Recommendation method and device, electronic equipment, storage medium and product

Abstract

The invention provides a recommendation method and device, electronic equipment, a storage medium and a product. The method comprises the following steps: receiving a recommendation request, wherein the recommendation request comprises a target user identifier and object feature information of a recommendation object; determining user feature information corresponding to the target user identifier according to the recommendation request; determining a deviation data set corresponding to the object feature information, and determining an unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator; determining a target recommendation model according to the deviation data set and the unbiased data set; and inputting the object feature information into the target recommendation model to obtain a target recommendation object. According to the scheme, the target recommendation model obtained by adjusting the deviation data set collected in real time and the non-deviation data conforming to the user characteristics can adapt to real-time changes of the user, and therefore recommendation accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a recommendation method, device, electronic device, storage medium and product. Background Art

[0002] In the application scenario of recommendation systems in the field of artificial intelligence technology, users input feature information of recommended objects into the recommendation system, and the recommendation system automatically provides personalized recommended objects. Recommended objects include but are not limited to videos, commodities, or social content.

[0003] In actual applications, due to uneven data distribution, historical behavior deviations, or model limitations, there are often various deviations in the recommended objects, which will affect the user experience. In related technologies, the causes of the deviations are analyzed manually, and the recommendation system parameters are adjusted accordingly.

[0004] However, manually analyzing the causes of deviations will introduce errors, leading to poor debiasing effects and, in turn, low recommendation accuracy. Summary of the invention

[0005] The present application provides a recommendation method, device, electronic device, storage medium and product to improve the accuracy of recommendations.

[0006] In a first aspect, the present application provides a recommendation method, comprising: receiving a recommendation request, the recommendation request comprising a target user identifier and object feature information of a recommended object; determining user feature information corresponding to the target user identifier based on the recommendation request; determining a biased data set corresponding to the object feature information, and determining an unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator; determining a target recommendation model based on the biased data set and the unbiased data set; and inputting the object feature information into the target recommendation model to obtain a target recommended object.

[0007] Optionally, the method as described above determines the user characteristic information corresponding to the target user identifier, including: determining multiple historical interaction records corresponding to the target user identifier, and multiple timestamps corresponding to the multiple historical interaction records; determining multiple target interaction records in the multiple historical interaction records based on a preset duration and the multiple timestamps; and determining the user characteristic information based on the multiple target interaction records.

[0008] Optionally, the method as described above, determines the deviation data set corresponding to the object feature information, including: determining multiple historical feature information corresponding to the multiple target interaction records; determining the deviation data set from the multiple target interaction records based on the object feature information and the multiple historical feature information, the similarity between the historical feature information corresponding to the deviation data set and the object feature information is greater than or equal to a first threshold.

[0009] Optionally, the method as described above determines the unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator, including: establishing the user simulator based on the user feature information, the user simulator being used to simulate the user's interactive behavior based on the user feature information; determining a plurality of candidate objects based on the object feature information, the feature information of the plurality of candidate objects having a similarity with the object feature information that is greater than or equal to a second threshold; generating an object list corresponding to the plurality of candidate objects through the unbiased recommendation algorithm, in the object list, each candidate object has the same exposure rate; and obtaining the unbiased data set from the object list through the user simulator.

[0010] Optionally, the method as described above, obtaining the unbiased data set from the object list through the user simulator, includes: determining the historical record of generating the data set based on the target user identifier; determining a target data range based on the historical record; and obtaining the unbiased data set from the object list through the user simulator based on the target data range, the data volume of the unbiased data set being within the target data range.

[0011] Optionally, the method as described above determines the target recommendation model based on the biased data set and the unbiased data set, including: determining the historical recommendation model corresponding to the target user identifier; training the historical recommendation model multiple times with the biased data set to obtain multiple intermediate models; testing the multiple intermediate models separately with the unbiased data set to obtain the intermediate loss function value corresponding to each intermediate model, until any intermediate loss function value is less than or equal to the loss function threshold, then the corresponding intermediate model is determined as the target recommendation model.

