CN112508613A

CN112508613A - Commodity recommendation method and device, electronic equipment and readable storage medium

Info

Publication number: CN112508613A
Application number: CN202011446682.0A
Authority: CN
Inventors: 张道甜; 刘宁东; 杨林; 唐明利
Original assignee: Tianjin Shituo Information Technology Co ltd
Current assignee: Tianjin Shituo Information Technology Co ltd
Priority date: 2020-12-09
Filing date: 2020-12-09
Publication date: 2021-03-16
Anticipated expiration: 2040-12-09
Also published as: CN112508613B

Abstract

The invention discloses a commodity recommendation method, a commodity recommendation device, electronic equipment and a readable storage medium, wherein the method comprises the following steps: acquiring target user log data, generating a first data set based on the target user log data, and inputting the first data set into the trained click rate prediction model to obtain a predicted first click rate; generating a second data set based on the target user log data and the first click rate, and inputting the second data set into a trained conversion rate prediction model to obtain a predicted conversion rate; and recommending the commodities to the target user according to the predicted first click rate and the predicted conversion rate. The method utilizes the idea of multi-target learning, simultaneously considers the click rate and the conversion rate, and obviously improves the recommendation quality; the method realizes multi-model fusion, inputs the predicted value of the click rate into the conversion rate model as the characteristic, constructs a new characteristic, effectively utilizes the result of click rate model learning, transmits the result to the conversion rate model, and improves the generalization ability of the model.

Description

Commodity recommendation method and device, electronic equipment and readable storage medium

Technical Field

The invention relates to the field of computer distributed processing, in particular to a commodity recommendation method, a commodity recommendation device, electronic equipment and a readable storage medium.

Background

Artificial intelligence technology has been integrated into the lives of the public, and various internet platforms appear to help users to find commodities with higher cost performance quickly in order to meet the purchase demands of the users, personalized recommendation needs to be performed, and commodities which the users may purchase are recommended to the users, so that the Click Rate (ratio of actual Click times of advertisements to the display amount of the advertisements) and the Conversion Rate (Conversion Rate, CVR, Conversion Rate of users who Click advertisements to become effective activation, registration or even paying users) of the commodities are improved. However, when commodity recommendation is made on some current platforms, some are based on human experience and manual labels to judge user preferences, some are based on click rate prediction recommendation through some simple linear models or tree models (such as logistic regression and decision trees), the models and methods have limitations, a large amount of manpower is needed to label users, the accuracy of labels cannot be guaranteed, meanwhile, a simple machine learning algorithm cannot fully represent attribute characteristics and behavior characteristics of the users, the matching degree of recommended commodities and the users is poor, only the click rate prediction model is considered, the conversion rate is not considered during modeling, the commodity click rate is high, but the conversion rate is not necessarily high, the recommendation effect is poor, and user experience on the platform is influenced.

Disclosure of Invention

In view of the above problems, the present invention has been made to provide an article recommendation method, apparatus, electronic device and readable storage medium that overcome or at least partially solve the above problems.

According to an aspect of the present invention, there is provided a commodity recommendation method, the method including:

acquiring target user log data, generating a first data set based on the target user log data, and inputting the first data set into the trained click rate prediction model to obtain a predicted first click rate;

generating a second data set based on the target user log data and the first click rate, and inputting the second data set into a trained conversion rate prediction model to obtain a predicted conversion rate;

recommending commodities to the target user according to the predicted first click rate and the predicted conversion rate.

Optionally, the click rate prediction model is a LightGBM model capable of realizing distributed gradient lifting based on a Histogram decision tree algorithm, and the conversion rate prediction model is a deep FM model that combines DNN and FM advantages and learns low-order and high-order combination characteristics at the same time.

Optionally, the trained click rate prediction model is obtained by training as follows:

the method comprises the steps of obtaining user log data in a first preset time period, determining user attribute features, commodity attribute features, user behavior data and labels whether to click or not which have corresponding relations based on the user log data, generating a first training set, and training a pre-constructed click rate prediction model by using the first training set, so that the trained click rate prediction model is obtained.

