily through the enterprise product on line, so that the demands of the user in various aspects such as education, room purchase, medical treatment, travel and the like are met, and the viscosity of the user and the enterprise is increased.

Taking a deposit-dependent user as an example, the server obtaining feature information of the deposit-dependent user according to sample index information of clustered sample users taking sample user categories as the deposit-dependent user includes: middle aged and elderly people who are on the verge of retirement or have retired, have moderate churn rates, have low amounts of held products, have the lowest income level, have higher deposit contributions, prefer counter channels, and pay attention to funds and safe payments. Based on the characteristic information, the server can recommend products with steady benefits such as low-risk financial products, insurance products, pension financial products and the like to the user of which the target user is a deposit-dependent user as products associated with the deposit-dependent user in an off-line face-to-face communication mode.

In this embodiment, the feature information of each user category of the pre-trained clustering model is determined by clustering sample index information of sample users, so that associated products are confirmed for each user category, and after target user categories of users to be recommended are obtained, associated products are accurately selected for the users, and product recommendation is completed. Based on the association relationship between the user category and the product obtained by the process, the accuracy of product recommendation can be improved.

In an exemplary embodiment, in the step, for each user category, according to the feature information of the user category, the association relationship between the user category and the product is confirmed, which specifically includes the following contents: predicting the product consumption behavior of the associated sample user of the user category according to the characteristic information of the user category aiming at each user category; identifying a product associated with a product consumption behavior of an associated sample user; and according to the user category and the product associated with the product consumption behavior of the associated sample user, confirming the association relationship between the user category and the product.

Specifically, the server predicts the product consumption behavior of the associated sample user of the user category in a future period according to the characteristic information of each user category; then, according to the predicted product consumption behavior, confirming the product related to the product consumption behavior; and finally, according to the user category and the product associated with the product consumption behavior of the user in the association sample of the user category, confirming the association relationship between the user category and the product.

Taking a high-end necessary-to-be-ensured user as an example, the server predicts the product consumption behaviors of the related sample users of the high-end necessary-to-be-ensured user according to the characteristic information of the high-end necessary-to-be-ensured user, wherein the product consumption behaviors comprise: preference of online channels, preference of investment products, and acceptance of investment products with high risk and high income; therefore, the server can use the investment product as the product associated with the high-end necessary-to-be-ensured user, so that the association relationship between the high-end necessary-to-be-ensured user and the corresponding product is confirmed.

Taking a deposit-dependent user as an example, the server predicts the product consumption behavior of the obtained correlation sample user of the deposit-dependent user according to the characteristic information of the deposit-dependent user, wherein the product consumption behavior comprises the following steps: the preference counter channel and deposit contribution are still main contributions, and the product with steady income is more preferred because of lower acceptance of financial products with high risk; therefore, the server can take the low-risk financial products, insurance products, pension financial products and other products as the products associated with the deposit-dependent user, thereby confirming the association relationship between the deposit-dependent user and the corresponding products.

In this embodiment, the server determines, for each user category, a product preferred by the user by predicting the product consumption behavior of the user based on the association sample of the user category, so as to establish an association relationship between each user category and the product, and further enable a subsequent process of accurately recommending the product for the user based on the target user category of the user, thereby improving accuracy of product recommendation.

In an exemplary embodiment, as shown in fig. 5, another product recommendation method is provided, which is described by taking the application of the method to a server as an example, and includes the following steps:

step S501, training a resource loss information prediction model to be trained according to sample resource information of a predicted sample user to obtain a pre-trained resource loss information prediction model.

Step S502, clustering the clustering model to be trained according to sample index information of the clustering sample users under a plurality of preset clustering indexes to obtain a pre-trained clustering model.

Step S503, predicting the product consumption behavior of the associated sample user of the user category according to the characteristic information of each user category of the pre-trained cluster model.

The characteristic information of the user category is obtained according to sample index information of the associated sample user of the user category under a plurality of preset clustering indexes; the associated sample users of the user categories are the clustered sample users used for training the pre-trained cluster model, and the corresponding sample user categories are the clustered sample users of the user categories.