[0012] In a second aspect, the present application provides a recommendation device, comprising: a receiving module, used to receive a recommendation request, the recommendation request including a target user identifier and object feature information of a recommended object; a determination module, used to determine, based on the recommendation request, the user feature information corresponding to the target user identifier; a data module, used to determine a deviation data set corresponding to the object feature information, and determine an unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator; a training module, used to determine a target recommendation model based on the deviation data set and the unbiased data set; and a recommendation module, used to input the object feature information into the target recommendation model to obtain a target recommended object.

[0013] Optionally, in the device as described above, the determination module is specifically used to determine multiple historical interaction records corresponding to the target user identifier, and multiple timestamps corresponding to the multiple historical interaction records; the determination module is also specifically used to determine multiple target interaction records in the multiple historical interaction records based on a preset duration and the multiple timestamps; the determination module is also specifically used to determine the user feature information based on the multiple target interaction records.

[0014] Optionally, as described above, the device further includes: a screening module, used to determine multiple historical feature information corresponding to the multiple target interaction records; the screening module is specifically used to determine the deviation data set from the multiple target interaction records based on the object feature information and the multiple historical feature information, the similarity between the historical feature information corresponding to the deviation data set and the object feature information is greater than or equal to a first threshold.

[0015] Optionally, in the device as described above, the first generating module is used to establish the user simulator based on the user characteristic information, and the user simulator is used to simulate the user's interactive behavior based on the user characteristic information; the first generating module is also used to determine multiple candidate objects based on the object characteristic information, and the similarity between the characteristic information of the multiple candidate objects and the object characteristic information is greater than or equal to a second threshold; the first generating module is also used to generate an object list corresponding to the multiple candidate objects through the unbiased recommendation algorithm, and in the object list, each candidate object has the same exposure rate; the first generating module is also used to obtain the unbiased data set from the object list through the user simulator.

[0016] Optionally, in the device as described above, the second generating module is used to determine the historical record of generating the data set according to the target user identifier; the second generating module is also used to determine the target data range according to the historical record; the second generating module is also used to obtain the unbiased data set from the object list through the user simulator according to the target data range, and the data volume of the unbiased data set is within the target data range.

[0017] Optionally, in the device as described above, the execution module is used to determine the historical recommendation model corresponding to the target user identifier; the execution module is also used to train the historical recommendation model multiple times through the deviation data set to obtain multiple intermediate models; the execution module is also used to test the multiple intermediate models separately through the unbiased data set to obtain the intermediate loss function value corresponding to each intermediate model, until any intermediate loss function value is less than or equal to the loss function threshold, then the corresponding intermediate model is determined as the target recommendation model.

[0018] In a third aspect, the present application provides an electronic device comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement any method described in the first aspect.

[0019] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by a processor as described in any one of the methods in the first aspect.

[0020] In a fifth aspect, the present application provides a computer program product, comprising a computer program, wherein the computer program is executed by a processor as in any one of the methods in the first aspect.

[0021] The recommendation method, device, electronic device, storage medium and product provided by the present application receive a recommendation request, wherein the recommendation request includes a target user identifier and object feature information of the recommended object; determine the user feature information corresponding to the target user identifier according to the recommendation request; determine the deviation data set corresponding to the object feature information, and determine the unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator; determine a target recommendation model based on the deviation data set and the unbiased data set; input the object feature information into the target recommendation model to obtain a target recommended object. The above scheme, through the target recommendation model obtained by adjusting the deviation data set collected in real time and the unbiased data that conforms to the user characteristics, can adapt to the real-time changes of the user, thereby improving the accuracy of the recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0023] Figure 1 A schematic diagram of an application scenario of a recommended method provided in an embodiment of the present application;

[0024] Figure 2 A flowchart of a recommended method provided in an embodiment of the present application;

[0025] Figure 3 A flowchart of a recommended method provided in an embodiment of the present application;

[0026] Figure 4 A schematic diagram of determining a deviation data set provided in an embodiment of the present application;

[0027] Figure 5 A schematic diagram of determining an unbiased data set provided in an embodiment of the present application;

[0028] Figure 6 A schematic diagram of a target recommendation model for determining an embodiment of the present application;

[0029] Figure 7 A schematic diagram of a recommended device provided in an embodiment of the present application;

[0030] Figure 8 A schematic diagram of a recommended device provided in an embodiment of the present application;

[0031] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0032] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0033] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0034] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0035] In addition, this application involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and the use of artificial intelligence technology for automated decision-making, and a technical solution for making decisions that have a significant impact on personal rights and interests based on the results of automated decision-making. The application provides users with corresponding operation entrances for them to choose to agree or reject the results of automated decision-making; if the user chooses to reject, the expert decision-making process will be entered.

