Recommended model training and recommendation method and device, storage medium and electronic equipment

Abstract

Embodiments of the present disclosure relate to the technical field of computer data processing, and more particularly, to a recommendation model training and recommendation method and device, a storage medium, and an electronic device. The method comprises: establishing a long-short-time index relationship model, making the input of the long-short-time index relationship model include a short-time behavior feature of a user on a to-be-processed object, and making the output of the long-short-time index relationship model be a long-time index ranking relationship value of the user on the to-be-processed object; training the long-short-time index relationship model with the goal that the ranking order of the long-time index value is consistent with the ranking order of the corresponding long-time behavior feature; and training a recommendation model based on the long-short-time index relationship model. The technical solution of the present disclosure can combine the influence of long-time behavior features into the training of the recommendation model, thereby optimizing the recommendation effect of the recommendation model.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of computer information processing technology. More specifically, the embodiments of the present disclosure relate to recommendation model training and recommendation methods and devices, storage media, and electronic devices. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the disclosure that are recited in the claims, and no statement herein is admitted to be prior art by inclusion in this section.

[0003] With the development of Internet information technology, recommendation systems have emerged. Recommendation systems are mainly based on user-related information, giving priority to providing or pushing the information that users need, thereby improving the efficiency of users' information acquisition. Summary of the Invention

[0004] Based on the collected information, the recommendation system outputs predicted values ​​for multiple objects. These values ​​can be calculated using a formula to obtain a recommendation score for the object. The recommendation system can then sort and recommend objects based on their recommendation scores, giving objects with higher recommendation scores higher display priority.

[0005] Current recommendation systems use recommendation models to obtain object prediction values, which are generally based on predicted short-term user behavior characteristics. In actual object recommendation, incorporating long-term user behavior characteristics into the calculation of recommendation scores would further improve the efficiency of obtaining user information. However, the data on long-term user behavior characteristics is more sparse than that on short-term behavior characteristics. Using sparse long-term user behavior characteristics to train a recommendation model results in poor prediction performance. Furthermore, it is difficult to incorporate long-term behavior characteristics into the calculation of recommendation values ​​by optimizing the recommendation model for short-term behavior characteristics.

[0006] To this end, there is a great need for an improved recommendation model training and recommendation method and device, storage medium, and electronic device that can incorporate the influence of long-term behavioral characteristics into the recommendation model training, thereby optimizing the recommendation effect of the recommendation model.

[0007] In this context, the embodiments of the present invention are intended to provide a recommendation model training and recommendation method and device, storage medium, electronic device

[0008] According to one aspect of the present disclosure, a recommendation model training method is provided, comprising:

[0009] Establishing a long-short-time indicator relationship model, wherein the input of the long-short-time indicator relationship model includes the short-time behavior characteristics of the user on the object to be processed, and the output of the long-short-time indicator relationship model is the long-time indicator ranking relationship value of the user on the object to be processed;

[0010] Training the long- and short-term indicator relationship model with the goal of ensuring that the sorting order of the long-term indicator values ​​is consistent with the sorting order of the corresponding long-term behavioral features;

[0011] Based on the long-short-time indicator relationship model, the recommendation model is trained so that the input of the recommendation model includes the user characteristics of the user and the object characteristics of the object to be processed, and the output of the recommendation model is the recommendation value of the object to be processed to the user.

[0012] According to one aspect of the present disclosure, there is provided an object recommendation method, comprising:

[0013] Inputting the user characteristics of the user and the object characteristics of the object to be processed into the recommendation model to obtain a plurality of recommendation values ​​of the object to be processed for the user, wherein the recommendation model is trained by the above method;

[0014] Sorting the plurality of objects to be processed according to the recommendation values ​​to obtain a sorting order of the objects to be processed;

[0015] The objects to be processed are recommended to the user according to the sorting order.

[0016] According to one aspect of the present disclosure, a recommendation model training device is provided, comprising:

[0017] An establishment module is used to establish a long-short-time indicator relationship model, so that the input of the long-short-time indicator relationship model includes the short-time behavior characteristics of the user on the object to be processed, and the output of the long-short-time indicator relationship model is the long-time indicator ranking relationship value of the user on the object to be processed;

[0018] A long-short-time indicator relationship model training module is used to train the long-short-time indicator relationship model with the goal of ensuring that the sorting order of the long-time indicator values ​​is consistent with the sorting order of the corresponding long-time behavior features;

[0019] The recommendation model training module is used to train the recommendation model based on the long-short-time indicator relationship model, so that the input of the recommendation model includes the user characteristics of the user and the object characteristics of the object to be processed, and the output of the recommendation model is the recommendation value of the object to be processed to the user.

[0020] According to one aspect of the present disclosure, there is provided an object recommendation device, comprising:

[0021] A recommendation value acquisition module, configured to input the user characteristics of the user and the object characteristics of the object to be processed into a recommendation model to obtain a plurality of recommendation values ​​of the object to be processed for the user, wherein the recommendation model is trained using the method described above;

[0022] a sorting module, configured to sort the plurality of objects to be processed according to the recommendation values ​​to obtain a sorting order of the objects to be processed;

[0023] A recommendation module is used to recommend the objects to be processed to the user according to the sorting order.

[0024] According to one aspect of the present disclosure, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned recommendation model training method and / or object recommendation method is implemented.

[0025] According to one aspect of the present disclosure, there is provided an electronic device, including:

[0026] processor; and

[0027] a memory for storing executable instructions of the processor;

[0028] The processor is configured to execute any one of the above-mentioned recommendation model training methods and / or object recommendation methods by executing the executable instructions.

