Material recommendation method and device

Abstract

The embodiment of the invention provides a material recommendation method and device, and the method comprises the steps: obtaining a recommendation accuracy rate corresponding to each group of initial fusion weight sets after the material recommendation is carried out on a user group based on a plurality of groups of initial fusion weight sets; determining a group of target fusion weight sets based on the recommendation accuracy corresponding to each group of initial fusion weight sets; and based on the target fusion weight set, carrying out material recommendation on target users except the multiple groups of users. Through the method, the accuracy of material recommendation can be improved.

Description

Technical Field

[0001] The present disclosure relates to but is not limited to the field of Internet technology, and in particular to a material recommendation method and device. Background Art

[0002] The recommendation system is an information filtering system that can recommend suitable materials for users to expose and consume from a large number of materials through layers of screening. Before a material is exposed to the user, it will go through multiple processes such as recall, rough sorting, fine sorting, re-sorting and / or strategy adjustment. In the above processes, there may be a process that requires screening materials that users may be interested in from multiple dimensions, which involves fusion parameters, which are used to perform weighted fusion processing on the results of different dimensions for material screening. However, fixed fusion parameters will affect the recommendation effect. Summary of the invention

[0003] In view of this, the embodiments of the present disclosure provide a material recommendation method and device, an electronic device, and a storage medium, which can improve the accuracy of recommendations.

[0004] The technical solution of the present disclosure is achieved as follows:

[0005] In a first aspect, an embodiment of the present disclosure provides a material recommendation method, comprising:

[0006] After obtaining material recommendations for user groups based on multiple groups of initial fusion weight sets, the recommendation accuracy corresponding to each group of initial fusion weight sets is obtained; wherein the recommendation accuracy is determined based on the preset behaviors of users in the group on the recommended materials; each weight in a group of initial fusion weight sets is associated with a preset behavior, and the weights in different groups of initial fusion weight sets are different; each group of initial fusion weight sets is used to perform weighted fusion on the predicted probability of each preset behavior of users in the group on the materials so as to sort the materials and make recommendations;

[0007] Based on the recommendation accuracy corresponding to each set of initial fusion weight sets, a set of target fusion weight sets is determined;

[0008] Based on the target fusion weight set, material recommendations are made to target users outside the user group; wherein the target fusion weight set is used to perform weighted fusion on the predicted probabilities of the target users predicting various preset behaviors for materials so as to sort the materials and make recommendations.

[0009] In some embodiments, determining a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets includes:

[0010] Determine whether each group of initial fusion weight sets meets a preset convergence condition; wherein satisfying the preset convergence condition means that the difference between each group of initial fusion weight sets is less than a preset difference threshold;

[0011] In the case where each group of initial fusion weight sets meets the preset convergence condition, the target fusion weight set is determined based on the converged multiple groups of weight sets; wherein the converged multiple groups of weight sets refer to the multiple groups of initial fusion weight sets;

[0012] When each group of initial fusion weight sets does not meet the preset convergence condition, a benchmark fusion weight set is selected from the multiple groups of initial fusion weight sets based on the recommendation accuracy corresponding to each group of initial fusion weight sets, and the target fusion weight set is determined based on the benchmark fusion weight set.

[0013] In some embodiments, the benchmark fusion weight set is an initial fusion weight set with the highest recommendation accuracy, and determining the target fusion weight set based on the benchmark fusion weight set includes:

[0014] Based on the benchmark fusion weight set, adjusting each group of initial fusion weight sets other than the benchmark fusion weight set until the multiple groups of adjusted fusion weight sets obtained by adjustment meet the preset convergence condition;

[0015] Based on the converged multiple groups of weight sets, the target fusion weight set is determined; wherein the converged multiple groups of weight sets refer to the multiple groups of adjusted fusion weight sets obtained by adjustment.

[0016] In some embodiments, adjusting each group of initial fusion weight sets other than the reference fusion weight set based on the reference fusion weight set until the multiple groups of adjusted fusion weight sets obtained by adjustment meet the preset convergence condition includes:

[0017] For each group of initial fusion weight sets other than the benchmark fusion weight set, the benchmark fusion weight set is approached successively based on a preset step size and a random number until the multiple groups of adjusted fusion weight sets obtained by adjustment meet the preset convergence condition.

