CN117893279A - Object recommendation method and device - Google Patents

Abstract

The disclosure relates to the technical field of data processing, and provides a method and a device for recommending objects. The method comprises the following steps: performing feature transformation on the initial feature vector of each candidate object and the initial feature vector of each interactive object through the large language model to obtain the feature vector of each candidate object and the feature vector of each interactive object; performing attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects to obtain enhanced feature vectors of the user; and predicting each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, calculating to obtain a first prediction probability of each candidate object, and selecting a target object recommended by the oriented user from the candidate object feature data set, thereby solving the problem of inaccurate cold start object recommendation in the prior art, effectively solving the problem of object recommendation under long tail distribution, and improving the accuracy of a platform recommendation system.

Description

Object recommendation method and device

Technical Field

The disclosure relates to the technical field of data processing, in particular to a method and a device for recommending objects.

Background

The recommendation system plays an important role in the present informatization age, and can be applied to data recommendation in online shopping, news information reading, or video watching. The recommendation system can predict the preference of the user according to some behaviors of the user and information of the user, so that products possibly interested by the user are recommended to the user. However, in the existing platform containing the recommendation system, the exposure rate of a large part of products is extremely low, and a small part of hot products occupy the vast majority of the exposure of the recommendation system. The existing recommendation system is mainly used for estimating user preference based on user characteristics and product characteristics, but the exposure of the cold door products is low, and the interaction characteristics with users are few, so that the problem of inaccurate cold door product estimation is caused due to insufficient training of the user characteristics and the product characteristics, and therefore recommendation positions of the cold door products are few, and the recommended cold starting effect of the cold door products is poor.

Disclosure of Invention

In view of the above, embodiments of the present disclosure provide a method, an apparatus, an electronic device, and a readable storage medium for recommending an object, so as to solve the problem in the prior art that recommending a cold start object is inaccurate.

In a first aspect of an embodiment of the present disclosure, there is provided an object recommendation method, including: acquiring user behavior sequence feature data and a candidate object feature data set, wherein the user behavior sequence feature data comprises a plurality of interactive object feature data interacted by a user; vector embedding is carried out on the feature data of each candidate object to obtain an initial feature vector of each candidate object, and vector embedding is carried out on the feature data of each interaction object to obtain an initial feature vector of each interaction object; performing feature transformation on the initial feature vectors of all candidate objects through the large language model to obtain feature vectors of all candidate objects, and performing feature transformation on the initial feature vectors of all interactive objects through the large language model to obtain feature vectors of all interactive objects; performing attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects to obtain enhanced feature vectors of users corresponding to the candidate objects; predicting each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculating to obtain a first prediction probability of each candidate object; a target object for user recommendation is selected from the candidate object feature data set based on the first prediction probabilities of the respective candidate objects.

In a second aspect of the embodiments of the present disclosure, there is provided an object recommendation apparatus, including: the system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring user behavior sequence characteristic data and a candidate object characteristic data set, and the user behavior sequence characteristic data comprises a plurality of interactive object characteristic data interacted by a user; the vector embedding module is used for carrying out vector embedding on the feature data of each candidate object to obtain an initial feature vector of each candidate object, and carrying out vector embedding on the feature data of each interaction object to obtain an initial feature vector of each interaction object; the feature transformation module is used for carrying out feature transformation on the initial feature vectors of all candidate objects through the large language model to obtain feature vectors of all candidate objects, and carrying out feature transformation on the initial feature vectors of all interactive objects through the large language model to obtain feature vectors of all interactive objects; the attention processing module is used for carrying out attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects to obtain enhanced feature vectors of the users corresponding to the candidate objects; the prediction module is used for predicting each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculating to obtain a first prediction probability of each candidate object; and the recommendation module is used for selecting target objects recommended by the oriented user from the candidate object characteristic data set based on the first prediction probability of each candidate object.

In a third aspect of the disclosed embodiments, an electronic device is provided, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the above method when executing the computer program.

In a fourth aspect of the disclosed embodiments, there is provided a readable storage medium storing a computer program which, when executed by a processor, implements the steps of the above method.

Compared with the prior art, the embodiment of the disclosure has the beneficial effects that: the candidate object feature data is input into a recommendation model to perform vector embedding on the candidate object feature data to obtain initial feature vectors of candidate objects, each interactive object feature data is subjected to vector embedding to obtain initial feature vectors of interactive objects, the initial feature vectors can be used as characterization of users, and the interactive objects are objects of interactive behaviors of the users and can comprise clicked, browsed or purchased objects of the users. The initial feature vectors of the candidate objects and the initial feature vectors of the interaction objects are input into a pre-trained large language model, and the strong text extraction capacity of the large language model is utilized to obtain more abundant and diverse feature vectors, namely the feature vectors of the candidate objects and the feature vectors of the interaction objects, so that the effect of subsequent tasks is improved. And carrying out attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects, carrying out weighted summation on the feature vectors of the interaction objects involved in each interaction action of the user to obtain enhanced feature vectors of the user, and carrying out prediction processing based on the enhanced feature vectors of the user and the feature vectors of the candidate objects to obtain first prediction probability of the candidate objects, wherein the first prediction probability is the probability of the recommendation system calculating to obtain the object clicked by the user. And selecting the target object with high first prediction probability as a target object to be recommended to the user based on the prediction probability of each candidate object. The large language model is obtained by pre-training based on the open source large model on the basis of platform data, dedicated optimization is made for the products of the platform, and the input feature vectors can be enriched and characterized, so that the features of candidate objects and the features of interactive objects are fully trained, the expressive of the feature vectors of the objects are improved, more accurate prediction results are obtained, more accurate recommendation is performed by combining the prediction results, the problem that in the prior art, the cold start object is inaccurate is solved, the accuracy of a platform recommendation system is improved, the object recommendation problem under long tail distribution is effectively solved, and the use experience of users is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following description will briefly explain the embodiments or the drawings required to be used in the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and other drawings may be obtained according to these drawings without inventive effort for a person of ordinary skill in the art.