Step S504, confirming the products associated with the product consumption behaviors of the associated sample users, and confirming the association relationship between the user categories and the products according to the user categories and the products associated with the product consumption behaviors of the associated sample users.

In step S505, the resource information of the user is subjected to resource loss prediction processing through a pre-trained resource loss information prediction model, so as to obtain the resource loss information of the user.

Step S506, under the condition that the resource loss information of the user meets the preset resource loss threshold, the user type of the user is confirmed to be the resource loss user.

In step S507, in the case that the user type of the user is a resource loss user, index information of the user under a plurality of basic feature indexes is obtained, and correlation information among the plurality of basic feature indexes is confirmed.

Step S508, fusion processing is carried out on the basic characteristic indexes of which the corresponding correlation information meets the preset correlation threshold value, so as to obtain a plurality of preset clustering indexes.

Step S509, for each preset clustering index, performing fusion processing on index information of the user under the basic characteristic index corresponding to the preset clustering index to obtain index information of the user under the preset clustering index.

Step S510, inputting index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering to obtain target user categories of the user.

Step S511, inquiring the association relationship, and confirming the product associated with the target user category as the target product.

In step S512, the target product is recommended to the user.

In the embodiment, a server trains a resource loss information prediction model to be trained by predicting resource information of a sample user to obtain a model capable of accurately predicting the resource loss information; meanwhile, training the clustering model to be trained through index information of the clustering sample user under a plurality of preset clustering indexes to obtain a clustering model with high clustering effect distinction degree and accurate clustering result; in addition, through sample index information of the clustering sample users, feature information of each user category of the pre-trained clustering model is determined, and product consumption information of the associated sample users of the user categories in a future period is predicted, so that associated products are confirmed for each user category. Based on the preparation process, the server confirms the basis of locking the target guest group, the basis of realizing accurate clustering of the user and the association relation between the user category and the product for the product recommendation process, so that the server can accurately confirm the target guest group needing product recommendation through a pre-trained resource loss information prediction model; and then confirming the user category of the user through index information of the user under a plurality of preset clustering indexes and a pre-trained clustering model, confirming a target product required to be recommended to the user according to the user category, and further completing product recommendation of the user. The product recommendation method based on the process can not be interfered by subjective influences, can realize accurate classification of users and product recommendation for the users, improves the accuracy of product recommendation, and further can effectively save the users with lost resources.

In order to more clearly illustrate the product recommendation method provided in the embodiments of the present application, a specific embodiment is described below specifically. In an exemplary embodiment, the present application further provides a cluster analysis method based on a middle-high-end fluid loss client, which specifically includes the following steps:

step 1: a target guest group is selected.

The server judges that the user with deposit resources of 5-20 ten thousand yuan is a middle-high end user from business, and defines the resource loss user as the reduction of 50% or more of the month-day average financial assets in the expression period relative to the month-day average financial assets in the last month of the observation period.

The server uses Keras to build a Sequential model, builds a deep neural network consisting of an input layer, three hidden layers and an output layer, imports data of all users in observation period, expression period and interval period into the model, trains user data in the observation period and interval period by adopting a back propagation algorithm, predicts whether users in the expression period can run off by using a classification algorithm, classifies the users into non-resource-loss users and resource-loss users according to results, and the resource-loss users are target guest groups for subsequent cluster analysis.

Step 2: and screening the clustering index.

The server selects indexes with guiding significance for user category division based on historical service information, and selects 6 core dimensions of resource reduction proportion, age, product coverage rate, income, channel preference and user contribution value as basic indexes of a clustering algorithm.

Step 3: and (5) deriving a clustering index.

The server analyzes and processes the asset descending proportion index, and derives the flow loss rate index according to the condition of the descending proportion of the average asset of the month and the day in the expression period at the end of the observation period; analyzing and processing the age index, and deriving life cycle index according to the age group; analyzing and processing the product coverage rate index, and deriving a product holding condition index according to the holding quantity condition of the product; analyzing and processing the income index, and deriving a income level index according to the condition of the wage amount sent for the generation of 6 months; analyzing and processing the channel preference index, and deriving the preference index according to the transaction number proportion condition of each channel; and analyzing and processing the user contribution value index, and deriving a contribution type index according to the contribution type index proportion situation.