[0018] In some embodiments, the determining the target fusion weight set based on the converged multiple weight sets includes:

[0019] Determine the weight set with the highest recommendation accuracy among the converged weight sets as the target fusion weight set; or,

[0020] Based on the weight associated with each preset behavior in each converged group of weight sets, a weight statistic corresponding to each preset behavior is determined, and the target fusion weight set is generated based on the weight statistic corresponding to each preset behavior.

[0021] In some embodiments, the method further comprises:

[0022] After completing the material recommendation for the target user based on the target fusion weight set, updating the multiple groups of initial fusion weight sets based on the target fusion weight set;

[0023] Based on the updated multiple groups of initial fusion weight sets, materials are recommended to the user group, and the recommendation accuracy corresponding to each group of updated initial fusion weight sets is obtained, so as to update the target fusion weight set and then recommend materials to the target user again.

[0024] In some embodiments, updating the multiple groups of initial fusion weight sets based on the target fusion weight set includes:

[0025] Based on the weight associated with any preset behavior in the target fusion weight set, determining a weight value range corresponding to any preset behavior;

[0026] For each group of initial fusion weight sets to be updated, the weights corresponding to any preset behavior are randomly generated based on the weight value range corresponding to any preset behavior, and based on the randomly generated weights corresponding to any preset behavior and the preset fitting equation, the weights corresponding to preset behaviors other than any preset behavior are fitted to obtain the updated initial fusion weight set.

[0027] In some embodiments, the fitting equation is an ellipse equation, the center of the ellipse represented by the ellipse equation is determined based on the target fusion weight set, the major semi-axis of the represented ellipse is the maximum weight in the target fusion weight set, and the minor semi-axis is the minimum weight in the target fusion weight set.

[0028] In some embodiments, the user group includes multiple groups, the multiple user groups correspond to the multiple initial fusion weight sets one by one, and the users in the multiple user groups are different; after obtaining material recommendations for the user groups based on the multiple initial fusion weight sets, the recommendation accuracy corresponding to each initial fusion weight set includes:

[0029] Obtain log data of recommended materials for each user group after material recommendations are made for each user group based on the corresponding initial fusion weight set;

[0030] Based on the data associated with each preset behavior in the log data of each user group and the preset accuracy detection index, the score value of each user group under the preset accuracy detection index is determined, and the score value of each user group is used as the recommendation accuracy corresponding to each initial fusion weight set.

[0031] In some embodiments, the recommending materials to target users outside the user group based on the target fusion weight set includes:

[0032] In response to the recommendation request of the target user, materials are recalled for the target user to obtain a recalled material candidate set and a plurality of prediction probabilities corresponding to each material in the material candidate set; wherein the prediction probability is a prediction probability of a preset behavior of the target user on the material, and one prediction probability corresponds to one preset behavior;

[0033] For each material in the material candidate set, weight the associated prediction probability using the weight corresponding to each preset behavior in the target fusion weight set, and add the weighted probability values ​​to obtain the recall score value of each material;

[0034] Based on the recall score value of each material in the material candidate set, all materials in the material candidate set are screened, and materials are recommended to the target user based on the screened and retained materials.

[0035] In a second aspect, an embodiment of the present disclosure provides a material recommendation device, including:

[0036] An acquisition module is configured to acquire the recommendation accuracy corresponding to each initial fusion weight set after recommending materials to a user group based on multiple initial fusion weight sets; wherein the recommendation accuracy is determined based on the preset behaviors of the users in the group on the recommended materials; each weight in a set of initial fusion weight sets is associated with a preset behavior, and the weights in the initial fusion weight sets of different groups are different; the initial fusion weight set is used to perform weighted fusion on the predicted probability of the users in the group on the materials to sort the materials and make recommendations;

[0037] A determination module configured to determine a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets;

[0038] The first prediction module is configured to recommend materials to target users outside the user group based on the target fusion weight set; wherein the target fusion weight set is used to perform weighted fusion on the predicted probability of the target user's occurrence of each preset behavior on the material to sort the materials and then make recommendations.