[0036] It should be noted that the methods, devices, electronic devices, storage media and products recommended in this application can be used in the field of artificial intelligence technology, and can also be used in any field other than artificial intelligence. The application fields of the methods, devices, electronic devices, storage media and products recommended in this application are not limited.

[0037] Figure 1 A schematic diagram of an application scenario of a recommendation method provided in an embodiment of the present application is given by way of example in combination with the illustrated scenario: a user inputs feature information into a recommendation system, and the recommendation system outputs relevant recommended objects based on the feature information.

[0038] The recommendation system may be a recommendation model, which recommends similar objects based on the user's past preferences and the features of the items.

[0039] In the related art, there is a certain deviation between the recommended objects output by the recommendation model and the actual needs of the user. The deviation is analyzed manually, and the recommendation model is adjusted according to the analysis results to improve the recommendation accuracy of the recommendation model.

[0040] However, user preferences change dynamically. The same recommended object may have met the user's needs in the past but not in the present. Manual analysis of deviations cannot obtain the user's changing preferences in real time, resulting in low recommendation accuracy of the recommendation model.

[0041] The recommended method provided in this application is intended to solve the above technical problems of the prior art.

[0042] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0043] Figure 2 A flowchart of a recommended method provided in an embodiment of the present application, the method comprises the following steps:

[0044] S201: Receive a recommendation request, where the recommendation request includes a target user identifier and object feature information of a recommended object.

[0045] Exemplarily, the target user identifier is a unique identifier of a user who initiates a recommendation request. The object feature information is information input by a user describing the recommended object.

[0046] Optionally, the target user identifier may be a unique identifier corresponding to an identity account used by the user. For example, a user logs in to an identity account to initiate a recommendation request, and the unique identifier corresponding to the logged-in identity account is the target user identifier.

[0047] Combined with the scenario example, the target user identifier can be used to accurately determine the unique corresponding user feature information, so that personalized recommendations can be accurately performed based on the user feature information.

[0048] S202: Determine user feature information corresponding to the target user identifier according to the recommendation request.

[0049] Among them, user characteristic information includes but is not limited to at least one of the following: user behavior information, user preferences, or user portrait data, etc.

[0050] Optionally, a correspondence between user characteristic information and user identification is pre-established, and the corresponding user characteristic information can be accurately determined through the target user identification and the correspondence.

[0051] Combined with the scenario examples, user feature information can accurately reflect user preferences. The recommendation model determined based on user feature information can make personalized recommendations based on user preferences, thereby improving the accuracy of recommendations.

[0052] S203: Determine a biased data set corresponding to the object feature information, and determine an unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator.

[0053] Among them, the biased dataset is a naturally collected dataset that includes bias, and the unbiased dataset is a generated dataset that does not include bias.

[0054] Optionally, the unbiased recommendation algorithm is a random strategy, which specifically refers to a strategy that has an equal chance of being exposed in the first position regardless of whether an object is good or bad. The user simulator is built using real user interaction data and the user's own feature information to interact with the specified recommendation strategy module and simulate the user's real feedback. This makes up for the lack of online unbiased data.

[0055] In the related art, the recommendation system tends to recommend popular objects because these objects have more user interaction data and are more likely to be clicked by users. This leads to overexposure of popular objects while unpopular objects are ignored.

[0056] However, the unbiased recommendation algorithm of this application ensures that each object has an equal exposure opportunity, regardless of its current popularity. This helps to discover and recommend some potentially high-quality but unpopular objects, improving the fairness of the recommendation.

[0057] In the related art, a data set is obtained through a user's historical records. If a user has fewer historical records, the data set will be smaller, which will lead to poor results in training a recommendation model.

[0058] However, this application generates a data set through a user simulator, which can enrich the number of samples and thus improve the accuracy of the recommendation model.