Optionally, the trained conversion rate prediction model is obtained by training in the following way:

generating a third data set based on the user log data in the second preset time period, and inputting the third data set into the trained click rate prediction model to obtain a predicted second click rate;

generating a second training set based on the user log data in the second preset time period, the converted label and the second click rate, and training a pre-constructed conversion rate prediction model by using the second training set so as to obtain a trained conversion rate prediction model;

wherein the second preset time period is closer to a point in time at which recommendation is made than the first preset time period.

Optionally, the recommending the goods to the target user according to the predicted first click rate and the predicted conversion rate includes:

calculating the product of the predicted first click rate and the predicted conversion rate, sorting the commodities according to the size of the product, and recommending one or more commodities to the target user; alternatively, the first and second electrodes may be,

calculating an average value of the predicted first click rate and the predicted conversion rate, sorting the commodities according to the size of the average value, and recommending one or more commodities to the target user.

Optionally, the generating a first data set based on the target user log data comprises:

determining target user attribute characteristics, recalled commodity attribute characteristics and behavior data aiming at recalled commodities which have corresponding relations based on the target user log data, and combining to form the first data set according to whether a user clicks a tag which is determined to be clicked or not;

wherein the target user attribute characteristics include at least one of: user attribution, age, gender, registration time, asset size, total browsing duration, total browsing times, purchasing times or total consumption amount;

the recalled commodity refers to a commodity browsed, clicked, collected, added to a shopping cart or purchased by a target user, and the attribute characteristics of the recalled commodity include at least one of the following characteristics: brand, model, type, place of origin or location;

the behavioral characteristics for recalled merchandise include at least one of: browse, click, collect, join a shopping cart, forward, consult, or purchase.

Optionally, the target user is a user who triggers a recall commodity purchase recommendation request; the recalled merchandise is merchandise that is in a state of being able to be purchased by the target user to control the quantity of the recalled merchandise; the behavior data aiming at the recalled commodity is behavior data of historical users to the recalled commodity.

According to another aspect of the present invention, there is provided an article recommendation apparatus including:

the click rate determining unit is used for acquiring target user log data, generating a first data set based on the target user log data, and inputting the first data set into the trained click rate prediction model to obtain a predicted first click rate;

a conversion rate determining unit which generates a second data set based on the target user log data and the first click rate, inputs the second data set to a trained conversion rate prediction model and obtains a predicted conversion rate;

and the commodity recommending unit recommends commodities to the target user according to the predicted first click rate and the predicted conversion rate.

In accordance with still another aspect of the present invention, there is provided an electronic apparatus including: a processor; and a memory arranged to store computer executable instructions that, when executed, cause the processor to perform a method as any one of the above.

According to a further aspect of the invention, there is provided a computer readable storage medium, wherein the computer readable storage medium stores one or more programs which, when executed by a processor, implement a method as any one of the above.

As can be seen from the above, the technical solution of the present invention can obtain the following beneficial effects:

according to the multi-target learning idea, double indexes of click rate and conversion rate are considered, and the recommended conversion rate is obviously improved. The single-target idea easily causes poor recommendation effect, for example, if only the click rate is considered, the click rate is high, the conversion rate is low, or if only the conversion rate is considered, the data sparsity is brought, and the recommendation effect is affected.

According to the multi-model fusion method, the predicted value of the click rate is input into the conversion rate model as the characteristic, on one hand, a new effective characteristic is constructed, on the other hand, the result experience of click rate model learning can be effectively utilized and transmitted to the conversion rate model, and the generalization capability of the model is improved.

Furthermore, the click rate model is preferably a LightGBM model, the LightGBM is based on a Historgram decision tree algorithm, and compared with a pre-sorted algorithm, the Histogram has advantages in memory consumption and calculation cost, the occupied memory is lower, and the complexity of data separation is lower; the conversion rate model is preferably a Deep FM model, Deep and FM are combined, FM is used as a low-order combination between features, Deep NN part is used as a high-order combination between features, and the two methods are combined in a parallel mode, so that the final architecture has the following characteristics: obtaining a hidden vector without pre-training FM; manual feature engineering is not required; the combined features of the low order and the high order can be learned simultaneously; the FM module and the Deep module share the Feature Embedding part, so that the training can be faster, the training and the learning can be more accurate, and more importantly, the two models are combined to obtain unexpected accuracy improvement.