[0039] In some embodiments, the determination module is configured to determine whether each group of initial fusion weight sets meets a preset convergence condition; wherein, meeting the preset convergence condition means that the difference between each group of initial fusion weight sets is less than a preset difference threshold; when each group of initial fusion weight sets meets the preset convergence condition, the target fusion weight set is determined based on the converged multiple groups of weight sets; wherein, the converged multiple groups of weight sets refer to the multiple groups of initial fusion weight sets; when each group of initial fusion weight sets does not meet the preset convergence condition, a baseline fusion weight set is selected from the multiple groups of initial fusion weight sets based on the recommendation accuracy corresponding to each group of initial fusion weight sets, and the target fusion weight set is determined based on the baseline fusion weight set.

[0040] In some embodiments, the determination module is configured to adjust each group of initial fusion weight sets other than the baseline fusion weight set based on the baseline fusion weight set until the multiple groups of adjusted fusion weight sets obtained after adjustment meet the preset convergence conditions; based on the converged multiple groups of weight sets, determine the target fusion weight set; wherein the converged multiple groups of weight sets refer to the multiple groups of adjusted fusion weight sets obtained by adjustment.

[0041] In some embodiments, the determination module is configured to approach the benchmark fusion weight set one by one based on a preset step size and a random number for each group of initial fusion weight sets other than the benchmark fusion weight set until the adjusted multiple groups of adjusted fusion weight sets meet the preset convergence condition.

[0042] In some embodiments, the determination module is configured to determine the weight set with the highest recommendation accuracy among the converged groups of weight sets as the target fusion weight set; or, based on the weight associated with each preset behavior in the converged groups of weight sets, determine the weight statistics corresponding to each preset behavior, and generate the target fusion weight set based on the weight statistics corresponding to each preset behavior.

[0043] In some embodiments, the apparatus further comprises:

[0044] An updating module, configured to update the multiple groups of initial fusion weight sets based on the target fusion weight set after completing the material recommendation for the target user based on the target fusion weight set;

[0045] The second recommendation module is configured to recommend materials to the user group based on the updated multiple groups of initial fusion weight sets, and obtain the recommendation accuracy corresponding to each group of updated initial fusion weight sets, so as to update the target fusion weight set and recommend materials to the target user again.

[0046] In some embodiments, the update module is configured to determine the weight value range corresponding to any preset behavior based on the weight associated with any preset behavior in the target fusion weight set; for each group of initial fusion weight sets to be updated, the weight corresponding to any preset behavior is randomly generated based on the weight value range corresponding to any preset behavior, and based on the randomly generated weight corresponding to any preset behavior and a preset fitting equation, the weight corresponding to the preset behavior other than any preset behavior is fitted to obtain an updated initial fusion weight set.

[0047] In some embodiments, the fitting equation is an ellipse equation, the center of the ellipse represented by the ellipse equation is determined based on the target fusion weight set, the major semi-axis of the represented ellipse is the maximum weight in the target fusion weight set, and the minor semi-axis is the minimum weight in the target fusion weight set.

[0048] In some embodiments, the user group includes multiple groups, the multiple user groups correspond one-to-one to the multiple initial fusion weight sets, and the users in the multiple user groups are different; the acquisition module is configured to obtain log data of recommended materials for each user group after material recommendations are made for each user group based on the corresponding initial fusion weight set; based on the data associated with each preset behavior in the log data of each user group, and the preset accuracy detection index, determine the score value of each user group under the preset accuracy detection index, and use the score value of each user group as the recommendation accuracy corresponding to each initial fusion weight set.

[0049] In some embodiments, the first recommendation module is configured to, in response to the recommendation request of the target user, recall materials for the target user, obtain a recalled material candidate set and multiple prediction probabilities corresponding to each material in the material candidate set; wherein the prediction probability is the prediction probability of the target user performing a preset behavior on the material, and one prediction probability corresponds to one preset behavior; for each material in the material candidate set, the associated prediction probability is weighted using the weight corresponding to each preset behavior in the target fusion weight set, and the weighted probability values ​​are added together to obtain a recall score value for each material; based on the recall score value of each material in the material candidate set, all materials in the material candidate set are screened, and material recommendations are made to the target user based on the screened and retained materials.