Fig. 1 is a scene schematic diagram of an application scene of an embodiment of the present disclosure;

FIG. 2 is a flowchart of an object recommendation method according to an embodiment of the present disclosure;

FIG. 3 is a flowchart of another object recommendation method according to an embodiment of the present disclosure;

FIG. 4 is a flowchart of another object recommendation method according to an embodiment of the present disclosure;

fig. 5 is a schematic structural diagram of an object recommendation apparatus according to an embodiment of the present disclosure;

fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure.

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system configurations, techniques, etc. in order to provide a thorough understanding of the disclosed embodiments. However, it will be apparent to one skilled in the art that the present disclosure may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present disclosure with unnecessary detail.

It should be noted that, the user information (including, but not limited to, terminal device information, user personal information, etc.) and the data (including, but not limited to, data for presentation, analyzed data, etc.) related to the present disclosure are information and data authorized by the user or sufficiently authorized by each party.

A recommendation method and apparatus for an object according to embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

Fig. 1 is a scene diagram of an application scene of an embodiment of the present disclosure. The application scenario may include terminal devices 1, 2 and 3, a server 4 and a network 5.

The terminal devices 1, 2 and 3 may be hardware or software. When the terminal devices 1, 2 and 3 are hardware, they may be various electronic devices having a display screen and supporting communication with the server 4, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like; when the terminal apparatuses 1, 2, and 3 are software, they can be installed in the electronic apparatus as above. The terminal devices 1, 2 and 3 may be implemented as a plurality of software or software modules, or as a single software or software module, to which the embodiments of the present disclosure are not limited. Further, various applications, such as a data processing application, an instant messaging tool, social platform software, a search class application, a shopping class application, and the like, may be installed on the terminal devices 1, 2, and 3.

The server 4 may be a server that provides various services, for example, a background server that receives a request transmitted from a terminal device with which communication connection is established, and the background server may perform processing such as receiving and analyzing the request transmitted from the terminal device and generate a processing result. The server 4 may be a server, a server cluster formed by a plurality of servers, or a cloud computing service center, which is not limited in the embodiment of the present disclosure.

The server 4 may be hardware or software. When the server 4 is hardware, it may be various electronic devices that provide various services to the terminal devices 1, 2, and 3. When the server 4 is software, it may be a plurality of software or software modules providing various services to the terminal devices 1, 2, and 3, or may be a single software or software module providing various services to the terminal devices 1, 2, and 3, which is not limited by the embodiments of the present disclosure.

The network 5 may be a wired network using coaxial cable, twisted pair wire, and optical fiber connection, or may be a wireless network that can implement interconnection of various communication devices without wiring, for example, bluetooth (Bluetooth), near field communication (Near Field Communication, NFC), infrared (Infrared), etc., which is not limited by the embodiment of the present disclosure.

The user can establish a communication connection with the server 4 via the network 5 through the terminal devices 1, 2, and 3 to receive or transmit information or the like. Specifically, the server 4 acquires user behavior sequence feature data and a candidate object feature data set, wherein the user behavior sequence feature data comprises a plurality of interactive object feature data interacted by a user; vector embedding is carried out on the feature data of each candidate object to obtain an initial feature vector of each candidate object, and vector embedding is carried out on the feature data of each interaction object to obtain an initial feature vector of each interaction object; performing feature transformation on the initial feature vectors of all candidate objects through the large language model to obtain feature vectors of all candidate objects, and performing feature transformation on the initial feature vectors of all interactive objects through the large language model to obtain feature vectors of all interactive objects; performing attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects to obtain enhanced feature vectors of users corresponding to the candidate objects; predicting each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculating to obtain a first prediction probability of each candidate object; a target object for user recommendation is selected from the candidate object feature data set based on the first prediction probabilities of the respective candidate objects.

It should be noted that the specific types, numbers and combinations of the terminal devices 1, 2 and 3, the server 4 and the network 5 may be adjusted according to the actual requirements of the application scenario, which is not limited by the embodiment of the present disclosure.

Fig. 2 is a flowchart of a method for object recommendation according to an embodiment of the present disclosure. The method of object recommendation of fig. 2 may be performed by the terminal device or the server of fig. 1. As shown in fig. 2, the method for recommending the object includes:

step 201, acquiring user behavior sequence feature data and a candidate object feature data set, wherein the user behavior sequence feature data comprises a plurality of interactive object feature data interacted by a user.

In some embodiments, the object recommendation method of the present disclosure may be applied to recommendation systems of various platforms, including shopping platforms, house renting platforms, information reading platforms, video platforms, and the like. When the disclosed method is applied to a shopping platform, the user behavior sequence feature data comprises a plurality of interactive object feature data interacted by a user, the plurality of interactive objects interacted by the user can be a plurality of interactive objects clicked by the user, purchased by the user, browsed by the user and collected by the user, the feature data of the interactive objects can be serial number features comprising the interactive objects, category features of the interactive objects, title text of the interactive objects and the like, the serial numbers of the interactive objects are unique identifiers of the objects in the platform, and the serial numbers of each object in the platform are unique and are used for identifying the identity information of the user in the platform. The candidate object characteristic data set is obtained from a plurality of objects in the platform database through a recall method, thousands of objects can be roughly selected from millions of objects, and the recall can be carried out by adopting a collaborative filtering and user image drawing method. The candidate feature data set includes a plurality of candidate data, and the feature data of the candidate may include a numbered feature of the candidate, a category feature of the candidate, a title text of the candidate, and the like. In the object recommendation method disclosed by the disclosure, the number of the plurality of interactive objects contained in the user behavior sequence feature data is not limited, and the plurality of interactive objects can be interacted by the user within three days or the plurality of objects interacted by the user within seven days. After user behavior sequence feature data and candidate object feature data sets are obtained, the user behavior sequence feature data and the candidate object feature data sets are input into a recommendation model, the recommendation model predicts probability of each candidate object in the candidate object feature data sets based on the user behavior sequence feature data and the candidate object feature data sets, predicts whether a user is interested in the candidate object, obtains prediction probability of each candidate object, and selects a plurality of target objects to recommend to the user based on the prediction probability of each candidate object.