And the modeling is carried out based on the user attributes, the commodity characteristics and the user historical behavior data, so that the manual marking process is reduced, and the problem that the accuracy of commodity recommendation is influenced by personal subjective factors is avoided.

The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.

Drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:

FIG. 1 is a flow chart diagram illustrating a method of merchandise recommendation in accordance with one embodiment of the present invention;

fig. 2 is a schematic structural view showing an article recommending apparatus according to an embodiment of the present invention;

FIG. 3 shows a schematic structural diagram of an electronic device according to one embodiment of the invention;

FIG. 4 shows a schematic structural diagram of a computer-readable storage medium according to one embodiment of the invention;

FIG. 5 illustrates a flow diagram of click rate acquisition according to one embodiment of the invention;

FIG. 6 shows a schematic flow diagram of conversion yield acquisition according to one embodiment of the present invention.

Detailed Description

Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

Fig. 1 is a schematic flow chart of a product recommendation method according to an embodiment of the present invention, where the method includes:

s110, obtaining target user log data, and generating a first data set based on the target user log data. The target user log data can be obtained from a distributed website platform or an application platform through a buried point, and feature data of user attributes (which can be recorded as user _ id) such as belonged place, age, sex, asset scale, browsing situation and the like can be obtained from the log data, and also commodity attribute features (which can be recorded as item _ id) to be recommended and recalled, such as commodity category, brand, model, belonged place and the like which are interesting to the user, and behavior data (which can be recorded as features) of the recalled commodity by the user, wherein the behavior data is preferably behavior data of historical users on the recalled commodity in a certain past time period, such as browsing times, browsing duration, collection, shopping cart adding, forwarding, consultation or purchasing times and the like, and certainly, data of whether the commodity clicks an incoming label and the like can be included, and the data are combined to form a first data set.

Then, the first data set is input into the trained click rate prediction model, and a predicted first click rate is obtained.

And S120, generating a second data set based on the target user log data and the first click rate, and inputting the second data set into a trained conversion rate prediction model to obtain a predicted conversion rate.

Of course, the second data set includes attribute features of the target user, attribute features of the recalled goods, behavior data, and the like.

It is worth to be noted that the second data set includes the first click rate corresponding to the target user, the idea of multi-target learning is embodied, the two models are fused, and the conversion rate prediction model can also react to the click rate.

S130, recommending commodities to the target user according to the predicted first click rate and the predicted conversion rate.

In the step, the recalled commodities are sorted according to the comprehensive high and low of the first click rate and the conversion rate, and the commodities are recommended to the user according to the sorting condition.

According to the technical scheme, the idea of multi-target learning is utilized, the click rate and the conversion rate are considered, and the recommendation quality is obviously improved; the method realizes multi-model fusion, inputs the predicted value of the click rate into the conversion rate model as the characteristic, constructs a new characteristic, effectively utilizes the result of click rate model learning, transmits the result to the conversion rate model, and improves the generalization ability of the model.

In a preferred embodiment, the click-through rate prediction model is a LightGBM model capable of realizing distributed gradient lifting based on a Histogram decision tree algorithm. Light Gradient Boosting Machine (LightGBM) is a Decision Tree algorithm based distributed Gradient Boosting (GBDT) framework sourced by the microsoft asia institute of distributed Machine learning toolkit (DMTK) team. The method is based on a decision tree algorithm of the Histogram, continuous floating point features are discretized into k discrete values, and the Histogram with the width of k is constructed. Then, the training data is traversed, and the cumulative statistics of each discrete value in the histogram are counted. When the feature selection is carried out, the optimal segmentation point is searched in a traversing way only according to the discrete value of the histogram; compared with a pre-sorted algorithm, the histogram has advantages in memory consumption and calculation cost, occupies lower memory and has lower complexity of data separation. Based on the histogram advantages, the following advantages can be obtained: reducing the calculation amount of the segmentation gain; training of the model is further accelerated by histogram subtraction: the histograms of the leaf nodes can be obtained by subtracting the histograms of the parent node and the adjacent node of the leaf node in the binary tree, and the cost is low; the use of the memory is reduced; the communication cost of parallel learning is reduced.