[0050] In a third aspect, an embodiment of the present disclosure provides an electronic device, including:

[0051] A memory for storing executable instructions;

[0052] The processor is used to implement the material recommendation method of the embodiment of the present disclosure when executing the executable instructions stored in the memory.

[0053] In a fourth aspect, an embodiment of the present disclosure provides a computer storage medium, wherein the computer storage instructions store a computer program or executable instructions, and when the computer program or executable instructions are executed by a processor, the material recommendation method of the embodiment of the present disclosure is implemented.

[0054] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, which implements the material recommendation method of the embodiment of the present disclosure when the computer program or instructions are executed by a processor.

[0055] In the embodiments of the present disclosure, based on the recommendation accuracy of multiple groups of initial fusion weight sets for material recommendation to user groups, the initial fusion weight sets are tuned to generate target fusion weight sets for material recommendation to target users outside the user group. In this way, the pattern changes of online data can be tracked in real time, which improves the problem of poor recommendation effect caused by fixed fusion parameters, and helps to improve the accuracy of material recommendation to target users. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flow chart of a material recommendation method provided in an embodiment of the present disclosure.

[0057] Figure 2 This is a schematic diagram of a material recommendation method in an embodiment of the present disclosure.

[0058] Figure 3 It is a schematic diagram of adjusting the initial fusion weight set in an embodiment of the present disclosure.

[0059] Figure 4 This is an example diagram of ellipse fitting in an embodiment of the present disclosure.

[0060] Figure 5 A material recommendation device diagram is provided for an embodiment of the present disclosure.

[0061] Figure 6 A schematic diagram of a hardware entity of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0062] The present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not used to limit the present disclosure.

[0063] In the subsequent description, the use of suffixes such as "module", "component" or "unit" to represent elements is only to facilitate the description of the present disclosure, and has no specific meaning in itself. Therefore, "module", "component" or "unit" can be used in a mixed manner. In addition, in the subsequent description, the use of prefixes such as "first" or "second" to identify information is only to facilitate the description of the present disclosure, and has no specific meaning in itself. In addition, in the subsequent description, "at least one" means one or more; "multiple" means two or more. "At least one" is one or more; "multiple" means two or more.

[0064] As mentioned above, fixed fusion parameters will affect the recommendation effect. On the one hand, it is difficult to search for fusion parameters, and a considerable amount of work needs to be invested in selecting the best ones. On the other hand, the distribution of user data and material data changes in real time, and fixed fusion parameters cannot dynamically adapt to the update of data patterns, which affects the final recommendation quality. For example, suppose the optimal fusion parameter is calculated to be 1:10, that is, under this fusion parameter, the platform observation indicators can be optimized and the user experience is the best. However, when the online data distribution changes, for example, due to the adjustment of the video ecology, the overall length of the video is extended, the old fusion parameters may cause the length indicator to increase, but the exposure indicator to decrease, which will have a certain negative impact on business indicators, user experience and the sustainable development of the platform.

[0065] Based on this, the present disclosure provides a material recommendation method. Figure 1 A flow chart of a material recommendation method provided in an embodiment of the present disclosure, such as Figure 1 As shown, the following steps are included:

[0066] S11. After recommending materials to user groups based on multiple groups of initial fusion weight sets, the recommendation accuracy corresponding to each group of initial fusion weight sets is obtained; wherein the recommendation accuracy is determined based on the preset behaviors of the users in the group on the recommended materials; each weight in a group of initial fusion weight sets is associated with a preset behavior, and the weights in the initial fusion weight sets of different groups are different; the initial fusion weight sets are used to perform weighted fusion on the predicted probabilities of the users in the group on the materials to sort the materials and make recommendations;

[0067] S12, determining a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets;

[0068] S13. Recommend materials to target users outside the user group based on the target fusion weight set; wherein the target fusion weight set is used to perform weighted fusion on the predicted probabilities of the target users performing various preset behaviors on the materials so as to sort the materials and make recommendations.

[0069] In the embodiment of the present disclosure, the material recommendation method can be executed by an electronic device, and the electronic device can be a server or a terminal device, which is not limited by the embodiment of the present disclosure.