[0059] S204: Determine a target recommendation model according to the biased data set and the unbiased data set.

[0060] Exemplarily, a model with bias is obtained by training with a biased dataset, and model parameters are optimized by back propagation with an unbiased dataset to reduce the bias, thereby obtaining a target recommendation model with small bias.

[0061] Combined with the scenario example, the biased dataset is collected naturally and is authentic, but the bias in the biased dataset will cause some objects to be over-recommended. Although the unbiased dataset eliminates various biases, it usually needs to be generated through simulation and lacks the richness and complexity of real user behavior. By combining the biased dataset and the unbiased dataset to determine the target recommendation model, the bias in the recommendation results can be reduced and the recommendation results can be made more in line with the real needs of users.

[0062] S205: Input the object feature information into the target recommendation model to obtain the target recommended object.

[0063] Combined with the scenario example, the target recommendation model determines the target recommendation object that matches the object based on the object feature information.

[0064] Optionally, there are multiple target recommendation objects.

[0065] Optionally, the multiple target recommended objects are sorted according to the degree of proximity to the object feature information. The sorting can effectively assist the user to quickly select a recommended object from the multiple target recommended objects.

[0066] Optionally, the target recommendation model outputs matching information of the target recommendation object. For example, if the object feature information includes feature A, feature B, and feature C, and the target recommendation object A is close to feature A and feature B, the matching information corresponding to the target recommendation object A may be "target recommendation object A matches feature A and feature B". The matching information can effectively assist the user in accurately selecting a recommended object from multiple target recommendation objects.

[0067] The recommendation method provided by the embodiment of the present application receives a recommendation request, the recommendation request includes a target user identifier and object feature information of the recommended object; determines the user feature information corresponding to the target user identifier according to the recommendation request; determines the deviation data set corresponding to the object feature information, and determines the unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator; determines the target recommendation model according to the deviation data set and the unbiased data set; inputs the object feature information into the target recommendation model to obtain the target recommended object. The above scheme, through the target recommendation model obtained by adjusting the deviation data set collected in real time and the unbiased data that conforms to the user characteristics, can adapt to the real-time changes of the user, thereby improving the accuracy of the recommendation.

[0068] Based on any of the above embodiments, Figure 3 , the detailed process of recommendation is explained.

[0069] Figure 3 The following is a flow chart of a recommended method provided in the embodiment of the present application. Figure 3 As shown, the method includes:

[0070] S301: Receive a recommendation request, where the recommendation request includes a target user identifier and object feature information of a recommended object.

[0071] It should be noted that the execution process of S301 refers to S201 and will not be repeated here.

[0072] S302: Determine a plurality of historical interaction records corresponding to the target user identifier and a plurality of timestamps corresponding to the plurality of historical interaction records.

[0073] The historical interaction record is a record of the user's historical use of the recommendation model corresponding to the target user identifier.

[0074] Exemplarily, the historical interaction records may include the user's click records, rating records, search records, browsing records, or favorite records, etc. The historical interaction records may reflect the user's personalized characteristics.

[0075] Exemplarily, each historical interaction record corresponds to a timestamp, and the timestamp is the time when the historical interaction record is generated.

[0076] S303: Determine multiple target interaction records from multiple historical interaction records according to a preset duration and multiple timestamps.

[0077] Exemplarily, the multiple historical interaction records include all interaction records of the user, and partial interaction records are obtained by screening from the multiple historical interaction records through preset durations and multiple timestamps.

[0078] Optionally, an interaction record that is closer to the current moment among the multiple historical interaction records is determined as the target interaction record.

[0079] Combined with the scenario example, the duration range is determined according to the current moment and the preset duration, and the interaction records with timestamps within the duration range are determined as target interaction records. It can be understood that the target interaction records closer to the current moment are more in line with the user's current preferences, and the target recommendation model determined on this basis can more accurately output personalized target recommendation objects.

[0080] S304: Determine user characteristic information according to multiple target interaction records.

[0081] Optionally, user feature information is extracted from multiple target interaction records through the model.