Moreover, the LightGBM adopts a leaf-wise growth strategy, finds out one leaf with the maximum splitting gain (generally, the maximum data volume) from all the current leaves at a time, then splits and circulates; but a deeper decision tree is grown to generate overfitting, so LightGBM adds a maximum depth display over leaf-wise to prevent overfitting while ensuring high efficiency.

The conversion rate prediction model is a DeepFM model which combines the advantages of DNN and FM and simultaneously learns the combined characteristics of low order and high order.

The Deep FM can be regarded as an algorithm derived from the FM, Deep and FM are combined, FM is used as a low order combination between features, Deep NN part is used as a high order combination between features, and the two methods are combined in a parallel mode, so that the final architecture has the following characteristics: obtaining a hidden vector without pre-training FM; manual feature engineering is not required; the combined features of the low order and the high order can be learned simultaneously; the FM module and the Deep module share the Feature Embedding part, so that the training can be faster, and the training and learning can be more accurate.

The network structure of the Deep FM model firstly divides the Deep FM model into a Deep neural network part and an FM factorization machine part, wherein the Deep neural network part can adopt a fully-connected feedforward neural network DNN, the DNN and the FM divide input user characteristics and attribute characteristics into a plurality of characteristic groups, each characteristic group corresponds to an embedding (embedding) vector, a characteristic splicing layer (concat) of the Deep neural network part splices all the embedding vectors, and then fully-connected layers (Fc (relu)) of two layers are added to realize the combination of high-order characteristics; the FM factorization machine carries out weighted summation (addition) on input original feature input such as user features and attribute features, and extracts feature combinations through an embedding vector inner product of each dimension to realize the combination of low-order features; and finally, combining the outputs of the Deep neural network and the FM factorization machine to obtain a prediction result (sigmoid).

The preferred embodiment combines the two excellent prediction models innovatively, the combined model has better generalization capability than a single model, the idea of ensemble learning is adopted, the advantages of various models are fully utilized, and the phenomena of large prediction deviation and overfitting of the single model are avoided.

In a preferred embodiment, referring to fig. 5, the trained click-through rate prediction model is obtained by training as follows:

obtaining historical user log data in a first preset time period, such as the last 30 days to 60 days, determining user attribute features with corresponding relations, commodity attribute features interested by the user and user behavior data of an operation data set of the user on the commodity based on the historical user log data in the time period, marking whether the user clicks the commodity details to form a label whether the user clicks, combining the label and the label to generate a first training set, and training a pre-constructed click rate prediction model, such as a LightGBM model, by using the first training set to obtain the trained click rate prediction model.

In a preferred embodiment, as can be seen from fig. 5 and 6, the trained conversion rate prediction model is obtained by training as follows:

according to fig. 5, a third data set is generated based on the user log data in the second preset time period, for example, the last 30 days, and the third data set may also include data items such as user attribute features, commodity attribute features, and behavior data, and then the third data set is input to the trained click rate prediction model to obtain a predicted second click rate.

Then, with reference to fig. 6, based on the data items of the user log data in the second preset time period, such as the user attribute features, the commodity attribute features, the behavior data, and the like, the label of whether to convert and the second click rate are spliced and combined to generate a second training set, and the second training set is used to train a pre-constructed conversion rate prediction model, so as to obtain a trained conversion rate prediction model.

Of course, the lengths and intervals of the first preset time period and the second preset time period are not limited to this, as long as the time periods that can achieve the training purpose are within the protection range if the conditions are satisfied, but the conditions that the second preset time period is closer to the time point when recommending to the target user than the first preset time period need to be satisfied.

In one embodiment, the recommending, to the target user, the item according to the predicted first click rate and the predicted conversion rate in S130 includes:

Of course, according to the actual situation, the recommendation can be made to the target user in a weighting manner or in a manner of some independent score.