[0070] In step S11, the electronic device obtains the recommendation accuracy corresponding to each initial fusion weight set after recommending materials to the user groups based on multiple initial fusion weight sets. The multiple initial fusion weight sets may be randomly preset weight sets or weight sets obtained based on guided initialization; the user groups may be any number of groups, such as one or more groups, and the number of users in a group may be arbitrary. In the embodiment of the present disclosure, when there are multiple user groups, the users in different user groups may be different, for example, 40 users, each with 10 people in a group, divided into 4 groups in total.

[0071] In the disclosed embodiment, when there is a group of user groups, multiple groups of initial fusion weight sets are all used for material recommendation for users in the user group; when there are pairs of user groups, the number of multiple groups of initial fusion weight sets is the same as the number of user groups, and one group of users corresponds to one group of initial fusion weight sets, but the weights in the initial fusion weight sets of different groups are different.

[0072] In the disclosed embodiment, a set of initial fusion weights includes multiple weights, each weight is associated with a preset behavior, for example, a weight corresponding to the completion rate and a weight corresponding to the duration, and each weight is used to perform weighted fusion on the probability of each preset behavior occurring among users in the group based on the material prediction, so as to sort the materials and make recommendations.

[0073] Generally, the recall stage of material recommendation needs to cover the materials that the recalled user is interested in as comprehensively as possible, so multiple recalls are often used, and each recall recalls certain materials from different angles. In the embodiments of the present disclosure, taking multiple user groups and the weights in the initial fusion weight set used for weighted fusion in the recall stage as an example, in some embodiments, material recommendations are made to user groups based on multiple initial fusion weight sets, including:

[0074] For each user group, materials are recalled for each user in the user group to obtain a candidate set of materials recalled for each user and multiple prediction probabilities corresponding to each material in the candidate set of materials; wherein the prediction probability is the probability of a preset behavior occurring based on the material predicted for the user, and one prediction probability corresponds to one preset behavior;

[0075] For each material in the material candidate set corresponding to each user, the associated prediction probability is weighted using the weights corresponding to each preset behavior in the initial fusion weight set corresponding to the user group to which the user belongs, and the weighted probability values ​​are added together to obtain the recall score value of each material;

[0076] Based on the recall score of each material in each user's material candidate set, the materials in the material candidate set are screened, and recommendations are made to the user based on the screened and retained materials.

[0077] In the disclosed embodiment, if there is only one user group, when recommending materials to the users in the user group, for each material in the material candidate set corresponding to each user, the weight corresponding to each preset behavior in each set of initial fusion weight sets is used to weight the associated prediction probability, and the weighted probability values ​​are added together to obtain the recall score value of each material. Then, for each user in the user group, each recalled material has a different recall score value based on a different initial fusion weight set. Based on the recall score value of each material corresponding to a set of initial fusion weight sets, the materials in the material candidate set can be screened, and recommendations can be made to the user based on the screened retained materials. In some embodiments, the user group includes multiple groups, and the multiple user groups correspond one-to-one to the multiple initial fusion weight sets, and the users in the multiple user groups are different;

[0078] The acquisition of the recommendation accuracy corresponding to each set of initial fusion weights after material recommendations are made to the user group based on multiple sets of initial fusion weights includes:

[0079] Obtain log data of recommended materials for each user group after material recommendations are made for each user group based on the corresponding initial fusion weight set;

[0080] Based on the data associated with each preset behavior in the log data of each user group and the preset accuracy detection index, the score value of each user group under the preset accuracy detection index is determined, and the score value of each user group is used as the recommendation accuracy corresponding to each initial fusion weight set.

[0081] In the disclosed embodiment, video recommendation is still used as an example for explanation. It is assumed that the probability of the occurrence of the preset behavior is weighted based on the weights corresponding to the two preset behaviors of completion and reaching the preset duration. If the predicted completion rate (i.e., the probability of completion of playback) of a certain material is 0.5, the duration rate (i.e., the probability of watching for 30 seconds) is 0.8, the weight of the completion rate is 0.6, and the weight of the duration rate is 0.4, then the recall score of the material is 0.64 (0.5*0.8+0.6*0.4). Based on the recall score of each material, a rough sorting can be performed, such as selecting the top 100 materials with larger recall scores as the materials after rough sorting and further performing fine sorting or strategy and other subsequent processes to further screen out the final materials and recommend them to each user.