[0029] According to the recommendation model training method and object recommendation method of the embodiment of the present disclosure,

[0030] By establishing and training a long-term and short-term indicator relationship model, it is possible to learn the long-term indicator ranking relationship value of the long-term behavior feature based on the short-term behavior feature, and train the recommendation model based on the long-term indicator ranking relationship value to learn the influence of the long-term indicator ranking relationship value of the long-term behavior feature on the object recommendation, thereby incorporating the influence of the long-term behavior feature on the object recommendation into the recommendation calculation of the recommendation model. Therefore, on the one hand, the present disclosure only needs to learn the long-term indicator ranking relationship value, and does not need to directly predict the long-term behavior feature. Compared with the long-term behavior feature, the long-term indicator ranking relationship value has lower accuracy requirements for the prediction model, which is conducive to the implementation of the long-term and short-term indicator relationship model; on the other hand, the influence of the long-term behavior feature on the object recommendation is incorporated into the recommendation calculation of the recommendation model to improve the recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood by reading the detailed description below with reference to the accompanying drawings, in which several embodiments of the present disclosure are shown by way of example and not limitation, wherein:

[0032] Figure 1 The following schematically shows a flow chart of a recommendation model training method according to an embodiment of the present disclosure;

[0033] Figure 2Schematically shows a schematic diagram of a training long-short time indicator relationship model according to an embodiment of the present disclosure;

[0034] Figure 3 Schematically shows a schematic diagram of a training long-short time indicator relationship model according to an embodiment of the present disclosure;

[0035] Figure 4 Schematically shows a schematic diagram of a training recommendation model according to an embodiment of the present disclosure;

[0036] Figure 5 Schematically shows a schematic diagram of a training recommendation model according to an embodiment of the present disclosure;

[0037] Figure 6 The following schematically shows a flow chart of an object recommendation method according to an embodiment of the present disclosure;

[0038] Figure 7 Schematically shows a schematic diagram of using a recommendation model according to an embodiment of the present disclosure;

[0039] Figure 8 Schematically shows a module diagram of a recommendation model training device according to an embodiment of the present disclosure;

[0040] Figure 9 Schematically shows a module diagram of an object recommendation device according to an embodiment of the present disclosure;

[0041] Figure 10 A schematic diagram illustrating a storage medium according to an embodiment of the present disclosure; and

[0042] Figure 11 A block diagram of an electronic device according to a disclosed embodiment is schematically shown.

[0043] In the drawings, the same or corresponding reference numerals denote the same or corresponding parts. DETAILED DESCRIPTION

[0044] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make this disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0045] Those skilled in the art will appreciate that the embodiments of the present disclosure may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present disclosure may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0046] According to an embodiment of the present disclosure, a recommendation model training and recommendation method and device, a storage medium, and an electronic device are provided.

[0047] In this document, any number of elements in the drawings is for illustration and not for limitation, and any naming is for distinction only and does not have any limiting meaning.

[0048] The principles and spirit of the present disclosure are described in detail below with reference to several representative embodiments of the present disclosure. SUMMARY

[0050] The inventors discovered that a recommendation system, based on collected information, outputs multiple-dimensional object predictions. These outputted values ​​can be calculated using a formula to obtain an object's recommendation score. The recommendation system can then sort and recommend objects based on their recommendation scores, giving objects with higher recommendation scores higher display priority.

[0051] Current recommendation systems use recommendation models to obtain object prediction values, which are generally predicted short-term user behavior characteristics, such as the probability of users clicking on an object and the length of time they read within a set short time period. In actual object recommendations, if the long-term user behavior characteristics are also included in the calculation of the recommendation value, the efficiency of obtaining user information will be further improved. However, the data of long-term user behavior characteristics is sparser than that of short-term behavior characteristics. Using sparse long-term user behavior characteristics to train a recommendation model will result in poor prediction performance of the recommendation model. At the same time, due to the different needs and scenarios of users for objects, short-term behavior characteristics cannot replace long-term behavior characteristics. Therefore, it is also difficult to incorporate long-term behavior characteristics into the calculation of recommendation values ​​by optimizing the recommendation model of short-term behavior characteristics.

[0052] In view of the above, the basic idea of ​​the present invention is: by establishing and training a long-term and short-term indicator relationship model, it is possible to learn the long-term indicator ranking relationship value of the long-term behavior feature based on the short-term behavior feature, and train the recommendation model based on the long-term indicator ranking relationship value, so as to be able to learn the influence of the long-term indicator ranking relationship value of the long-term behavior feature on the object recommendation, thereby incorporating the influence of the long-term behavior feature on the object recommendation into the recommendation calculation of the recommendation model. Therefore, on the one hand, the present invention only needs to learn the long-term indicator ranking relationship value, and does not need to directly predict the long-term behavior feature. The long-term indicator ranking relationship value has lower accuracy requirements for the prediction model than the long-term behavior feature, which is conducive to the realization of the long-term and short-term indicator relationship model; on the other hand, the present invention incorporates the influence of the long-term behavior feature on the object recommendation into the recommendation calculation of the recommendation model to improve the recommendation effect.

[0053] After introducing the basic principles of the present application, the various non-limiting embodiments of the present application will be specifically introduced below.

[0054] Exemplary methods

[0055] The recommendation model training method according to the exemplary embodiments of the present disclosure will be described below in combination with Figure 1 Reference is made to FIG. 1, which shows a flowchart of the recommendation model training method according to the exemplary embodiments of the present disclosure. The recommendation model training method can include the following steps: Figure 1

[0056] Step S110: establishing a long-short-time index relationship model, making the input of the long-short-time index relationship model include short-time behavior characteristics of a user on a to-be-processed object, and making the output of the long-short-time index relationship model be a long-time index ranking relationship value of the user on the to-be-processed object;

[0057] Step S120: training the long-short-time index relationship model with the goal of making the ranking order of the long-time index value consistent with the ranking order of the corresponding long-time behavior characteristics;

[0058] Step S130: training the recommendation model based on the long-short-time index relationship model, making the input of the recommendation model include user characteristics of the user and object characteristics of the to-be-processed object, and making the output of the recommendation model be a recommendation value of the to-be-processed object to the user.