In some specific embodiments, the generating a first data set based on the target user log data comprises:

wherein the target user attribute characteristics include at least one of: user attribution, age, gender, registration time, asset size, total browsing duration, total browsing times, purchasing times or total consumption amount; the recalled commodity refers to a commodity browsed, clicked, collected, added to a shopping cart or purchased by a target user, and the attribute characteristics of the recalled commodity include at least one of the following characteristics: brand, model, type, place of origin or location; the behavioral characteristics for recalled merchandise include at least one of: browse, click, collect, join a shopping cart, forward, consult, or purchase.

Further, according to the requirements of the training, the following can be determined: the target user is a user triggering a recall commodity purchase recommendation request; the recalled merchandise is merchandise that is in a state of being able to be purchased by the target user to control the quantity of the recalled merchandise; the behavior data for recalled merchandise is behavior data for historical users, such as within the last 30-60 days, for the recalled merchandise. In the above situation, it is easier to obtain a model for the purpose of predicting a click rate with higher accuracy.

Fig. 2 is a schematic structural diagram illustrating an article recommendation device 200 according to an embodiment of the present invention, where the device 200 includes:

the click rate determining unit 210 is adapted to obtain target user log data, and generate a first data set based on the target user log data. The target user log data can be obtained from a distributed website platform or an application platform through a buried point, and feature data of user attributes (marked as user _ id) such as belonged places, ages, sexes, asset scales, browsing situations and the like can be obtained from the log data, and attribute features (marked as item _ id) of commodities to be recommended and recalled, which are interesting to users, such as commodity types, brands, models, belonged places and the like, and behavior data (marked as features) of the recalled commodities are also obtained from the log data.

A conversion rate determining unit 220 adapted to generate a second data set based on the target user log data and the first click rate, input the second data set to a trained conversion rate prediction model, and obtain a predicted conversion rate.

The second data set comprises target user attribute characteristics, recalled commodity attribute characteristics, behavior data and the like.

A commodity recommending unit 230 adapted to recommend commodities to the target user according to the predicted first click rate and the predicted conversion rate.

In the unit, the recalled commodities are sorted according to the comprehensive high and low of the first click rate and the conversion rate, and commodity recommendation is carried out on the commodities to the user according to the sorting condition.

In a preferred embodiment, the click-through rate prediction model is a LightGBM model capable of realizing distributed gradient lifting based on a Histogram decision tree algorithm.

In a preferred embodiment, referring to fig. 5, in the click rate determining unit 210, the trained click rate prediction model is obtained by training as follows:

In a preferred embodiment, as can be seen from fig. 5 and fig. 6, in the conversion rate determining unit 220, the trained conversion rate prediction model is obtained by training as follows:

In one embodiment, the goods recommending unit 230 is specifically adapted to:

In some specific embodiments, the click-through rate determining unit 210 is specifically adapted to:

It should be noted that, for the specific implementation of each apparatus embodiment, reference may be made to the specific implementation of the corresponding method embodiment, which is not described herein again.

In summary, according to the technical scheme of the present invention, first, target user log data is obtained, a first data set is generated based on the target user log data, and the first data set is input to the trained click rate prediction model to obtain a predicted first click rate; then generating a second data set based on the target user log data and the first click rate, and inputting the second data set into a trained conversion rate prediction model to obtain a predicted conversion rate; and finally, recommending commodities to the target user according to the predicted first click rate and the predicted conversion rate. According to the technical scheme, according to the thought of multi-target learning, the double indexes of click rate and conversion rate are considered at the same time, and the recommended conversion rate is obviously improved; and according to a multi-model fusion method, the predicted value of the click rate is input into the conversion rate model as a feature, on one hand, a new effective feature is constructed, on the other hand, the result experience of the click rate model learning can be effectively utilized and transmitted to the conversion rate model, and the generalization capability of the model is improved.

It should be noted that:

the algorithms and displays presented herein are not inherently related to any particular computer, virtual machine, or other apparatus. Various general purpose devices may be used with the teachings herein. The required structure for constructing such a device will be apparent from the description above. Moreover, the present invention is not directed to any particular programming language. It is appreciated that a variety of programming languages may be used to implement the teachings of the present invention as described herein, and any descriptions of specific languages are provided above to disclose the best mode of the invention.