Step 4: and (5) fusing the clustering indexes.

The server adopts a random gradient descent algorithm to carry out weight analysis on the four indexes of the life cycle index, the income level index, the product holding condition index and the contribution type index which are derived in the step 3, the life cycle index and the income level index are fused into a user value index, and the product holding condition index and the contribution type index are fused into a user contribution index.

Step 5: preprocessing data.

The server respectively performs extremely poor standardization processing on the data of the user under the loss rate index, the user value index, the user contribution index and the preference index, so that the processed variables are subjected to standard normal distribution.

Step 6: and establishing a clustering model.

The server adopts a K-Means algorithm to establish a clustering model, and inputs four index data subjected to the range normalization processing in the step 5, so as to train the model. The time sequence analysis method is used for sequencing the user contribution index data in the model according to time sequence, extracting the data of the historical consumption habit data of the user, forming the prediction of the future consumption behavior of the user, and providing basis for the subsequent establishment of marketing strategies. In order to optimize the model distinguishing effect, the clustering number is set to 4 to 9, clustering analysis is carried out respectively, the clustering effect is evaluated by using the Calinski-Harabasz index, and when the clustering number is 4, the Calinski-Harabasz score is highest, the clustering effect is best, and the good distinguishing degree is achieved. And meanwhile, the clustering result is subjected to dimension reduction by using the visualization tool TSNE, so that the clustering result is more visual.

Step 7: and describing a clustering result.

The server performs user feature description on each cluster of the cluster model respectively, and describes the result of the cluster model as: pre-pay constraint type users, potential installable type users, high-end obligatory type users and deposit dependent type users.

Pre-support constraint users: the life cycle of the user is in a steady growth stage, and the method has the characteristics of lowest loss rate, more held products, higher payroll income level, highest pre-payment contribution, preference of using online channels for trading, extremely low arrival rate and the like.

Potential deadline user: the life cycle of the user is in the early-out period of the cottage, and the method has the characteristics of highest loss rate, more held products, medium payroll income level, higher pre-payment contribution, preference for online banking channel transaction, most frequent use of three-party payment platform transaction and the like.

High-end must keep-type users: the life cycle of the users is in an acceleration rising stage, the enterprise high management proportion is the largest, the five-star level and above users have the highest proportion, the loss rate is lower, the number of held products is the largest, the payroll income level is the highest, the contribution of medium-income and credit cards is the largest, the contribution of pre-charge is the lowest, mobile banking transactions are favored, the three-party binding cards are the largest, in-line transfer and cross-line transfer transactions are frequent, financial transactions are the largest, and the like.

Deposit dependent users: the life cycle of the user is in an endangered retirement stage, and the method has the characteristics of medium loss rate, minimum quantity of held products, minimum payroll income level, high arrival rate, extremely good counter exchange, less contribution, less transaction amount, minimum liability, good deposit preference, extremely little use of three-party payment, paying attention to funds, payment safety and the like.

Step 8: recommending products for users.

And the server recommends products and services for the user according to the prediction of the future consumption behavior of the user and the user portrait characteristic description in the clustering model.

Pre-support constraint users: according to the preference characteristics of the users, an online marketing strategy is generated, online transaction channels which are used recently by the users are screened, and the recommendation of investment products such as financial accounting is put in when the users access through the channels, so that the probability of becoming financial users is increased. Through the mobile banking pushing the rewarding questionnaire or the short message pushing to the store and other activities, the user is invited and attracted to the store, and the opportunity of face-to-face communication with the user is increased, so that more sales possibilities can be found. According to the contribution characteristics of the users, other pre-support products are recommended to the users in a mode of mobile banking or short message pushing and the like three months before the pre-support of the users expires, so that new pre-support constraint is added, and the user saving is realized.