[0082] After recommending materials to users, the electronic device will obtain log data for each group of users for the recommended materials. The log data includes data on whether the user has performed preset behaviors, such as data on the completion of broadcasting and reaching the preset duration. The electronic device determines the recommendation accuracy of each group of users based on the data associated with each preset behavior and combined with the preset accuracy detection index. The log data can be kafka log data, and the log data can also include user characteristics and / or material characteristics.

[0083] Exemplarily, the preset accuracy detection indicators include scale, efficiency and single-channel indicators. For each group of users, the score value of each group of users under the preset accuracy detection indicators can be determined based on the total completion volume and / or the total amount reaching the preset duration in the scale indicator, combined with the per capita completion rate in the efficiency indicator and the amount of exposed materials in the single-channel indicator. The score value represents the recommendation accuracy. For example, the larger the score value, the higher the recommendation accuracy.

[0084] It should be noted that the embodiments of the present disclosure are not limited to the application of the initial fusion weight set to the multi-way recall stage, but can also be applied to other stages after the recall, which is not limited by the embodiments of the present disclosure.

[0085] In the disclosed embodiment, when a group of user groups are recommended materials using multiple sets of initial fusion weight sets, for each set of initial fusion weight sets, the recalled materials can be used to calculate the recall score value and perform material screening and further recommendation, and then the recommendation accuracy of the user group under each set of initial fusion weight sets is determined based on the log data of the user group for the recommended materials. It should be noted that in the disclosed embodiment, different recall scores are obtained after weighting the probabilities of recalled materials based on different initial fusion weight sets. Different recall scores will result in different results of subsequent rough sorting, fine sorting, etc., so that different materials are recommended to each user, and the corresponding recommendation accuracy will also be different.

[0086] In step S12, the electronic device determines a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets, and the target fusion weight set also includes weights associated with each preset behavior, and one weight is associated with one preset behavior. Exemplarily, the initial fusion weight set corresponding to the user group with the highest recommendation accuracy can be directly used as the target fusion weight set, or a set of target fusion weight sets can be determined based on multiple sets of initial fusion weight sets.

[0087] In step S13, the electronic device recommends materials to target users outside the user group based on the target fusion weight set. The target user may be a user who has a recommendation request synchronously with the user group, or a user who has a recommendation request after the user group's recommendation request, which is not limited in the embodiment of the present disclosure.

[0088] Taking the application of the target fusion weight set in the recall stage as an example, in some embodiments, recommending materials to target users outside the user group based on the target fusion weight set includes:

[0089] In response to the recommendation request of the target user, materials are recalled for the target user to obtain a recalled material candidate set and a plurality of prediction probabilities corresponding to each material in the material candidate set; wherein the prediction probability is a prediction probability of a preset behavior of the target user on the material, and one prediction probability corresponds to one preset behavior;

[0090] For each material in the material candidate set, weight the associated prediction probability using the weight corresponding to each preset behavior in the target fusion weight set, and add the weighted probability values ​​to obtain the recall score value of each material;

[0091] Based on the recall score value of each material in the material candidate set, all materials in the material candidate set are screened, and materials are recommended to the target user based on the screened and retained materials.

[0092] It should be noted that in the disclosed embodiment, the material recommendation method for the target user is the same as the material recommendation method for the user group, except that the fusion weight set involved in the recommendation process is different. In addition, in the disclosed embodiment, the recommendation accuracy corresponding to each set of initial fusion weight sets can be quickly determined based on the real-time computing framework Flink, so that the target fusion weight set can be determined in a relatively short time.

[0093] It can be understood that in the embodiments of the present disclosure, based on the recommendation accuracy of multiple groups of initial fusion weight sets for material recommendations to user groups, the initial fusion weight sets are tuned to generate target fusion weight sets for material recommendations to target users outside the user group. In this way, the pattern changes of online data can be tracked in real time, which improves the problem of poor recommendation effect caused by fixed fusion parameters, and helps to improve the accuracy of material recommendations to target users.