[0059] In the recommendation model training method of the embodiments of the present disclosure, by establishing and training the long-short-time index relationship model, the long-time index ranking relationship value of the long-time behavior characteristics can be learned based on the short-time behavior characteristics, the recommendation model is trained based on the long-time index ranking relationship value, so as to learn the influence of the long-time behavior characteristics on the object recommendation, thereby the influence of the long-time behavior characteristics on the object recommendation is brought into the recommendation calculation of the recommendation model. Thus, on the one hand, the present disclosure only needs to learn the long-time index ranking relationship value, without directly predicting the long-time behavior characteristics, the long-time index ranking relationship value has lower accuracy requirement on the prediction model than the long-time behavior characteristics, which is conducive to the implementation of the long-short-time index relationship model; on the other hand, the influence of the long-time behavior characteristics on the object recommendation is brought into the recommendation calculation of the recommendation model, so as to improve the recommendation effect.

[0060] In the exemplary embodiments of the present disclosure, the range of the short-time and long-time corresponding time periods of the short-time behavior characteristics and the long-time behavior characteristics can be set by the user. The short-time time period is less than the long-time time period. For example, the short-time time period can be 1 hour, 6 hours, 12 hours, etc., and the long-time time period can be 24 hours, one week, two weeks, etc., which is not limited by the present disclosure. In further embodiments of the present disclosure, the long-time time period can be an integer multiple of the short-time time period. ​

[0061] In exemplary embodiments of the present disclosure, short-term and long-term behavioral features may include the same behavioral feature items. For example, short-term behavioral features may include the number of clicks, browsing duration, touch duration, and non-touch duration of a user within a short period of time; correspondingly, long-term behavioral features may include the number of clicks, browsing duration, touch duration, and non-touch duration of a user within a long period of time. Specific behavioral features can be set as needed, and the present disclosure is not limited thereto.

[0062] In exemplary embodiments of the present disclosure, the long-term indicator ranking relationship value is not a long-term behavior feature; it only needs to indicate the ranking relationship of the long-term behavior features. For example, the long-term indicator ranking relationship value can be the ranking sequence number of the long-term behavior feature, thereby being used to express the ranking relationship of the long-term behavior features. For another example, the long-term indicator ranking relationship value can have a corresponding functional relationship with the ranking sequence number of the long-term behavior feature, thereby being used to express the ranking relationship of the long-term behavior features.

[0063] In an exemplary embodiment of the present disclosure, since it is difficult to directly use short-term behavior features to predict long-term behavior features, the present disclosure adopts a long-short-term indicator relationship model to predict the sorting order of long-term behavior features. To this end, the training goal of the long-short-term indicator relationship model of the present disclosure is to ensure that the sorting order of the long-term indicator values ​​of the same object to be processed by multiple users output by the long-short-term indicator relationship model is consistent with the sorting order of the long-term behavior features of the same object to be processed by the multiple users. For example, the first user's total reading time for the object to be processed within a week is 20 hours, and the second user's total reading time for the object to be processed within a week is 5 hours. The long-short-term indicator relationship model does not need to predict the total time of the first user and the second user within a week, but only needs to predict that the long-term behavior features of the first user are greater than the long-term behavior features of the second user. Assume that the short-term behavior features of the user are S x,t,1 ...S x,t,k , where x represents the user, t represents the time, and k represents the number of short-term behavior features. The long-term indicator ranking relationship value output by the long-term and short-term indicator relationship model can be expressed as SLnet(S x,t,1 ...S x,t,k ), where SLnet represents the long-term and short-term indicator relationship model. Therefore, according to the short-term behavior characteristics of user x at each moment t within the long-term time period of the long-term behavior characteristics, the long-term indicator ranking relationship value SLnet(S at that moment t) can be predicted and output by the long-term and short-term indicator relationship model. x,t,1 ...S x,t,k ), according to the long-term index ranking relationship value SLnet(S x,t,1 ...S x,t,k) can be used as the long-term indicator ranking relationship value F(x) of user x in the long-term period, which is expressed as follows:

[0064]

[0065] Among them, SLnet represents the long-short-term indicator relationship model, and T is the number of short-term time periods of short-term behavior characteristics within the long-term time period of long-term behavior characteristics.

[0066] In exemplary embodiments of the present disclosure, the long-term and short-term indicator relationship model can be any machine learning model. For example, the long-term and short-term indicator relationship model can be a multi-layer neural network (MLP). In exemplary embodiments of the present disclosure, the recommendation model can be any machine learning model. For example, the recommendation model can be a multi-expert gated multi-objective model (MMoE) based on a deep neural network. The present disclosure is not limited to this.

[0067] In exemplary embodiments of the present disclosure, user characteristics may include, but are not limited to, one or more of the following: basic user information, user preference information, and historical user behavior characteristics. Basic user information includes, but is not limited to, user age, gender, city, and occupation. User preference information includes, but is not limited to, object characteristics of user-preferred objects. Object characteristics of user-preferred objects can be obtained based on big data analysis of basic user information and historical user behavior characteristics. Historical user behavior characteristics include, but are not limited to, behavioral characteristics of user operations on one or more objects within a historical time period.

[0068] In the exemplary embodiments of the present disclosure, objects may include but are not limited to commodities, pictures, text, web pages, information, and functions. Other physical or non-physical objects are within the scope of protection of the present disclosure. Object features may include but are not limited to one or more of the following: object name, object type, object attribute information, and user characteristics of users who prefer the object. Depending on the object, the object may have different object attribute information. For example, for a picture, its object attribute information may include picture size, picture resolution, picture type, etc.; for a web page, the object attribute information may include web page keywords, URLs, number of web page links, etc. Users who prefer the object can be obtained through big data analysis based on object features and user historical behavior characteristics.