In the description provided herein, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the invention and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: that the invention as claimed requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.

Those skilled in the art will appreciate that the modules in the device in an embodiment may be adaptively changed and disposed in one or more devices different from the embodiment. The modules or units or components of the embodiments may be combined into one module or unit or component, and furthermore they may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.

The various component embodiments of the invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It will be appreciated by those skilled in the art that a microprocessor or Digital Signal Processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components of the merchandise recommendation device according to embodiments of the present invention. The present invention may also be embodied as apparatus or device programs (e.g., computer programs and computer program products) for performing a portion or all of the methods described herein. Such programs implementing the present invention may be stored on computer-readable media or may be in the form of one or more signals. Such a signal may be downloaded from an internet website or provided on a carrier signal or in any other form.

For example, fig. 3 shows a schematic structural diagram of an electronic device according to an embodiment of the invention. The electronic device 300 comprises a processor 310 and a memory 320 arranged to store computer executable instructions (computer readable program code). The memory 320 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read only memory), an EPROM, a hard disk, or a ROM. The memory 320 has a storage space 330 storing computer readable program code 331 for performing any of the method steps described above. For example, the storage space 330 for storing the computer readable program code may comprise respective computer readable program codes 331 for respectively implementing various steps in the above method. The computer readable program code 331 may be read from or written to one or more computer program products. These computer program products comprise a program code carrier such as a hard disk, a Compact Disc (CD), a memory card or a floppy disk. Such a computer program product is typically a computer readable storage medium such as described in fig. 4. Fig. 4 shows a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present invention. The computer readable storage medium 400 has stored thereon a computer readable program code 331 for performing the steps of the method according to the invention, readable by a processor 310 of the electronic device 300, which computer readable program code 331, when executed by the electronic device 300, causes the electronic device 300 to perform the steps of the method described above, in particular the computer readable program code 331 stored on the computer readable storage medium may perform the method shown in any of the embodiments described above. The computer readable program code 331 may be compressed in a suitable form.

It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The usage of the words first, second and third, etcetera do not indicate any ordering. These words may be interpreted as names.

Claims

1. A method for recommending an article, the method comprising:

acquiring target user log data, generating a first data set based on the target user log data, and inputting the first data set into a trained click rate prediction model to obtain a predicted first click rate;

2. The method of claim 1, wherein the click-through rate prediction model is a LightGBM model capable of achieving distributed gradient boosting based on a Histogram decision tree algorithm, and the conversion rate prediction model is a deep FM model that combines DNN and FM advantages to simultaneously learn low-order and high-order combination features.

3. The method of claim 1, wherein the trained click-through rate prediction model is trained as follows:

4. The method of claim 3, wherein the trained conversion prediction model is obtained by training as follows:

generating a third data set based on user log data in a second preset time period, and inputting the third data set into the trained click rate prediction model to obtain a predicted second click rate;

5. The method of any of claims 1-4, wherein said recommending a good to the target user based on the predicted first click rate and the predicted conversion rate comprises:

6. The method of any of claims 1-4, wherein generating the first data set based on the target user log data comprises:

7. The method of claim 6, wherein the target user is a user who triggers a recall merchandise purchase recommendation request; the recalled merchandise is merchandise that is in a state of being able to be purchased by the target user to control the quantity of the recalled merchandise; the behavior data aiming at the recalled commodity is behavior data of historical users to the recalled commodity.

8. An article recommendation device, the device comprising:

the click rate determining unit is used for acquiring target user log data, generating a first data set based on the target user log data, and inputting the first data set into a trained click rate prediction model to obtain a predicted first click rate;

9. An electronic device, wherein the electronic device comprises: a processor; and a memory arranged to store computer-executable instructions that, when executed, cause the processor to perform the method of any one of claims 1-7.

10. A computer readable storage medium, wherein the computer readable storage medium stores one or more programs which, when executed by a processor, implement the method of any of claims 1-7.