Step 102, performing vector embedding on the feature data of each candidate object to obtain an initial feature vector of each candidate object, and performing vector embedding on the feature data of each interaction object to obtain an initial feature vector of each interaction object.

In some embodiments, the recommendation model includes a first embedded and a second embedded layer. The candidate feature data includes a plurality of continuous features and a plurality of discrete features, in the embodiment of the present disclosure, the continuous features may be prices of the candidate objects, warmth of the candidate objects, and the like, where the continuous features refer to that the values of the features are continuous, and the discrete features of the candidate objects may be classification variables or binary variables of the candidate objects, for example, categories of the candidate objects, and codes of the candidate objects. Hash coding is carried out on each discrete feature of each candidate object to obtain a coding result of each discrete feature of each candidate object, wherein the hash coding is a coding mode for converting input data into a unique identifier with a fixed length, and the hash coding is used for calculating each discrete feature of each candidate object based on a hash function to generate a hash value with the fixed length, namely the coding result of each discrete feature of each candidate object. The recommendation model comprises a plurality of first embedding layers and second embedding layers, wherein the first embedding layers are the generic names of embedding layers for carrying out vector embedding on the feature data of each candidate object, and the second embedding layers are the generic names of embedding layers for carrying out vector embedding on the feature data of each interaction object. And inputting the coding results of the discrete features of each candidate object into a first embedding layer of the recommendation model to carry out vector embedding, and carrying out vector embedding on each discrete feature coding result to obtain feature vectors of the discrete features of each candidate object through independent first embedding layers, wherein parameters in each first embedding layer are different. And processing each continuous feature of each candidate object, and respectively carrying out barrel classification processing on each continuous feature of each candidate object to obtain barrel classification results of each continuous feature of each candidate object. The continuous feature is binned, i.e., the continuous values are mapped into a plurality of finite bins, e.g., the price of the candidate is split into a plurality of bins, such as "0-50", "50-150", "150-250", and so on. And inputting the barrel division result of each continuous feature of each candidate object into a first embellishing layer of the recommendation model to perform vector embedding, so as to obtain feature vectors of each continuous feature of each candidate object. And splicing the feature vector of each discrete feature of each candidate object with the feature vector of each continuous feature of each candidate object to obtain the initial feature vector of each candidate object.

In some embodiments, the interactive object feature data includes a plurality of continuous features and a plurality of discrete features, and in embodiments of the present disclosure, the continuous features of the interactive object may be a price of the interactive object, a heat of the interactive object, and the like, where the continuous features refer to that a value of the feature is continuous, and the discrete features of the interactive object may be a classification variable or a binary variable of the interactive object, for example, a category of the interactive object, an encoding of the interactive object, and a title text of the interactive object. Hash coding is carried out on each discrete feature of each interactive object, and coding results of each discrete feature of each interactive object are obtained. And inputting the coding results of the discrete features of each interactive object into a second embellishing layer of the recommendation model to carry out vector embedding, wherein the vector embedding is carried out on each discrete feature coding result through a single second embellishing layer, and parameters in each second embellishing layer are different, so that feature vectors of the discrete features of each interactive object are obtained. And processing each continuous feature of each interactive object, and respectively carrying out barrel separation processing on each continuous feature of each interactive object to obtain barrel separation results of each continuous feature of each interactive object. The sequential features are binned, i.e., sequential values are mapped into a plurality of finite bins, e.g., the price of the interactive object is split into a plurality of bins, such as "0-50", "50-150", "150-250", and so on. And inputting the barrel division result of each continuous feature of each interactive object into a second embellishing layer of the recommendation model to perform vector embedding, so as to obtain feature vectors of each continuous feature of each interactive object. And splicing the feature vector of each discrete feature of each interactive object with the feature vector of each continuous feature of each interactive object to obtain the initial feature vector of each interactive object.

The candidate object feature data and the interactive object feature data are embedded in vectors, semantic and structural information of the data can be captured, and data of different types and formats can be uniformly converted into vector representation, so that the recommendation model can better process and understand the data, and accuracy of the recommendation model is improved.

And 103, performing feature transformation on the initial feature vectors of the candidate objects through the large language model to obtain feature vectors of the candidate objects, and performing feature transformation on the initial feature vectors of the interactive objects through the large language model to obtain feature vectors of the interactive objects.

In some embodiments, the recommendation model further includes a large language model, the large language model may be based on an open-source chatGLM, the chatGLM may be pre-trained based on database data of the shopping platform, dedicated optimization may be performed on vector characterization capability of an e-commerce scene of the shopping platform, and parameter adjustment and optimization may be performed on the large language model according to distribution and characteristics of the database data of the shopping platform during training, so as to better adapt to data characteristics and recommendation requirements of the shopping platform, and obtain the large language model. And inputting the initial feature vector of each candidate object into a large language model, and enhancing the connotation expression of the input feature vector based on the large language model to obtain the feature vector of each candidate object. And inputting the initial feature vector of each interactive object into a large language model to perform feature transformation to obtain a richer and more diverse feature vector, namely the feature vector of each interactive object. Training the large language model based on shopping platform database data, placing the trained large language model in a recommendation model, enhancing connotation expression of feature vectors through the large language model, and improving accuracy of the recommendation model.

And 104, performing attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects to obtain enhanced feature vectors of the users corresponding to the candidate objects.