[0082] Exemplarily, extracting user feature information from multiple target interaction records may include: feature extraction, including extracting basic features: extracting basic features from interaction records, such as interaction types, etc. Extracting advanced features: extracting more advanced features according to interaction types, such as the number of clicks, browsing frequency, and average ratings, etc. Behavior frequency analysis, such as calculating the click frequency of users clicking on various types of content and calculating the purchase frequency of users purchasing various types of goods. Calculating the frequency of users searching for various keywords. Calculating the frequency of users browsing various pages. Behavior time analysis, such as analyzing the active time period of users in a day and identifying the periodic behavior of users. User preference inference, such as topic preference, category preference, interaction preference, and rating preference. User portrait construction, such as obtaining age, gender, geographic location, interest tags, behavior patterns, social circles, interaction frequency, etc. Feature engineering, including correlation analysis, selecting features that are highly relevant to the recommendation task, and removing irrelevant or redundant features. Feature dimensionality reduction, reducing the number of features and improving model efficiency. Feature conversion, such as normalization and encoding, etc. Model training, such as selecting a model, selecting a suitable recommendation model, such as collaborative filtering, deep learning model, etc., training the model: training the recommendation model using the extracted features.

[0083] S305: Determine multiple pieces of historical feature information corresponding to multiple target interaction records.

[0084] The multiple historical feature information are feature information corresponding to multiple historical objects.

[0085] Combined with the scenario example, the object corresponding to each target interaction record is determined from multiple target interaction records, and the feature information corresponding to each object is determined. For example, if any target interaction record is a click object, the object clicked by the user is determined, and the feature information of the clicked object is determined.

[0086] S306: Determine a deviation data set from multiple target interaction records according to the object feature information and multiple historical feature information.

[0087] The similarity between the historical feature information and the object feature information corresponding to the deviation data set is greater than or equal to a first threshold.

[0088] Next, combine Figure 4 The determination bias data set is described.

[0089] Figure 4 Schematic diagram of determining a deviation data set provided in an embodiment of the present application. Figure 4 As shown, multiple target interaction records are obtained from multiple historical interaction records according to preset duration and timestamp, multiple historical feature information corresponding to the multiple target interaction records are determined, and target historical feature information is determined from the multiple historical feature information, and the similarity between the target historical feature information and the object feature information is greater than or equal to a first threshold. The interaction record corresponding to the target historical feature information is determined as a deviation data set.

[0090] In combination with the scenario example, the object feature information includes feature A, feature B, and feature C, and each data in the deviation data set includes at least one of feature A, feature B, and feature C.

[0091] Based on the above implementation, by controlling the similarity between the historical feature information corresponding to the deviation data set and the object feature information, the deviation data set can be made to meet the preference corresponding to the recommendation request, thereby improving the accuracy of the recommendation.

[0092] A feasible implementation method can determine an unbiased data set through the following method: establish a user simulator based on user feature information, and the user simulator is used to simulate the user's interactive behavior based on the user feature information; determine multiple candidates based on object feature information, and the similarity between the feature information of the multiple candidates and the object feature information is greater than or equal to a second threshold; generate an object list corresponding to the multiple candidates through an unbiased recommendation algorithm, and in the object list, each candidate has the same exposure rate; and obtain an unbiased data set from the object list through the user simulator.

[0093] Exemplarily, the user's interactive behavior habits are determined based on the user's characteristic information, and a user simulator is established based on the interactive behavior habits. The user simulator can simulate the user's interactive behavior habits and select an object that the user may select from multiple objects.

[0094] Next, combine Figure 5 Determining an unbiased dataset is explained.

[0095] Figure 5 A schematic diagram of determining an unbiased data set provided in an embodiment of the present application. Figure 5 As shown, a corresponding user simulator that can simulate user interaction behavior is determined according to user feature information. Multiple candidate objects corresponding to object feature information are determined. An object list is generated according to the multiple candidate objects through an unbiased recommendation algorithm to ensure that the objects in the object list have the same exposure rate. The user simulator is used to simulate the user's interaction behavior to obtain unbiased data from the object list.

[0096] Combined with scenario examples, when the data set corresponding to a user is small, generating an unbiased data set by simulating the user's interactive behavior through a user simulator can effectively expand the number of data sets.