[0069] In exemplary embodiments of the present disclosure, because the recommendation model is trained based on a long-term and short-term indicator relationship model, the parameters at each layer of the recommendation model implicitly reflect the influence of long-term behavioral characteristics. Consequently, the recommendation model can directly output a recommendation value for the object to be processed based on the user characteristics of the user and the object characteristics of the object to be processed. In some variations, the recommendation model may first directly output short-term behavioral characteristics that predict the influence of the user's long-term behavioral characteristics on the object to be processed based on the user characteristics of the user and the object characteristics of the object to be processed, and then perform statistical calculations based on the predicted short-term behavioral characteristics to obtain the recommendation value for the object to be processed. The statistical calculation may include normalizing each short-term behavioral characteristic to the same numerical range and performing a weighted sum or calculating an average of the normalized short-term behavioral characteristics to obtain the recommendation value for the object to be processed. Furthermore, the weights of each short-term behavioral characteristic can also be obtained through training of the recommendation model. The recommendation value is used to evaluate the value of recommending the object to the user. In some embodiments, the higher the recommendation value, the higher the priority of recommending the object to the user.

[0070] In the exemplary embodiment of the present disclosure, the long-short time indicator relationship model can be trained by using a paired training method. Figure 2 As shown, in this embodiment, the long-short-time indicator relationship model 210 can be trained using a first sample set. The first sample set includes multiple first samples, and each first sample includes a feature pair and a sorting label. The feature pair includes the short-time behavior features of the first user on the object to be processed and the short-time behavior features of the second user on the object to be processed. The sorting label is a sorting classification label of the long-time behavior features of the first user and the second user on the object to be processed. For example, when the long-time behavior features of the first user on the object to be processed are greater than the long-time behavior features of the second user on the object to be processed, the sorting classification label can be set to 1; when the long-time behavior features of the first user on the object to be processed are not greater than the long-time behavior features of the second user on the object to be processed, the sorting classification label can be set to 0. Thus, the long-short-time indicator relationship model can learn the short-time behavior features of the two users input therein to determine the sorting classification of the long-time behavior features of the two users.

[0071] Furthermore, the model parameters of the long-short-time indicator relationship model can be adjusted according to the comparison between the ranking relationship of the long-time indicator ranking relationship value output by the long-short-time indicator relationship model based on the feature pair and the corresponding ranking label.

[0072] In this embodiment, the loss function of the long-short time indicator relationship model is, for example:

[0073]

[0074] Where x and x' are randomly sampled users. g() is a nonlinear function. In one example:

[0075]

[0076] Here, m is a custom constant. Therefore, the parameters of the long-term and short-term indicator relationship model can be adjusted and optimized by maximizing the loss function.

[0077] In the exemplary embodiment of the present disclosure, a sequence training method can be used to train the long-short time indicator relationship model. Figure 3 As shown, a second sample set is used to train the long-short-time indicator relationship model 210, the second sample set includes multiple second samples, each of the second samples includes a feature sequence, the feature sequence includes the short-time behavior features of multiple users on the same object to be processed, and the multiple short-time behavior features in the feature sequence are sorted according to the long-time behavior features of multiple users on the same object to be processed.

[0078] Furthermore, the model parameters of the long-short-time indicator relationship model can be adjusted based on a comparison between the sorting order of the long-time indicator sorting relationship values ​​output by the long-short-time indicator relationship model based on the feature sequence and the sorting order of the corresponding feature sequence. For example, a loss function of the long-short-time indicator relationship model can be calculated based on the sorting order of the long-time indicator sorting relationship values ​​output by the long-short-time indicator relationship model based on the feature sequence and the sorting order of the corresponding feature sequence, and the model parameters of the long-short-time indicator relationship model can be adjusted and optimized based on the calculated loss function.

[0079] For example, the long-term indicator relationship model can be input with 1, the short-term behavior characteristics of user B; 2, the short-term behavior characteristics of user C; 3, the short-term behavior characteristics of user A in the order of the long-term behavior characteristics of each user. The long-term indicator relationship model outputs the long-term indicator ranking relationship values ​​of user A, user B, and user C based on the feature sequence. The sorting order of the long-term indicator ranking relationship values ​​output by the training long-term indicator relationship model is consistent with the sorting of the input feature sequence. For example, the long-term indicator relationship model outputs the long-term indicator ranking relationship values ​​of 5, 1, and 3 for user A, user B, and user C, respectively. The long-term indicator ranking relationship values ​​output by the long-term indicator relationship model are sorted from small to large as user B, user C, and user A, which is consistent with the sorting of the input feature sequence.

[0080] In an exemplary embodiment of the present disclosure, the long-term indicator ranking relationship values ​​learned by the long-term indicator relationship model can be incorporated into the recommendation model by calculating the loss function of the recommendation model based on the output of the long-term indicator relationship model. As a result, when making object recommendations, only the recommendation model can be used, without the involvement of the long-term indicator relationship model. This reduces the storage capacity required by the model, shortens the model's execution time, and improves the efficiency of object recommendations.

[0081] See also Figure 4 , Figure 4 The diagram of the training recommendation model according to the embodiment of the present disclosure is schematically shown. In this embodiment, the long-short-time indicator relationship model is trained before the recommendation model. When the recommendation model is trained, the long-short-time indicator relationship model used is already trained. In this embodiment, the recommendation model 220 predicts the user's predicted short-term behavior characteristics for the object to be processed based on the user characteristics and the object characteristics, and calculates the user's recommendation value for the object to be processed based on the predicted short-term behavior characteristics.

[0082] Thus, the recommendation model 220 can be trained based on the third sample set. The third sample set includes multiple third samples, each of which includes user features of the user, object features of the object to be processed, short-term behavioral features of the user with respect to the object to be processed, and a recommendation value of the object to be processed to the user.