In some embodiments, for a candidate object, each similarity score between the candidate object and the feature vector of each interaction object is calculated, and each similarity score is used as a weight to carry out weighted summation with the feature vector of each interaction object, so as to obtain an enhanced feature vector of the user corresponding to the candidate object, wherein the enhanced feature vector of the user integrates the similarity and the corresponding features of the user and different interaction objects, and the interests and the preferences of the user can be more comprehensively reflected. For each candidate object, the enhancement feature vector of the user corresponding to each candidate object can be obtained. For one candidate object, the similarity score between the candidate object and the feature vector of each interaction object is calculated, and each similarity score is used as a weight to carry out weighted summation with the feature vector of the candidate object, so that the enhanced feature vector of the user corresponding to the candidate object is obtained, the accuracy and the reliability of a recommendation system can be improved, and personalized recommendation service can be provided for the user better.

And 105, carrying out prediction processing on each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculating to obtain a first prediction probability of each candidate object.

In some embodiments, the first prediction probability of the candidate object may be obtained by predicting whether the user will click on the candidate object based on the enhanced feature vector of the user and the feature vector of the candidate object, where the first prediction probability may be a specific value between 0 and 1. The first prediction probability of each candidate object is a precondition for carrying out subsequent recommendation, and tens of candidate objects with higher first prediction probability can be selected from thousands of candidate objects to be recommended to the user.

And 106, selecting target objects recommended by the oriented users from the candidate object characteristic data set based on the first prediction probability of each candidate object.

In some embodiments, thousands of level candidates may be recalled from a platform database of over ten thousands of level objects, and a first prediction probability for each candidate is predicted by the recommendation model, the first prediction probability being the likelihood that the candidate was clicked by the user calculated by the recommendation model. A preset threshold value can be set, candidate objects with the first prediction probability larger than the preset threshold value are selected as target objects, and the target objects are scattered and rearranged to be recommended to the user.

According to the object recommendation method, the candidate object feature data are subjected to vector embedding to obtain initial feature vectors of the candidate objects, the interactive object feature data are subjected to vector embedding to obtain initial feature vectors of the interactive objects, the initial feature vectors can be used as characterization of users, and the interactive objects are objects for the users to perform interactive behaviors and can comprise the clicked, browsed or purchased objects of the users. The initial feature vectors of the candidate objects and the initial feature vectors of the interaction objects are input into a pre-trained large language model, and the strong text extraction capacity of the large language model is utilized to obtain more abundant and diverse feature vectors, namely the feature vectors of the candidate objects and the feature vectors of the interaction objects, so that the effect of subsequent tasks is improved. And carrying out attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects, carrying out weighted summation on the feature vectors of the interaction objects involved in each interaction action of the user to obtain enhanced feature vectors of the user, and carrying out prediction processing based on the enhanced feature vectors of the user and the feature vectors of the candidate objects to obtain first prediction probability of the candidate objects, wherein the first prediction probability is the probability of the recommendation system calculating to obtain the object clicked by the user. And selecting the target object with high first prediction probability as a target object to be recommended to the user based on the prediction probability of each candidate object. The large language model is obtained by pre-training based on the open source large model on the basis of platform data, dedicated optimization is made for the products of the platform, and the input feature vectors can be enriched and characterized, so that the features of candidate objects and the features of interactive objects are fully trained, the expressive of the feature vectors of the objects are improved, more accurate prediction results are obtained, more accurate recommendation is performed by combining the prediction results, the problem that in the prior art, the cold start object is inaccurate is solved, the accuracy of a platform recommendation system is improved, the object recommendation problem under long tail distribution is effectively solved, and the use experience of users is improved.

In some embodiments, attention processing is performed based on feature vectors of each candidate object and feature vectors of each interactive object to obtain enhanced feature vectors of users corresponding to each candidate object, including: calculating the similarity between the feature vector of each candidate object and the feature vector of each interaction object to obtain each first similarity score corresponding to each candidate object; and carrying out weighted summation on the basis of the first similarity scores corresponding to the candidate objects and the feature vectors of the candidate objects to obtain the enhanced feature vectors of the users corresponding to the candidate objects.

According to the method for the attention mechanism, attention degrees of users on different objects can be captured, the first similarity between each interaction object and each candidate object is calculated, the weighted summation is carried out on the representations of each interaction object by utilizing the first similarity to obtain the overall interest representation of the users on all interactions in the current state.

In some embodiments, the predicting processing is performed on each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and the calculating to obtain the first prediction probability of each candidate object includes: splicing the enhanced feature vector of each user and the feature vector of each candidate object to obtain each fusion feature vector; and inputting each fusion feature vector into a multi-layer perceptron, and carrying out prediction processing on each candidate object to obtain a first prediction probability of each candidate object.

In some embodiments, the enhancement feature vector of each user and the feature vector of each candidate object are spliced, and the information of the enhancement feature vector of each user and the information of the feature vector of each corresponding candidate object are fused to obtain each fusion feature vector. The multi-layer perceptron is a neural network formed by a plurality of fully-connected layers, each layer comprises a certain number of neurons, input is activated through a nonlinear activation function (such as ReLU and the like), fusion feature vectors are transmitted layer by layer in the multi-layer perceptron, abstract features of different layers of the input fusion feature vectors are learned and extracted, finally, a first prediction probability for each candidate object is output, wherein the first prediction probability is calculated by a recommendation system to obtain the probability that the user clicks the object, and the probability can be a numerical value between 0 and 1.