[0097] In this feasible implementation, an unbiased recommendation algorithm is used to ensure that each candidate object has the same exposure rate in the object list, eliminating popularity bias, position bias, etc. The user's interactive behavior is simulated according to the user's characteristic information to generate an unbiased data set, ensuring that each object has an equal exposure opportunity, making the recommendation result fairer and avoiding some objects being overexposed due to popularity and other objects being ignored due to unpopularity.

[0098] A feasible implementation method is to obtain an unbiased data set from an object list through a user simulator in the following way: determine the historical record of generating a data set based on a target user identifier; determine a target data range based on the historical record; and obtain an unbiased data set from an object list through a user simulator based on the target data range, wherein the data volume of the unbiased data set is within the target data range.

[0099] The historical record is a record of each time a user corresponding to the target user ID generates a data set. Each historical record corresponds to a data set generated once.

[0100] Optionally, the more historical records there are, the smaller the target data range will be. The fewer historical records there are, the larger the target data range will be.

[0101] Combined with the scenario example, the more historical records there are, the more times the historical recommendation model has been trained for the personalized needs of users, and the more the recommended objects output by the historical recommendation model meet the needs of users. Further training based on the historical recommendation model does not require a large amount of data, which can avoid model overfitting and reduce the number of training times. The fewer historical records there are, the more data can be increased by increasing the target data range, thereby improving the training effect of the model.

[0102] In this feasible implementation, the target data range is determined according to historical records, which can reduce the number of training times and improve training efficiency on the basis of achieving the target training results.

[0103] S307: Determine a target recommendation model according to the biased data set and the unbiased data set.

[0104] A feasible implementation method can determine the target recommendation model through the following method: determine the historical recommendation model corresponding to the target user identifier; train the historical recommendation model multiple times through the biased data set to obtain multiple intermediate models; test the multiple intermediate models separately through the unbiased data set to obtain the intermediate loss function value corresponding to each intermediate model, until any intermediate loss function value is less than or equal to the loss function threshold, then the corresponding intermediate model is determined as the target recommendation model.

[0105] Next, combine Figure 6 The target recommendation model is described.

[0106] Figure 6 Schematic diagram of a target recommendation model provided in an embodiment of the present application. Figure 6 As shown, determine the historical recommendation model corresponding to the target user identifier. The historical recommendation model is trained multiple times using the biased data set, and an intermediate model is obtained each time. Each intermediate model is tested using the unbiased data set to obtain the intermediate loss function value corresponding to each intermediate model. If the intermediate loss function value is greater than the loss function threshold, continue to train and test the historical recommendation model. If the intermediate loss function value is less than or equal to the loss function threshold, stop training and testing, and determine the current intermediate model as the target recommendation model.

[0107] Combined with the scenario example, the historical recommendation model is the recommendation model obtained from the most recent training corresponding to the target user identifier. The recommendation model obtained from the most recent training meets the user's preferences. Further training on this basis can effectively improve the training effect.

[0108] In this feasible implementation, multiple intermediate models are tested through unbiased datasets to ensure that the final selected model performs well on the unbiased dataset, avoiding overexposure of some objects due to popularity and neglect of other objects due to unpopularity, thereby reducing the bias of the model and improving the recommendation accuracy of the model.

[0109] S308: Input the object feature information into the target recommendation model to obtain the target recommended object.

[0110] It should be noted that the execution process of S308 refers to S205 and will not be repeated here.

[0111] Figure 7 This is a schematic diagram of a recommended device provided in an embodiment of the present application. Figure 7 As shown, the recommendation device 70 may include: a receiving module 71, a determining module 72, a data module 73, a training module 74, and a recommendation module 75, wherein:

[0112] The receiving module 71 is used to receive a recommendation request, where the recommendation request includes a target user identifier and object feature information of a recommended object.

[0113] The determination module 72 is used to determine the user characteristic information corresponding to the target user identification according to the recommendation request.

[0114] The data module 73 is used to determine the biased data set corresponding to the object feature information, and determine the unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator.

[0115] The training module 74 is used to determine a target recommendation model based on the biased data set and the unbiased data set.

[0116] The recommendation module 75 is used to input the object feature information into the target recommendation model to obtain the target recommended object.