[0083] During training, the user features and object features in the third sample set are input into the recommendation model 220, and the recommendation model outputs the predicted short-term behavior features of the user for the object to be processed and the predicted recommendation value calculated based on the predicted short-term behavior features. The short-term behavior features in the third sample are input into the long-short-term indicator relationship model 210, and the long-short-term indicator relationship model 210 outputs the long-term indicator ranking relationship value based on the short-term behavior features. The loss function calculation module 230 calculates the loss function of the recommendation model 220. The loss function of the recommendation model 220 may include two parts. One part is the loss function of the recommendation model 220 itself, that is, the loss function calculated based on the short-term behavior features of the user for the object to be processed predicted by the recommendation model 220 and the predicted recommendation value calculated based on the predicted short-term behavior features, as well as the short-term behavior features of the user for the object to be processed in the third sample and the recommendation value of the object to be processed to the user. The other part is the loss function calculated based on the long-term indicator ranking relationship value of the user for the object to be processed output by the long-short-term indicator relationship model 210. The loss function of the recommendation model 220 Loss new For example, it can be calculated as follows:

[0084] Loss new=Loss old -SLnet(x current )

[0085] Among them, Loss old is the loss function of the recommendation model 220 itself, that is, the loss function calculated based on the short-term behavior characteristics of the user for the object to be processed predicted by the recommendation model 220 and the predicted recommendation value calculated based on the predicted short-term behavior characteristics, as well as the short-term behavior characteristics of the user in the third sample for the object to be processed and the recommendation value of the object to be processed to the user. SLnet(x current ) is the long-term indicator ranking relationship value of the user for the object to be processed, which is the output of the long-term and short-term indicator relationship model 210.

[0086] Therefore, through the form of a loss function, the relationship between the long-term and short-term behavior features learned by the long-term and short-term indicator relationship model 210 is combined with the recommendation model 220, so that when the recommendation model 220 recommends an object, it can combine the factors of the long-term and short-term behavior features to calculate the recommendation value of the object for the user.

[0087] See also Figure 5 , Figure 5 The figure schematically shows a diagram of training a recommendation model according to an embodiment of the present disclosure.

[0088] In this embodiment, the long-short-term indicator relationship model is trained before the recommendation model. During recommendation model training, the long-short-term indicator relationship model used is already trained. In this embodiment, the recommendation model 220 predicts the user's predicted short-term behavior characteristics for the object to be processed based on the user characteristics and the object characteristics, and calculates the user's recommendation value for the object to be processed based on the predicted short-term behavior characteristics.

[0089] Thus, the recommendation model 220 can be trained based on the fourth sample set. The fourth sample set includes a plurality of fourth samples, each of which includes user features of the user, object features of the object to be processed, and a recommendation value of the object to be processed to the user.

[0090] During training, the user features of the user in the fourth sample set and the object features of the object to be processed are input into the recommendation model 220 to obtain the predicted short-term behavior features of the user for the object to be processed and the predicted recommendation value predicted by the recommendation model 220. The predicted short-term behavior features of the user for the object to be processed predicted by the recommendation model 220 are input into the long-short-time indicator relationship model 210 to obtain the long-short-time indicator ranking relationship value of the user for the object to be processed output by the long-short-time indicator relationship model 210.

[0091] The loss function calculation module 230 calculates the loss function of the recommendation model 220. The loss function of the recommendation model 220 may include two parts. One part is the loss function of the recommendation model 220 itself, that is, the loss function calculated based on the short-term behavioral characteristics of the user for the object to be processed predicted by the recommendation model 220, the predicted recommendation value calculated based on the predicted short-term behavioral characteristics, and the recommendation value of the object to be processed to the user in the fourth sample. The other part is the loss function calculated based on the long-term indicator ranking relationship value of the user for the object to be processed, which is output from the long-term and short-term indicator relationship model 210.

[0092] Furthermore, in other embodiments, the long-term indicator ranking relationship value of the user for the object to be processed output by the long-term indicator relationship model 210 may be directly used as the loss function of the recommendation model 220. The present disclosure can implement more variations.

[0093] In embodiments of the present disclosure, the long-term and short-term indicator relationship model can be periodically trained. The training period can be one day, one week, or the like, and the present disclosure is not limited thereto. By periodically training the long-term and short-term indicator relationship model, it is possible to optimize the long-term and short-term indicator relationship model while also optimizing the recommendation model's recommendation performance.

[0094] The following combination Figure 6 An object recommendation method according to an exemplary embodiment of the present disclosure will be described. Figure 6 The following steps are shown:

[0095] Step S310: Input the user characteristics of the user and the object characteristics of the object to be processed into the recommendation model to obtain a plurality of recommendation values ​​of the objects to be processed for the user. The recommendation model is trained by the method described above.

[0096] Step S320: Sort the plurality of objects to be processed according to the recommendation values ​​to obtain a sorting order of the objects to be processed.

[0097] Step S330: Recommending the objects to be processed to the user according to the sorting order.

[0098] Further integration Figure 7 When recommending objects, there is no need for the long-short-time indicator relationship model to be involved. Only the recommendation model is needed to obtain the recommendation value of the object to be processed to the user, based on the predicted recommendation value, so as to recommend the object to be processed with a higher priority to the user.

[0099] The above is merely an illustrative description of various embodiments provided by the present disclosure, and the present disclosure is not limited thereto. Each embodiment may be used alone or in combination.

[0100] Exemplary apparatus

[0101] After introducing the recommendation model training method of the exemplary embodiment of the present disclosure, next, Figure 8 The recommendation model training device according to an exemplary embodiment of the present disclosure is described. Figure 8 As shown, the recommendation model training device 410 of the exemplary embodiment of the present disclosure may include: a building module 411, a long-short-time indicator relationship model training module 412 and a recommendation model training module 413.

[0102] The establishment module 411 is used to establish a long-short-time indicator relationship model, so that the input of the long-short-time indicator relationship model includes the short-time behavior characteristics of the user on the object to be processed, and the output of the long-short-time indicator relationship model is the long-time indicator ranking relationship value of the user on the object to be processed;

[0103] The long-short-time indicator relationship model training module 412 is used to train the long-short-time indicator relationship model with the goal of making the sorting order of the long-time indicator values ​​consistent with the sorting order of the corresponding long-time behavior features;

[0104] The recommendation model training module 413 is used to train the recommendation model based on the long-short-time indicator relationship model, so that the input of the recommendation model includes the user characteristics of the user and the object characteristics of the object to be processed, and the output of the recommendation model is the recommendation value of the object to be processed to the user.