In some embodiments, the recommended model structure includes a first embedding layer 301, a second embedding layer 302, a large language model 303, an attention processing module 304, a stitching module 305, and a multi-layer perceptron 306, where candidate object feature data is input into the first embedding layer 301 to perform vector embedding to obtain initial feature vectors of candidate objects, and each interactive object feature data is input into the second embedding layer 302 to perform vector embedding to obtain initial feature vectors of each interactive object. The initial feature vector of the candidate object is input into the large language model 303 to perform feature transformation to obtain the feature vector of the candidate object, and the initial feature vector of each interaction object is input into the large language model 303 to perform feature transformation to obtain the feature vector of each interaction object. And inputting the feature vectors of the candidate objects and the feature vectors of the interaction objects into the attention processing module 304 for attention processing to obtain the enhanced feature vectors of the users corresponding to the candidate objects. And inputting the enhanced feature vector of the user corresponding to the candidate object and the feature vector of the candidate object into a splicing module 305 for splicing to obtain a fusion feature vector. And inputting the fusion feature vector into the multi-layer perceptron 306 to predict the candidate object to obtain a first prediction probability of the candidate object. And selecting a target object from the candidate object data set based on the first prediction probabilities of the respective candidate objects. The large language model 303 is obtained by pre-training based on open-source large language model on the basis of platform data, dedicated optimization is made for a product of a platform, and input feature vectors can be enriched and characterized, so that the features of candidate objects and the features of interactive objects are fully trained, the expressive of the feature vectors of the objects are improved, more accurate prediction results are obtained, more accurate recommendation is performed by combining the prediction results, the problem that cold start object recommendation is inaccurate in the prior art is solved, the accuracy of a platform recommendation system is improved, and the problem of object recommendation under long tail distribution is effectively solved.

In some embodiments, selecting a target object from the candidate object feature data set for user recommendation based on the first prediction probabilities of the respective candidate objects, comprises: comparing the first prediction probability of each candidate object with a preset threshold value; and if the first prediction probability of the candidate object is greater than or equal to a preset threshold value, determining the candidate object as a target object recommended to the user.

In some embodiments, the preset threshold may be a value between 0 and 1, and the specific value is not limited in this disclosure. In the recommendation system, the first prediction probability refers to the probability that the recommendation model predicts the interest or the behavior of the user in a certain candidate object according to the input user behavior sequence characteristic data and the candidate object characteristic data set. The first predictive probability may be a value between 0 and 1 representing the probability that the user clicks, purchases or is interested in the candidate. And if the first prediction probability of a certain candidate object is greater than or equal to a preset threshold value, determining the candidate object as a target object recommended to the user. The candidate object with the first prediction probability being greater than or equal to the preset threshold value is considered to be an object possibly interested by the user, and has high recommendation value. And screening out candidate objects most likely to attract the interest of the user by comparing the first prediction probability of each candidate object with a preset threshold value, and taking the candidate objects as recommended target objects. The screening process described above may improve the accuracy and user experience of the recommendation system, presenting the most relevant and attractive content to the user.

In some embodiments, before performing attention processing based on the feature vector of each candidate object and the feature vector of each interactive object to obtain the enhanced feature vector of the user corresponding to each candidate object, the method further includes: acquiring search words; vector embedding is carried out on the search words, and initial feature vectors of the search words are obtained; performing feature transformation on the initial feature vector of the search word through the large language model to obtain the feature vector of the search word; performing attention processing based on the feature vectors of the candidate objects, the feature vectors of the interaction objects and the feature vectors of the search words to obtain enhanced feature vectors of the candidate objects; predicting each candidate object based on the enhanced feature vector of each candidate object and the feature vector of the search word, and calculating to obtain a second prediction probability of each candidate object; a target object recommended by the oriented user is selected from the candidate object feature data set based on the second prediction probabilities of the respective candidate objects.

In some embodiments, the search terms are input into a search recommendation model, and probabilistic predictions are made for each candidate based on the user behavior sequence feature data, the candidate feature data set, and the search terms. The search term refers to a query condition or a keyword input by a user in a search box in the present disclosure, reflects interest and demand of the user on a certain theme or content, and can be used as important input information in a search recommendation model. In a search recommendation model, search terms may help a computer recognize and understand the intent and interests of a user, thereby providing relevant search results and recommendations to the user. By analyzing the search terms, the search recommendation model can know the attention points of the user, and the results are ordered and filtered. The search recommendation model further comprises a third enabling layer and a large language model, the search words are input into the third enabling layer to conduct feature embedding, the search words are converted into high-dimensional vectors so that the recommendation system can conveniently process the search words, initial feature vectors of the search words are obtained, the initial feature vectors of the search words are input into the large language model to conduct feature transformation, and meaning expression of the feature vectors is enhanced through the large language model, so that feature vectors of the search words are obtained. The method comprises the steps of carrying out attention processing based on feature vectors of candidate objects, feature vectors of interaction objects and feature vectors of search words, focusing on information most relevant to a current task, distributing a weight to each feature vector, carrying out weighted calculation based on each distributed weight and the feature vectors of candidate objects, the feature vectors of interaction objects and the feature vectors of search words, carrying out depth fusion on historical behavior data of users, the current search words and the features of the candidate objects, obtaining more accurate and personalized object vector expression, namely enhancement feature vectors of candidate objects, wherein the size of the weight can be dynamically adjusted along with the change of the feature vectors, so that a search recommendation model can adaptively process different input data, obtain enhancement feature vectors of candidate objects, understand interests and demands of users more comprehensively, help capture dynamic interest changes of users, provide recommendations more in accordance with demands of users, and improve accuracy of recommendation systems. And predicting each candidate object based on the enhanced feature vector of each candidate object and the feature vector of the search word, splicing the enhanced feature vector of each candidate object and the feature vector of the search word to obtain corresponding second fusion feature vectors, and inputting each second fusion feature vector into a multi-layer perceptron of the search recommendation model to predict so as to obtain second prediction probability of each candidate object. The second predicted probability for each candidate may represent a probability that the user clicks, purchases, or is interested in the candidate, and may be a value between 0 and 1. Based on the second prediction probability of each candidate object, selecting a target object recommended to the user from the candidate object feature data set, and compared with the recommendation model in the embodiment, the search recommendation model is further added with search words as input, so that the accuracy of a recommendation system can be further improved, and the use experience of the user is improved.

In some embodiments, the method further includes, before performing prediction processing on each candidate object based on the enhanced feature vector of each candidate object and the feature vector of the search term, calculating a second prediction probability of each candidate object: acquiring user characteristic data; vector embedding is carried out on the user characteristic data to obtain a characteristic vector of the user; predicting each candidate object based on the enhanced feature vector of each candidate object, the feature vector of the user and the feature vector of the search word, and calculating to obtain a third prediction probability of each candidate object; a target object recommended by the oriented user is selected from the candidate object feature data set based on the third prediction probability of each candidate object.