[0117] Optionally, the receiving module 71 may execute Figure 2 S201 in the embodiment.

[0118] Optionally, the determination module 72 may execute Figure 2 S202 in the embodiment.

[0119] Optionally, the data module 73 may execute Figure 2 S203 in the embodiment.

[0120] Optionally, the training module 74 may perform Figure 2 S204 in the embodiment.

[0121] Optionally, the recommendation module 75 may perform Figure 2 S205 in the embodiment.

[0122] It should be noted that the recommendation device shown in the embodiment of the present application can execute the technical solution shown in the above method embodiment, and its implementation principle and beneficial effects are similar, which will not be repeated here.

[0123] In a possible implementation, the determination module 72 is specifically configured to:

[0124] Determine a plurality of historical interaction records corresponding to the target user identifier and a plurality of timestamps corresponding to the plurality of historical interaction records;

[0125] Determine multiple target interaction records from multiple historical interaction records according to a preset duration and multiple timestamps;

[0126] Determine user feature information based on multiple target interaction records.

[0127] Figure 8 A schematic diagram of a recommended device provided in an embodiment of the present application. Figure 7 Based on the embodiment shown, Figure 8 As shown, the recommendation device 80 further includes: a screening module 76, a first generating module 77, a second generating module 78, and an execution module 79, wherein:

[0128] The screening module 76 is used to:

[0129] Determine multiple historical feature information corresponding to multiple target interaction records;

[0130] According to the object feature information and the multiple historical feature information, a deviation data set is determined from the multiple target interaction records, and the similarity between the historical feature information corresponding to the deviation data set and the object feature information is greater than or equal to a first threshold.

[0131] The first generating module 77 is used for:

[0132] A user simulator is established according to the user characteristic information, and the user simulator is used to simulate the user's interactive behavior according to the user characteristic information;

[0133] Determine a plurality of objects to be selected according to the object feature information, wherein the similarity between the feature information of the plurality of objects to be selected and the object feature information is greater than or equal to a second threshold;

[0134] Generate an object list corresponding to multiple candidate objects through an unbiased recommendation algorithm, in which each candidate object has the same exposure rate;

[0135] Obtain an unbiased dataset from the object list through the user simulator.

[0136] The second generating module 78 is used for:

[0137] According to the target user identifier, determine the historical record of generating the data set;

[0138] Determine the target data range based on historical records;

[0139] According to the target data range, an unbiased dataset is obtained from the object list through a user simulator, and the data amount of the unbiased dataset is within the target data range.

[0140] The execution module 79 is used to:

[0141] Determine a historical recommendation model corresponding to the target user identifier;

[0142] The historical recommendation model is trained multiple times using the deviation dataset to obtain multiple intermediate models.

[0143] Multiple intermediate models are tested separately using an unbiased data set to obtain the intermediate loss function value corresponding to each intermediate model. When any intermediate loss function value is less than or equal to the loss function threshold, the corresponding intermediate model is determined as the target recommendation model.

[0144] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is shown in FIG. Fig. 9 As shown, the electronic device includes:

[0145] The electronic device includes a processor 291 and a memory 292; it may also include a communication interface 293 and a bus 294. The processor 291, the memory 292, and the communication interface 293 may communicate with each other through the bus 294. The communication interface 293 may be used for information transmission. The processor 291 may call the logic instructions in the memory 292 to execute the method of the above embodiment.

[0146] In addition, the logic instructions in the above-mentioned memory 292 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0147] The memory 292 is a computer-readable storage medium that can be used to store software programs and computer executable programs, such as program instructions / modules corresponding to the methods in the embodiments of the present application. The processor 291 executes functional applications and data processing by running the software programs, instructions, and modules stored in the memory 292, that is, implementing the methods in the above method embodiments.

[0148] The memory 292 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 292 may include a high-speed random access memory and may also include a non-volatile memory.

[0149] An embodiment of the present application provides a non-temporary computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the method as described in the above embodiment.

[0150] An embodiment of the present application provides a computer program product, including a computer program, which implements the method of the above embodiment when the computer program is executed by a processor.