[0094] In addition, in the embodiments of the present disclosure, a target fusion weight set is determined based on the recommendation accuracy of material recommendations for multiple user groups based on multiple initial fusion weight sets. Since multiple user groups cover more users, the recommendation accuracy corresponding to each initial fusion weight set will be more convincing, and the target fusion weight set determined on this basis will also be more accurate. In the embodiments of the present disclosure, material recommendations can also be made to multiple user groups in parallel based on multiple initial fusion weight sets. Without affecting the recommendations for the users included in the multiple user groups, the speed of determining the target fusion weight set can be further accelerated, thereby accelerating the material recommendations to the target users.

[0095] Figure 2Schematic diagram of a material recommendation method in an embodiment of the present disclosure. Figure 2 As shown, if the electronic device receives a user request L21, the recommendation service is started, and user features and material features are extracted based on feature engineering L22, and then the completion rate score (i.e., completion rate probability) and duration score (i.e., duration probability) are obtained based on the trained model L23. The model can be a multi-target recall model, one target is the recall based on the completion rate, and the other target is the recall based on the duration. Based on the recalled materials, the electronic device further performs subsequent weighted fusion based on the recommendation engine L24, that is, the output probability of the corresponding preset behavior (i.e., the target) is weighted with the weights in the initial fusion weight set to obtain the recall score value of the recalled materials, and the materials are subsequently subjected to rough sorting, fine sorting, and other processes for recommendation based on the recall score value. Based on the recommended materials, the electronic device obtains log data L25, where the log data includes log data of multiple user groups requesting the recommendation service, and may also include log data of a group of user groups for each group of initial fusion weight sets. The electronic device determines the recommendation accuracy of each initial fusion weight set based on the data of preset behaviors associated with each user in the log data and the preset accuracy detection index, and finally determines a set of target fusion weight sets through the adaptive fusion parameter search method L26, that is, determines a set of completion rate weights and duration weights, and re-sends them to the recommendation engine L24, so as to make material recommendations to the target users after receiving recommendation requests from target users outside the user group.

[0096] As mentioned above, the log data L25 may also include user characteristics and / or material characteristics. The user characteristics may include the user's behavior characteristics for the material after the recommendation. The material characteristics may include the material popularity determined based on the user behavior data of all user groups, etc. Therefore, based on the user characteristics and / or material characteristics in the log data L25, the model L23 may also be retrained to optimize the model. In the above schematic diagram, the adaptive fusion parameter search method L26, that is, the method for determining the target fusion weight set, can be described as follows.

[0097] In some embodiments, determining a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets includes:

[0098] Determine whether each group of initial fusion weight sets meets a preset convergence condition; wherein satisfying the preset convergence condition means that the difference between each group of initial fusion weight sets is less than a preset difference threshold;

[0099] In the case where each group of initial fusion weight sets meets the preset convergence condition, the target fusion weight set is determined based on the converged multiple groups of weight sets; wherein the converged multiple groups of weight sets refer to the multiple groups of initial fusion weight sets;

[0100] When each group of initial fusion weight sets does not meet the preset convergence condition, a benchmark fusion weight set is selected from the multiple groups of initial fusion weight sets based on the recommendation accuracy corresponding to each group of initial fusion weight sets, and the target fusion weight set is determined based on the benchmark fusion weight set.

[0101] In the disclosed embodiment, the electronic device determines the target fusion weight set in a targeted manner according to whether the multiple groups of initial fusion weight sets converge. Wherein, when the multiple groups of initial fusion weight sets converge, because the difference between the initial fusion weight sets of each group is small, the influence of different initial fusion weight sets on the recommendation accuracy will not be too great. On this basis, the target fusion weight set is directly determined based on the multiple groups of initial fusion weight sets, which can simplify the determination process of the target fusion weight set and take into account the accuracy of the target fusion weight set; and when the multiple groups of initial fusion weight sets do not converge, because the difference between the initial fusion weight sets of each group is large, the influence of different initial fusion weight sets on the recommendation accuracy may also be large. Based on this, the electronic device selects a reference fusion weight set from the multiple groups of initial fusion weight sets based on the recommendation accuracy corresponding to each group of initial fusion weight sets, and determines the target fusion weight set based on the reference fusion weight set. Wherein, the reference fusion weight set can be the initial fusion weight set corresponding to the user group with the maximum recommendation accuracy, or the initial fusion weight set corresponding to the user group with the intermediate recommendation accuracy, and the target fusion weight set can be the reference fusion weight set itself, or a weight set further calculated and determined based on the reference fusion weight set.