Potential deadline user: according to the preference characteristics of the users, online marketing strategies are generated, innovative service platforms such as cloud network points are invited to the users through channels such as mobile banking and short messages, and the users are guided to complete new technical experiences including business handling, scenic spot ticket purchasing and other personalized services. According to the contribution characteristics of the user, the activities of pushing the three-party payment platform to bind the bank card to win the money reduction activity, consuming the money reduction activity by using the bank card on the three-party payment platform and the like are carried out, so that the probability of binding the bank card on the three-party payment platform and the frequency of paying by using the bank card of the user are improved, and the user can save the money.

High-end must keep-type users: according to preference characteristics of the users, online and offline marketing strategies are generated, financial services aiming at different scenes are formulated, the users are guided to pay traffic fine, property fees, electric fees and the like daily through mobile banking online, the rigidity requirements of the users in education, house purchasing, endowment, medical treatment, delivery, travel and the like are met, and the viscosity of the users is increased in more application scenes. By being equipped with a dedicated user manager offline, the opportunity of face-to-face contact with the user is increased, and the product combination is designed according to the characteristics of the user, so that the user is guided to hold investment products.

Deposit dependent users: according to the preference characteristics of the users, an offline marketing strategy is generated, deposit business transactions of the users are maintained in an offline face-to-face communication mode, and a teller or hall user manager guides the users to select and configure products with steady benefits such as low-risk financial products, insurance products and pension financial products so as to replace deposit with increasingly reduced benefits, thereby realizing user saving.

In this embodiment, the server performs modeling analysis on core index data of the middle-high-end lost user based on an artificial intelligence model technology, so as to construct a clustered image of the lost user, characterize the data, analyze deep reasons behind the data, reasonably and effectively allocate recommended resources, and develop a basis for targeted recommended measures, thereby achieving the purpose of maximally improving recovery rate of the lost user.

It should be understood that, although the steps in the flowcharts related to the above embodiments are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

Based on the same inventive concept, the embodiment of the application also provides a product recommendation device for realizing the above related product recommendation method. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation of one or more embodiments of the product recommendation device provided below may refer to the limitation of the product recommendation method hereinabove, and will not be repeated here.

In an exemplary embodiment, as shown in fig. 6, there is provided a product recommendation apparatus including: a loss information prediction module 602, an index information acquisition module 604, a user category clustering module 606, and a target product recommendation module 608, wherein:

the loss information prediction module 602 is configured to predict the user's resource loss information according to a pre-trained resource loss information prediction model, and confirm the user type of the user according to the user's resource loss information.

The index information obtaining module 604 is configured to obtain index information of the user under a plurality of preset cluster indexes when the user type of the user is a resource loss user.

The user category clustering module 606 is configured to input index information of the user under a plurality of preset clustering indexes into a pre-trained clustering model for clustering, so as to obtain a target user category of the user.

The target product recommending module 608 is configured to recommend a target product associated with a target user category to a user.

In an exemplary embodiment, the attrition information prediction module 602 is further configured to obtain resource information of a user; performing resource loss prediction processing on the resource information of the user through a pre-trained resource loss information prediction model to obtain the resource loss information of the user; and under the condition that the resource loss information of the user meets the preset resource loss threshold value, determining the user type of the user as the resource loss user.

In an exemplary embodiment, the product recommendation device further includes a resource loss information prediction model training module, configured to obtain sample resource information of each prediction sample user; for each predicted sample user, dividing the sample resource information into observation period resource information, interval period resource information and presentation period resource information according to the time sequence of the sample resource information of the predicted sample user; inputting the observation period resource information, the interval period resource information and the expression period resource information into a resource loss information prediction model to be trained to obtain expression period resource prediction information of a prediction sample user; and training the resource loss information prediction model to be trained through the difference between the expression period resource prediction information and the expression period resource information to obtain a trained resource loss information prediction model, and taking the trained resource loss information prediction model as a pre-trained resource loss information prediction model.