[0151] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0152] It should be further noted that, although the various steps in the flowchart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0153] It should be understood that the above-mentioned device embodiments are only illustrative, and the device of the present application can also be implemented in other ways. For example, the division of units / modules in the above-mentioned embodiments is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units, modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed.

[0154] In addition, unless otherwise specified, each functional unit / module in each embodiment of the present application may be integrated into one unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The above-mentioned integrated unit / module may be implemented in the form of hardware or in the form of a software program module.

[0155] If the integrated unit / module is implemented in the form of hardware, the hardware may be a digital circuit, an analog circuit, etc. The physical implementation of the hardware structure includes but is not limited to transistors, memristors, etc. The processor may be any appropriate hardware processor, such as a CPU, a GPU, an FPGA, a DSP, and an ASIC, etc. The storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc.

[0156] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0157] In the above embodiments, the description of each embodiment has its own emphasis. For the part not described in detail in a certain embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0158] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0159] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A recommendation method, characterized in that: include: receiving a recommendation request, the recommendation request including a target user identifier and object feature information of a recommended object; Determining user feature information corresponding to the target user identifier according to the recommendation request; Determine a biased data set corresponding to the object feature information, and determine an unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator; Determining a target recommendation model according to the biased data set and the unbiased data set; The object feature information is input into the target recommendation model to obtain a target recommendation object.

2. The method according to claim 1, characterized in that Determining user characteristic information corresponding to the target user identifier includes: Determine a plurality of historical interaction records corresponding to the target user identifier and a plurality of timestamps corresponding to the plurality of historical interaction records; Determining a plurality of target interaction records from the plurality of historical interaction records according to a preset duration and the plurality of timestamps; The user characteristic information is determined according to the multiple target interaction records.

3. The method according to claim 2, characterized in that Determining a deviation data set corresponding to the object feature information includes: Determining multiple historical feature information corresponding to the multiple target interaction records; The deviation data set is determined from the multiple target interaction records according to the object feature information and the multiple historical feature information, and the similarity between the historical feature information corresponding to the deviation data set and the object feature information is greater than or equal to a first threshold.

4. The method according to claim 1 or 2, characterized in that: Determining an unbiased data set corresponding to the user feature information by using an unbiased recommendation algorithm and a user simulator includes: Establishing the user simulator according to the user characteristic information, wherein the user simulator is used to simulate the user's interactive behavior according to the user characteristic information; Determine a plurality of objects to be selected according to the object feature information, wherein the similarity between the feature information of the plurality of objects to be selected and the feature information of the object is greater than or equal to a second threshold; Generate an object list corresponding to the plurality of candidate objects by using the unbiased recommendation algorithm, wherein each candidate object in the object list has the same exposure rate; The unbiased data set is obtained from the object list by the user simulator.

5. The method according to claim 4, characterized in that Acquiring the unbiased data set from the object list through the user simulator includes: Determining, according to the target user identifier, a historical record of generating a data set; determining a target data range based on the historical records; According to the target data range, the unbiased data set is acquired from the object list through the user simulator, and the data volume of the unbiased data set is within the target data range.

6. The method according to any one of claims 1 to 5, characterized in that Determining a target recommendation model according to the biased data set and the unbiased data set includes: Determine a historical recommendation model corresponding to the target user identifier; The historical recommendation model is trained multiple times using the deviation data set to obtain multiple intermediate models; The multiple intermediate models are tested and processed separately using the unbiased data set to obtain an intermediate loss function value corresponding to each intermediate model, until any intermediate loss function value is less than or equal to a loss function threshold, the corresponding intermediate model is determined as the target recommendation model.

7. A recommendation device, characterized in that: include: A receiving module, configured to receive a recommendation request, wherein the recommendation request includes a target user identifier and object feature information of a recommended object; A determination module, configured to determine user characteristic information corresponding to the target user identifier according to the recommendation request; A data module, used to determine a biased data set corresponding to the object feature information, and determine an unbiased data set corresponding to the user feature information through an unbiased recommendation algorithm and a user simulator; A training module, used for determining a target recommendation model according to the biased data set and the unbiased data set; The recommendation module is used to input the object feature information into the target recommendation model to obtain a target recommended object.

8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 6 when executed by a processor.

10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 6 when being executed by a processor.

Images