[0105] According to an exemplary embodiment of the present disclosure, the long-short time indicator relationship model training module includes:

[0106] The first sample set training module is used to train the long-short-time indicator relationship model using the first sample set, the first sample set includes multiple first samples, each of the first samples includes a feature pair and a sorting label, the feature pair includes the short-time behavior characteristics of the first user for the object to be processed and the short-time behavior characteristics of the second user for the object to be processed, and the sorting label is the sorting classification label of the long-time behavior characteristics of the first user and the second user for the object to be processed.

[0107] According to an exemplary embodiment of the present disclosure, the first sample set training module includes: a first parameter adjustment module, which is used to adjust the model parameters of the long-short-time indicator relationship model based on the comparison of the sorting relationship of the long-time indicator sorting relationship value output by the long-short-time indicator relationship model based on the feature pair and the corresponding sorting label.

[0108] According to an exemplary embodiment of the present disclosure, the recommendation model training module includes: a second sample set training module, which is used to train the long-short-time indicator relationship model using a second sample set, the second sample set includes multiple second samples, each of the second samples includes a feature sequence, the feature sequence includes the short-time behavior characteristics of multiple users on the same object to be processed, and the multiple short-time behavior features in the feature sequence are sorted according to the long-time behavior characteristics of multiple users on the same object to be processed.

[0109] According to an exemplary embodiment of the present disclosure, the second sample set training module includes: a second parameter adjustment module, which is used to adjust the model parameters of the long-short-time indicator relationship model based on the comparison of the sorting order of the long-time indicator sorting relationship value output by the long-short-time indicator relationship model based on the feature sequence and the sorting order of the corresponding feature sequence.

[0110] According to an exemplary embodiment of the present disclosure, the recommendation model training module includes: a loss function calculation module, which is used to calculate the loss function of the recommendation model based on the output of the long-short-time indicator relationship model.

[0111] According to an exemplary embodiment of the present disclosure, the recommendation model predicts and obtains the predicted short-time behavior characteristics of the user for the object to be processed based on the user characteristics and the object characteristics, and calculates and obtains the recommendation value of the user for the object to be processed based on the predicted short-time behavior characteristics. The recommendation model is trained with a third sample set, and the third sample set includes multiple third samples, each of the third samples includes the user characteristics of the user, the object characteristics of the object to be processed, the short-time behavior characteristics of the user for the object to be processed, and the recommendation value of the object to be processed for the user. The loss function calculation module includes: a first input module, which is used to input the short-time behavior characteristics of the user for the object to be processed in the third sample set into the long-short-time indicator relationship model; a first loss function calculation module, which is used to calculate the loss function of the recommendation model based on the long-time indicator ranking relationship value of the user for the object to be processed output by the long-short-time indicator relationship model.

[0112] According to an exemplary embodiment of the present disclosure, the recommendation model predicts and obtains the predicted short-time behavior characteristics of the user for the object to be processed based on the user characteristics and the object characteristics, and calculates and obtains the recommendation value of the user for the object to be processed based on the predicted short-time behavior characteristics. The recommendation model is trained with a fourth sample set, and the fourth sample set includes multiple fourth samples, each of the fourth samples includes the user characteristics of the user and the object characteristics of the object to be processed and the recommendation value of the object to be processed for the user. The loss function calculation module includes: a second input module, which is used to input the user characteristics of the user and the object characteristics of the object to be processed in the fourth sample set into the recommendation model to obtain the predicted short-time behavior characteristics of the user for the object to be processed predicted by the recommendation model; a third input module, which is used to input the predicted short-time behavior characteristics into the long-short-time indicator relationship model; and a second loss function calculation module, which is used to calculate the loss function of the recommendation model based on the long-time indicator ranking relationship value of the user for the object to be processed output by the long-short-time indicator relationship model.

[0113] According to an exemplary embodiment of the present disclosure, the long-short-time indicator relationship model is trained periodically.

[0114] Next, refer to Figure 9 The object recommendation device according to an exemplary embodiment of the present disclosure is described. Figure 9 As shown, the object recommendation device 420 of the exemplary embodiment of the present disclosure may include: a recommendation value obtaining module 421, a sorting module 422 and a recommendation module 423.

[0115] The recommendation value obtaining module 421 is used to input the user characteristics of the user and the object characteristics of the object to be processed into the recommendation model to obtain the recommendation values ​​of multiple objects to be processed for the user. The recommendation model is trained by the method described above.

[0116] The sorting module 422 is used to sort the plurality of objects to be processed according to the recommendation values ​​to obtain a sorting order of the objects to be processed;

[0117] The recommendation module 423 is configured to recommend the objects to be processed to the user according to the sorting order.

[0118] Since the functional modules of the recommendation model training device and the object recommendation device in the disclosed embodiments are the same as those in the disclosed embodiments of the recommendation model training method and the object recommendation method, they will not be described in detail here.

[0119] Exemplary storage media

[0120] After introducing the recommendation model training and recommendation method and device of the exemplary embodiment of the present disclosure, next, refer to Figure 10 A storage medium according to an exemplary embodiment of the present disclosure is described.

[0121] refer to Figure 10 As shown, a program product 1000 for implementing the above method according to an embodiment of the present disclosure is described. The program product 1000 may be a portable compact disc read-only memory (CD-ROM) and includes program code, and may be run on a device such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0122] The program product may be implemented in any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0123] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0124] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0125] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0126] Exemplary electronic device

[0127] After introducing the storage medium of the exemplary embodiment of the present disclosure, next, reference is made to Figure 11 An electronic device according to an exemplary embodiment of the present disclosure will be described.

[0128] Figure 11 The electronic device 800 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0129] like Figure 11 As shown, electronic device 800 is implemented as a general-purpose computing device. Components of electronic device 800 may include, but are not limited to, the aforementioned at least one processing unit 810, the aforementioned at least one storage unit 820, a bus 830 connecting various system components (including storage unit 820 and processing unit 810), and a display unit 840.