In some embodiments, the search recommendation model further includes a fourth embedded layer, the user feature data including a plurality of discrete features and a plurality of continuous features. In the embodiment of the disclosure, the continuous feature may be the age of the user, the consumption amount of the user for three months, and the like, where the continuous feature refers to that the value of the feature is continuous, and the discrete feature of the user may be a classification variable or a binary variable of the user, for example, the gender of the user, and the code of the user. Hash coding is carried out on each discrete feature of the user to obtain a coding result of each discrete feature of the user, and the coding result of each discrete feature of the user is obtained. And inputting the coding results of the discrete features of the user into a fourth embellishing layer of the search recommendation model for vector embedding, and carrying out vector embedding on each discrete feature coding result to obtain feature vectors of the discrete features of the user through the independent fourth embellishing layer, wherein parameters in each embellishing layer are different. And processing each continuous feature of the user, and respectively carrying out barrel separation processing on each continuous feature of the user to obtain barrel separation results of each continuous feature of the user. The sequential features are binned, i.e., sequential values are mapped into a plurality of finite bins, e.g., the user's age is split into a plurality of bins, such as "0-18 years", "18-25 years", "25-35 years", and so forth. And inputting the barrel division result of each continuous feature of the user into a fourth embellishing layer of the recommendation model to perform vector embedding, so as to obtain feature vectors of each continuous feature of the user. And splicing the feature vector of each discrete feature of the user with the feature vector of each continuous feature of the user to obtain the feature vector of the user. And splicing the enhanced feature vector of each candidate object, the feature vector of the user and the feature vector of the search word to obtain each third fusion feature vector, inputting each third fusion feature vector into a multi-layer perceptron to carry out probability prediction on each candidate object to obtain third prediction probability of each candidate object, and selecting a target object recommended by the oriented user from the candidate object feature data set based on the third prediction probability of each candidate object. In some embodiments, the feature data of the user is input into the search recommendation model, which is beneficial to improving the accuracy of the recommendation system and improving the accuracy of the search recommendation model.

In some embodiments, the structure of the search recommendation model is shown in fig. 4, and includes a first embedding layer 401, a second embedding layer 402, a third embedding layer 403, a fourth embedding layer 404, a large language model 405, an attention processing module 406, a stitching module 407, and a multi-layer perceptron 408, where candidate object feature data is input into the first embedding layer 401 to perform vector embedding to obtain initial feature vectors of candidate objects, each interactive object feature data is input into the second embedding layer 402 to perform vector embedding to obtain initial feature vectors of each interactive object, a search word is input into the third embedding layer 403 to perform vector embedding to obtain initial feature vectors of the search word, and user feature data is input into the fourth embedding layer 404 to perform vector embedding to obtain feature vectors of users. The initial feature vectors of the candidate objects are input into the large language model 405 to perform feature transformation to obtain feature vectors of the candidate objects, the initial feature vectors of the interaction objects are input into the large language model 405 to perform feature transformation to obtain feature vectors of the interaction objects, and the initial feature vectors of the search words are input into the large language model 405 to perform feature transformation to obtain feature vectors of the search words. The feature vectors of the candidate objects, the feature vectors of the interaction objects and the feature vectors of the search words are input into the attention processing module 406 for attention processing, so that the enhanced feature vectors of the candidate objects are obtained. The enhanced feature vector of the candidate object, the feature vector of the search word and the feature vector of the user are input into a splicing module 407 for splicing to obtain a third fusion feature vector, and the third fusion feature vector is input into a multi-layer perceptron 408 for carrying out probability prediction on the candidate object to obtain a third prediction probability of the candidate object. A target object recommended by the oriented user is selected from the candidate object feature data set based on the third prediction probability of each candidate object. The feature data and the search words of the user are input into the search recommendation model, so that the accuracy of a recommendation system is improved, and the accuracy of the search recommendation model is improved.

In some embodiments, after selecting the target object recommended by the oriented user from the candidate object feature data set, further comprises: acquiring feedback information of a user; parameters of the large language model are updated in real time based on the feedback information.

In some embodiments, the interests of the online users are in the short term, so that positive and negative feedback of the users is introduced, the positive and negative feedback of the users in the shopping platform is captured in real time, and parameters of the large language model are updated in real time through reinforcement learning, so that vector expression of object features and search word features is better performed, and the performance of a recommendation system and the accuracy of the online recommendation model can be effectively improved.

Any combination of the above-mentioned optional solutions may be adopted to form an optional embodiment of the present disclosure, which is not described herein in detail.

The following are device embodiments of the present disclosure that may be used to perform method embodiments of the present disclosure. For details not disclosed in the embodiments of the apparatus of the present disclosure, please refer to the embodiments of the method of the present disclosure.

Fig. 5 is a schematic diagram of an object recommendation apparatus according to an embodiment of the present disclosure. As shown in fig. 5, the object recommendation apparatus includes:

an obtaining module 501, configured to obtain user behavior sequence feature data and a candidate object feature data set, where the user behavior sequence feature data includes a plurality of interactive object feature data interacted with by a user;

The vector embedding module 502 is configured to perform vector embedding on feature data of each candidate object to obtain an initial feature vector of each candidate object, and perform vector embedding on feature data of each interactive object to obtain an initial feature vector of each interactive object;

the feature transformation module 503 is configured to perform feature transformation on the initial feature vector of each candidate object through the large language model to obtain a feature vector of each candidate object, and perform feature transformation on the initial feature vector of each interactive object through the large language model to obtain a feature vector of each interactive object;

the attention processing module 504 is configured to perform attention processing based on the feature vectors of each candidate object and the feature vectors of each interactive object, so as to obtain enhanced feature vectors of the users corresponding to each candidate object;

the prediction module 505 is configured to perform prediction processing on each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculate a first prediction probability of each candidate object;

a recommendation module 506 is configured to select a target object recommended by the user from the candidate object feature data set based on the first prediction probability of each candidate object.