In an exemplary embodiment, the product recommendation device further includes a cluster model training module, configured to obtain sample index information of a clustered sample user under a plurality of preset cluster indexes; clustering sample index information of a clustered sample user under a plurality of preset clustering indexes is subjected to clustering processing through a clustering model to be trained, so that sample user categories of the clustered sample user are obtained; confirming category difference information among the sample user categories; and training the clustering model to be trained under the condition that the category difference information is smaller than a preset difference threshold value, so as to obtain a training-completed clustering model, and taking the training-completed clustering model as a pre-trained clustering model.

In an exemplary embodiment, the index information obtaining module 604 is further configured to obtain index information of the user under a plurality of basic feature indexes; confirming correlation information among a plurality of basic characteristic indexes; fusing the basic characteristic indexes of which the corresponding correlation information meets the preset correlation threshold value to obtain a plurality of preset clustering indexes; and aiming at each preset clustering index, carrying out fusion processing on index information of the user under the basic characteristic index corresponding to the preset clustering index to obtain the index information of the user under the preset clustering index.

In an exemplary embodiment, the product recommendation device further includes an association relationship confirmation module for confirming each user category of the pre-trained cluster model; acquiring characteristic information of each user category; the characteristic information of the user category is obtained according to sample index information of the associated sample user of the user category under a plurality of preset clustering indexes; the associated sample users of the user categories are the clustered sample users used for training a pre-trained clustered model, and the corresponding sample user categories are the clustered sample users of the user categories; and aiming at each user category, according to the characteristic information of the user category, confirming the association relationship between the user category and the product.

The target product recommendation module 608 is further configured to query the association relationship, and confirm a product associated with the target user category as a target product; and recommending the target product to the user.

In an exemplary embodiment, the association relationship confirmation module is further configured to predict, for each user category, product consumption behaviors of an associated sample user of the user category according to feature information of the user category; identifying a product associated with a product consumption behavior of an associated sample user; and according to the user category and the product associated with the product consumption behavior of the associated sample user, confirming the association relationship between the user category and the product.

The respective modules in the above-described product recommendation apparatus may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In an exemplary embodiment, a computer device is provided, which may be a server, and an internal structure thereof may be as shown in fig. 7. The computer device includes a processor, a memory, an Input/Output interface (I/O) and a communication interface. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer equipment is used for storing data such as resource loss information of a user, index information of the user, characteristic data of products and the like. The input/output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a product recommendation method.

It will be appreciated by those skilled in the art that the structure shown in fig. 7 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In an exemplary embodiment, a computer device is also provided, comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the method embodiments described above when the computer program is executed.

In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method embodiments described above.

In an exemplary embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of the method embodiments described above.

It should be noted that, the user information (including, but not limited to, user equipment information, user personal information, etc.) and the data (including, but not limited to, data for analysis, stored data, presented data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party, and the collection, use and processing of the related data are required to comply with the related laws and regulations and standards of the related countries and regions.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the various embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the various embodiments provided herein may include at least one of relational databases and non-relational databases. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic units, quantum computing-based data processing logic units, etc., without being limited thereto.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The foregoing examples represent only a few embodiments of the present application, which are described in more detail and are not thereby to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application shall be subject to the appended claims.

Claims

1. A method of product recommendation, the method comprising:

2. The method according to claim 1, wherein predicting the resource loss information of the user through a pre-trained resource loss information prediction model, and determining the user type of the user according to the resource loss information of the user, comprises:

acquiring resource information of a user;

3. The method of claim 1, wherein the pre-trained resource churn information prediction model is trained by:

sample resource information of each prediction sample user is obtained;

4. The method of claim 1, wherein the pre-trained cluster model is trained by:

confirming category difference information among the sample user categories;

5. The method according to claim 1, wherein the obtaining the index information of the user under a plurality of preset cluster indexes includes:

6. The method of any one of claims 1 to 5, further comprising, prior to recommending the target product associated with the target user category to the user:

confirming each user category of the pre-trained cluster model;

recommending the target product to the user.

7. The method according to claim 6, wherein for each user category, determining the association relationship between the user category and the product according to the feature information of the user category includes:

8. A product recommendation device, the device comprising:

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.

11. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.