[0130] The storage unit stores program codes, which can be executed by the processing unit 810, so that the processing unit 810 performs the steps according to various exemplary embodiments of the present invention described in the "Exemplary Method" section above. For example, the processing unit 810 can perform the following steps: Figure 1 Follow the steps shown in .

[0131] The storage unit 820 may include a volatile storage unit, such as a random access memory unit (RAM) 8201 and / or a cache memory unit 8202 , and may further include a read-only memory unit (ROM) 8203 .

[0132] The storage unit 820 may also include a program / utility 8204 having a set (at least one) of program modules 8205, such program modules 8205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0133] The bus 830 may include a data bus, an address bus, and a control bus.

[0134] The electronic device 800 can also communicate with one or more external devices 900 (e.g., a keyboard, a pointing device, a Bluetooth device, etc.), and such communication can be performed via an input / output (I / O) interface 850. The electronic device 800 also includes a display unit 840, which is connected to the input / output (I / O) interface 850 for display. In addition, the electronic device 800 can also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 860. As shown, the network adapter 860 communicates with other modules of the electronic device 800 via a bus 830. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0135] It should be noted that although several modules or submodules of the recommendation model training device and the object recommendation device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be concretized in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided to be concretized by multiple units / modules.

[0136] Furthermore, although the operations of the disclosed method are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in this particular order, or that all illustrated operations must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0137] Although the spirit and principles of the present invention have been described with reference to several specific embodiments, it should be understood that the present invention is not limited to the specific embodiments disclosed, and the division into various aspects does not mean that the features of these aspects cannot be combined to benefit. Such division is only for the convenience of expression. The present invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A recommendation model training method, characterized in that: include: Establishing a long-short-time indicator relationship model, wherein the input of the long-short-time indicator relationship model includes the short-time behavior characteristics of multiple users on the object to be processed, and the output of the long-short-time indicator relationship model is the long-time indicator ranking relationship value of the multiple users on the object to be processed; Training the long- and short-term indicator relationship model with the goal of ensuring that the sorting order of the long-term indicator sorting relationship values ​​is consistent with the sorting order of the corresponding long-term behavior features; Based on the long-short-time indicator relationship model, training the recommendation model so that the input of the recommendation model includes the user characteristics of the user and the object characteristics of the object to be processed, and the output of the recommendation model is the recommendation value of the object to be processed to the user; The step of training the recommendation model based on the long-short-time indicator relationship model includes: The loss function of the recommendation model is calculated based on the output of the long-short-time indicator relationship model.

2. The recommendation model training method according to claim 1, characterized in that The training of the long-short time indicator relationship model includes: A first sample set is used to train the long-short-time indicator relationship model, the first sample set includes multiple first samples, each of the first samples includes a feature pair and a sorting label, the feature pair includes the short-time behavior characteristics of the first user on the object to be processed and the short-time behavior characteristics of the second user on the object to be processed, and the sorting label is the sorting classification label of the long-time behavior characteristics of the first user and the second user on the object to be processed.

3. The recommendation model training method according to claim 2, characterized in that The adopting of the first sample set to train the long-short time indicator relationship model comprises: According to the comparison between the ranking relationship of the long-term indicator ranking relationship value output by the long-term indicator relationship model based on the feature pair and the corresponding ranking label, the model parameters of the long-term indicator relationship model are adjusted.

4. The recommendation model training method according to claim 1, characterized in that The training of the long-short time indicator relationship model includes: A second sample set is used to train the long-short-time indicator relationship model, the second sample set includes multiple second samples, each second sample includes a feature sequence, the feature sequence includes the short-time behavior features of multiple users on the same object to be processed, and the multiple short-time behavior features in the feature sequence are sorted according to the long-time behavior features of multiple users on the same object to be processed.

5. The recommendation model training method according to claim 4, characterized in that The adopting of the second sample set to train the long-short time indicator relationship model includes: According to the comparison between the sorting order of the long-term indicator sorting relationship value output by the long-term indicator relationship model based on the feature sequence and the sorting order of the corresponding feature sequence, the model parameters of the long-term indicator relationship model are adjusted.

6. The recommendation model training method according to claim 1, characterized in that The recommendation model predicts the user's predicted short-term behavior characteristics for the object to be processed based on the user characteristics and the object characteristics, and calculates the user's recommendation value for the object to be processed based on the predicted short-term behavior characteristics. The recommendation model is trained with a third sample set, the third sample set including a plurality of third samples, each of the third samples including user features of the user, object features of the object to be processed, short-term behavior features of the user with respect to the object to be processed, and a recommendation value of the object to be processed to the user, and the loss function of the recommendation model is calculated based on the output of the long-short-term indicator relationship model. Inputting the short-term behavior characteristics of the users in the third sample set to the object to be processed into the long-short-term indicator relationship model; The loss function of the recommendation model is calculated based on the long-term indicator ranking relationship value of the user to the object to be processed output by the long-term and short-term indicator relationship model.

7. The recommendation model training method according to claim 1, characterized in that The recommendation model predicts the user's predicted short-term behavior characteristics for the object to be processed based on the user characteristics and the object characteristics, and calculates the user's recommendation value for the object to be processed based on the predicted short-term behavior characteristics. The recommendation model is trained with a fourth sample set, the fourth sample set including a plurality of fourth samples, each of the fourth samples including user features of the user, object features of the object to be processed, and a recommendation value of the object to be processed to the user, and the loss function of the recommendation model calculated based on the output of the long-short-time indicator relationship model includes: Inputting the user features of the user in the fourth sample set and the object features of the object to be processed into the recommendation model to obtain the predicted short-term behavior features of the user on the object to be processed predicted by the recommendation model; Inputting the predicted short-term behavior characteristics into the long-short-term indicator relationship model; The loss function of the recommendation model is calculated based on the long-term indicator ranking relationship value of the user to the object to be processed output by the long-term and short-term indicator relationship model.