According to the technical scheme provided by the embodiment of the disclosure, the vector embedding module 502 is used for carrying out vector embedding on the candidate object feature data to obtain an initial feature vector of the candidate object, and carrying out vector embedding on each interactive object feature data to obtain an initial feature vector of the interactive object, which can be used as a representation of a user, wherein the interactive object is an object of which the user has interactive behaviors and can comprise an object clicked, browsed or purchased by the user. The initial feature vector of the candidate object and the initial feature vector of the interaction object are input to the pre-trained large language model through the feature transformation module 503, and the feature vectors which are richer and more diverse are obtained by utilizing the strong text extraction capability of the large language model, namely the feature vector of the candidate object and the feature vector of the interaction object, so that the effect of the follow-up task is improved. The attention processing module 504 performs attention processing based on the feature vectors of the candidate objects and the feature vectors of the interactive objects, performs weighted summation on the feature vectors of the interactive objects involved in each interactive behavior of the user to obtain enhanced feature vectors of the user, and the prediction module 505 performs prediction processing based on the enhanced feature vectors of the user and the feature vectors of the candidate objects to obtain a first prediction probability of the candidate objects, where the first prediction probability is a probability that the recommendation system calculates to obtain the object clicked by the user. The recommendation module 506 selects, based on the prediction probabilities of the respective candidate objects, the target object with the first prediction probability being high as the recommendation to the user. The large language model is obtained by pre-training based on the open source large model on the basis of platform data, dedicated optimization is made for the products of the platform, and the input feature vectors can be enriched and characterized, so that the features of candidate objects and the features of interactive objects are fully trained, the expressive of the feature vectors of the objects are improved, more accurate prediction results are obtained, more accurate recommendation is performed by combining the prediction results, the problem that in the prior art, the cold start object is inaccurate is solved, the accuracy of a platform recommendation system is improved, the object recommendation problem under long tail distribution is effectively solved, and the use experience of users is improved.

In some embodiments, the attention processing module 504 is configured to calculate a similarity between the feature vector of each candidate object and the feature vector of each interaction object, resulting in each first similarity score corresponding to each candidate object; and carrying out weighted summation on the basis of the first similarity scores corresponding to the candidate objects and the feature vectors of the candidate objects to obtain the enhanced feature vectors of the users corresponding to the candidate objects.

In some embodiments, the prediction module 505 is configured to splice the enhancement feature vector of each user with the feature vector of each candidate object to obtain each fusion feature vector; and inputting each fusion feature vector into a multi-layer perceptron, and carrying out prediction processing on each candidate object to obtain a first prediction probability of each candidate object.

In some embodiments, recommendation module 506 is configured to compare the first prediction probabilities of the respective candidate objects to a preset threshold; and if the first prediction probability of the candidate object is greater than or equal to a preset threshold value, determining the candidate object as a target object recommended to the user.

In some embodiments, the object recommendation apparatus is configured to acquire a search term before performing attention processing based on the feature vector of each candidate object and the feature vector of each interaction object to obtain an enhanced feature vector of a user corresponding to each candidate object; vector embedding is carried out on the search words, and initial feature vectors of the search words are obtained; performing feature transformation on the initial feature vector of the search word through the large language model to obtain the feature vector of the search word; performing attention processing based on the feature vectors of the candidate objects, the feature vectors of the interaction objects and the feature vectors of the search words to obtain enhanced feature vectors of the candidate objects; predicting each candidate object based on the enhanced feature vector of each candidate object and the feature vector of the search word, and calculating to obtain a second prediction probability of each candidate object; a target object recommended by the oriented user is selected from the candidate object feature data set based on the second prediction probabilities of the respective candidate objects.

In some embodiments, the object recommendation device is configured to predict each candidate object based on the enhanced feature vector of each candidate object and the feature vector of the search term, and acquire user feature data before calculating a second prediction probability of each candidate object; vector embedding is carried out on the user characteristic data to obtain a characteristic vector of the user; predicting each candidate object based on the enhanced feature vector of each candidate object, the feature vector of the user and the feature vector of the search word, and calculating to obtain a third prediction probability of each candidate object; a target object recommended by the oriented user is selected from the candidate object feature data set based on the third prediction probability of each candidate object.

In some embodiments, the object recommendation device is configured to obtain feedback information of the user after selecting a target object recommended to the user from the candidate object feature data set; parameters of the large language model are updated in real time based on the feedback information.

It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic of each process, and should not constitute any limitation on the implementation process of the embodiments of the disclosure.

Fig. 6 is a schematic diagram of an electronic device 6 provided by an embodiment of the present disclosure. As shown in fig. 6, the electronic device 6 of this embodiment includes: a processor 601, a memory 602 and a computer program 603 stored in the memory 602 and executable on the processor 601. The steps of the various method embodiments described above are implemented by the processor 601 when executing the computer program 603. Alternatively, the processor 601, when executing the computer program 603, performs the functions of the modules/units of the apparatus embodiments described above.

The electronic device 6 may be a desktop computer, a notebook computer, a palm computer, a cloud server, or the like. The electronic device 6 may include, but is not limited to, a processor 601 and a memory 602. It will be appreciated by those skilled in the art that fig. 6 is merely an example of the electronic device 6 and is not limiting of the electronic device 6 and may include more or fewer components than shown, or different components.

The processor 601 may be a central processing unit (Central Processing Unit, CPU) or other general purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like.

The memory 602 may be an internal storage unit of the electronic device 6, for example, a hard disk or a memory of the electronic device 6. The memory 602 may also be an external storage device of the electronic device 6, for example, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Card (Flash Card) or the like, which are provided on the electronic device 6. The memory 602 may also include both internal and external storage units of the electronic device 6. The memory 602 is used to store computer programs and other programs and data required by the electronic device.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit.

The integrated modules/units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a readable storage medium (e.g., a computer readable storage medium). Based on such understanding, the present disclosure may implement all or part of the flow of the method of the above-described embodiments, or may be implemented by a computer program to instruct related hardware, and the computer program may be stored in a computer readable storage medium, where the computer program, when executed by a processor, may implement the steps of the method embodiments described above. The computer program may comprise computer program code, which may be in source code form, object code form, executable file or in some intermediate form, etc. The computer readable storage medium may include: any entity or device capable of carrying computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth.

The above embodiments are merely for illustrating the technical solution of the present disclosure, and are not limiting thereof; although the present disclosure has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the disclosure, and are intended to be included in the scope of the present disclosure.

Claims (10)

1. A method of object recommendation, comprising:

acquiring user behavior sequence feature data and a candidate object feature data set, wherein the user behavior sequence feature data comprises a plurality of interactive object feature data interacted by the user;

vector embedding is carried out on the candidate object feature data to obtain initial feature vectors of the candidate objects, and vector embedding is carried out on the interactive object feature data to obtain initial feature vectors of the interactive objects;

performing feature transformation on the initial feature vector of each candidate object through a large language model to obtain the feature vector of each candidate object, and performing feature transformation on the initial feature vector of each interaction object through the large language model to obtain the feature vector of each interaction object;

performing attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects to obtain enhanced feature vectors of the users corresponding to the candidate objects;

performing prediction processing on each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculating to obtain a first prediction probability of each candidate object;

A target object is selected from the candidate object feature data set that is oriented towards the user recommendation based on a first prediction probability for each of the candidate objects.

2. The method according to claim 1, wherein the performing attention processing based on the feature vector of each candidate object and the feature vector of each interactive object to obtain the enhanced feature vector of the user corresponding to each candidate object includes:

calculating the similarity between the feature vector of each candidate object and the feature vector of each interaction object to obtain each first similarity score corresponding to each candidate object;

and carrying out weighted summation on the basis of each first similarity score corresponding to each candidate object and the feature vector of each candidate object to obtain the enhanced feature vector of the user corresponding to each candidate object.

3. The method according to claim 1, wherein the predicting each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculating to obtain a first prediction probability of each candidate object, includes:

Splicing the enhancement feature vectors of the users and the feature vectors of the candidate objects to obtain fusion feature vectors;

and inputting each fusion feature vector into a multi-layer perceptron, and carrying out prediction processing on each candidate object to obtain a first prediction probability of each candidate object.

4. The method of claim 1, wherein selecting a target object from the candidate object feature data set that is targeted for the user recommendation based on the first predictive probabilities for each of the candidate objects comprises:

comparing the first prediction probability of each candidate object with a preset threshold value;

and if the first prediction probability of the candidate object is greater than or equal to the preset threshold, determining the candidate object as a target object recommended to the user.

5. The method according to claim 1, wherein before performing attention processing based on the feature vector of each candidate object and the feature vector of each interactive object to obtain the enhanced feature vector of the user corresponding to each candidate object, the method further comprises:

acquiring search words;

Vector embedding is carried out on the search word, and an initial feature vector of the search word is obtained;

performing feature transformation on the initial feature vector of the search word through the large language model to obtain the feature vector of the search word;

performing attention processing based on the feature vectors of the candidate objects, the feature vectors of the interaction objects and the feature vectors of the search words to obtain enhanced feature vectors of the candidate objects;

performing prediction processing on each candidate object based on the enhanced feature vector of each candidate object and the feature vector of the search word, and calculating to obtain a second prediction probability of each candidate object;

a target object is selected from the candidate object feature data set that is oriented towards the user recommendation based on a second prediction probability for each of the candidate objects.

6. The method according to claim 5, wherein before predicting each candidate object based on the enhanced feature vector of each candidate object and the feature vector of the search term, calculating a second prediction probability of each candidate object, further comprises:

Acquiring user characteristic data;

performing vector embedding on the user characteristic data to obtain a characteristic vector of the user;

performing prediction processing on each candidate object based on the enhanced feature vector of each candidate object, the feature vector of the user and the feature vector of the search word, and calculating to obtain a third prediction probability of each candidate object;

a target object from the candidate object feature data set is selected that is oriented towards the user recommendation based on a third prediction probability for each of the candidate objects.

7. The method of object recommendation according to claim 1, further comprising, after selecting a target object from the candidate object feature data set that is oriented to the user recommendation:

acquiring feedback information of a user;

and updating parameters of the large language model in real time based on the feedback information.

8. An object recommendation device, characterized by comprising:

the acquisition module is used for acquiring user behavior sequence characteristic data and a candidate object characteristic data set, wherein the user behavior sequence characteristic data comprises a plurality of interactive object characteristic data interacted by the user;

the vector embedding module is used for carrying out vector embedding on the characteristic data of each candidate object to obtain an initial characteristic vector of each candidate object, and carrying out vector embedding on the characteristic data of each interaction object to obtain an initial characteristic vector of each interaction object;

The feature transformation module is used for carrying out feature transformation on the initial feature vectors of the candidate objects through a large language model to obtain feature vectors of the candidate objects, and carrying out feature transformation on the initial feature vectors of the interaction objects through the large language model to obtain feature vectors of the interaction objects;

the attention processing module is used for carrying out attention processing based on the feature vectors of the candidate objects and the feature vectors of the interaction objects to obtain enhanced feature vectors of the users corresponding to the candidate objects;

the prediction module is used for carrying out prediction processing on each candidate object based on the enhanced feature vector of each user and the feature vector of each candidate object, and calculating to obtain a first prediction probability of each candidate object;

and the recommendation module is used for selecting a target object recommended by the user from the candidate object characteristic data set based on the first prediction probability of each candidate object.

9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 7 when the computer program is executed.

10. A readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 7.