8. The recommendation model training method according to any one of claims 1 to 7, characterized in that: The long-short-time indicator relationship model is trained periodically.

9. An object recommendation method, characterized in that: include: Inputting user features of the user and object features of the object to be processed into a recommendation model to obtain a plurality of recommendation values ​​of the object to be processed for the user, wherein the recommendation model is trained by the method according to any one of claims 1 to 8; Sorting the plurality of objects to be processed according to the recommendation values ​​to obtain a sorting order of the objects to be processed; The objects to be processed are recommended to the user according to the sorting order.

10. A recommendation model training device, characterized in that: include: An establishment module is used to establish a long-short-time indicator relationship model, so that the input of the long-short-time indicator relationship model includes the short-time behavior characteristics of multiple users on the object to be processed, and the output of the long-short-time indicator relationship model is the long-time indicator ranking relationship value of the multiple users on the object to be processed; A long-short-time indicator relationship model training module is used to train the long-short-time indicator relationship model with the goal of ensuring that the sorting order of the long-time indicator sorting relationship values ​​is consistent with the sorting order of the corresponding long-time behavior features; A recommendation model training module, configured to train the recommendation model based on the long-short-time indicator relationship model, such that the input of the recommendation model includes the user characteristics of the user and the object characteristics of the object to be processed, and the output of the recommendation model is the recommendation value of the object to be processed to the user; The recommendation model training module includes: A loss function calculation module is used to calculate the loss function of the recommendation model based on the output of the long-short-time indicator relationship model.

11. The recommendation model training device according to claim 10, characterized in that: The long-short time indicator relationship model training module includes: The first sample set training module is used to train the long-short-time indicator relationship model using the first sample set. The first sample set includes multiple first samples, each of the first samples includes a feature pair and a sorting label, the feature pair includes the short-time behavior characteristics of the first user for the object to be processed and the short-time behavior characteristics of the second user for the object to be processed, and the sorting label is the sorting classification label of the long-time behavior characteristics of the first user and the second user for the object to be processed.

12. The recommendation model training device according to claim 11, characterized in that: The first sample set training module includes: The first parameter adjustment module is used to adjust the model parameters of the long-short-time indicator relationship model according to the comparison between the sorting relationship of the long-time indicator sorting relationship value output by the long-short-time indicator relationship model based on the feature pair and the corresponding sorting label.

13. The recommendation model training device according to claim 10, characterized in that: The recommendation model training module includes: The second sample set training module is used to train the long-short-time indicator relationship model using a second sample set, the second sample set includes multiple second samples, each of the second samples includes a feature sequence, the feature sequence includes the short-time behavior characteristics of multiple users on the same object to be processed, and the multiple short-time behavior features in the feature sequence are sorted according to the long-time behavior characteristics of multiple users on the same object to be processed.

14. The recommendation model training device according to claim 13, characterized in that: The second sample set training module includes: The second parameter adjustment module is used to adjust the model parameters of the long-short-time indicator relationship model according to the comparison between the sorting order of the long-time indicator sorting relationship value output by the long-short-time indicator relationship model based on the feature sequence and the sorting order of the corresponding feature sequence.

15. The recommendation model training device according to claim 10, characterized in that: The recommendation model predicts and obtains the predicted short-term behavior characteristics of the user for the object to be processed based on the user characteristics and the object characteristics, and calculates and obtains the recommendation value of the user for the object to be processed based on the predicted short-term behavior characteristics. The recommendation model is trained with a third sample set, and the third sample set includes a plurality of third samples, each of the third samples includes the user characteristics of the user, the object characteristics of the object to be processed, the short-term behavior characteristics of the user for the object to be processed, and the recommendation value of the object to be processed for the user. The loss function calculation module includes: A first input module, configured to input the short-term behavior characteristics of the user in the third sample set towards the object to be processed into the long-short-term indicator relationship model; The first loss function calculation module is used to calculate the loss function of the recommendation model based on the long-term indicator ranking relationship value of the user to the object to be processed output by the long-term and short-term indicator relationship model.

16. The recommendation model training device according to claim 10, characterized in that: The recommendation model predicts the user's predicted short-term behavior characteristics for the object to be processed based on the user characteristics and the object characteristics, and calculates the user's recommendation value for the object to be processed based on the predicted short-term behavior characteristics. The recommendation model is trained with a fourth sample set, the fourth sample set including a plurality of fourth samples, each of the fourth samples including a user feature of the user, an object feature of the object to be processed, and a recommendation value of the object to be processed to the user, and the loss function calculation module including: A second input module is configured to input the user features of the user in the fourth sample set and the object features of the object to be processed into the recommendation model to obtain the predicted short-term behavior features of the user with respect to the object to be processed predicted by the recommendation model; A third input module is used to input the predicted short-term behavior characteristics into the long-term and short-term indicator relationship model; The second loss function calculation module is used to calculate the loss function of the recommendation model based on the long-term indicator ranking relationship value of the user to the object to be processed output by the long-term and short-term indicator relationship model.

17. The recommendation model training device according to any one of claims 10 to 16, characterized in that: The long-short-time indicator relationship model is trained periodically.

18. An object recommendation device, characterized in that: include: a recommendation value obtaining module, configured to input user characteristics of a user and object characteristics of an object to be processed into a recommendation model to obtain recommendation values ​​of a plurality of objects to be processed for the user, wherein the recommendation model is trained by the method according to any one of claims 1 to 8; a sorting module, configured to sort the plurality of objects to be processed according to the recommendation values ​​to obtain a sorting order of the objects to be processed; A recommendation module is used to recommend the objects to be processed to the user according to the sorting order.

19. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it realizes: The recommendation model training method according to any one of claims 1 to 8; and / or The object recommendation method according to claim 9.

20. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; The processor is configured to execute the following instructions: The recommendation model training method according to any one of claims 1 to 8; and / or The object recommendation method according to claim 9.

Images