[0102] In some embodiments, the benchmark fusion weight set is an initial fusion weight set with the highest recommendation accuracy, and determining the target fusion weight set based on the benchmark fusion weight set includes:

[0103] Based on the benchmark fusion weight set, adjusting each group of initial fusion weight sets other than the benchmark fusion weight set until the multiple groups of adjusted fusion weight sets obtained after adjustment meet the preset convergence condition;

[0104] Based on the converged multiple groups of weight sets, the target fusion weight set is determined; wherein the converged multiple groups of weight sets refer to the multiple groups of adjusted fusion weight sets obtained by adjustment.

[0105] In an embodiment of the present disclosure, when multiple groups of initial fusion weight sets do not converge, the determined baseline fusion weight set is the initial fusion weight set with the highest recommendation accuracy. The electronic device adjusts other initial fusion weight sets based on the baseline fusion weight set until the adjusted multiple groups of adjusted fusion weight sets converge, and then determines the target fusion weight set based on the converged multiple groups of weight sets.

[0106] In some embodiments, the determining the target fusion weight set based on the converged multiple weight sets includes:

[0107] Determine the weight set with the highest recommendation accuracy among the converged weight sets as the target fusion weight set; or,

[0108] Based on the weight associated with each preset behavior in each converged group of weight sets, a weight statistic corresponding to each preset behavior is determined, and the target fusion weight set is generated based on the weight statistic corresponding to each preset behavior.

[0109] In the disclosed embodiment, the converged multiple sets of weights may be the aforementioned converged multiple sets of initial fusion weights, or may be the multiple sets of weights that converge after adjustment, wherein the multiple sets of weights that converge after adjustment include a baseline fusion weight set, i.e., an initial fusion weight set with the highest recommendation accuracy. Since the highest recommendation accuracy indicates the best recommendation effect, the weight set with the highest recommendation accuracy is the weight set that is suitable for the current material recommendation. Based on this, in some embodiments, the weight set with the highest recommendation accuracy can be directly determined as the target fusion weight set.

[0110] In other embodiments, considering that the weight set with the highest recommendation accuracy may not be the true optimal solution, for example, it may be affected by interference such as calculation accuracy. Therefore, in the embodiments of the present disclosure, the weight statistics corresponding to each preset behavior can be determined based on the weights associated with each preset behavior in the converged groups of weight sets, and the target fusion weight set can be generated based on the weight statistics corresponding to each preset behavior; wherein the weight statistics can be the mean or median, etc., which is not limited by the embodiments of the present disclosure.

[0111] It can be understood that in the embodiments of the present disclosure, the electronic device specifically determines the target fusion weight set, which is highly intelligent.

[0112] In some embodiments, adjusting each group of initial fusion weight sets other than the reference fusion weight set based on the reference fusion weight set until the multiple groups of adjusted fusion weight sets obtained after adjustment meet the preset convergence condition includes:

[0113] For each group of initial fusion weight sets other than the benchmark fusion weight set, the benchmark fusion weight set is approached successively based on a preset step size and a random number until the multiple groups of adjusted fusion weight sets obtained by adjustment meet the preset convergence condition.

[0114] In the embodiment of the present disclosure, when adjusting each initial fusion weight set other than the reference fusion weight set to approach the reference fusion weight set, the adjustment may be performed based on the following formula (1):

[0115] ω i =(ω opt -ωi )*lr+ω i +rand,i=1,2,3....n-1 (1)

[0116] in, is the initial fusion weight set corresponding to the user group with the highest recommendation accuracy, that is, the benchmark fusion weight set, also called the "optimal group". represents the fusion weight of the completion rate, Indicates the fusion weight of duration.

[0117] In addition, lr in formula (1) represents the step length, which determines how far each step is taken when approaching the optimal group; rand is a Gaussian random number. Because the current "optimal group" is not necessarily the optimal, this random number is introduced to increase randomness and give a certain probability to search around itself.

[0118] In the disclosed embodiment, the electronic device approaches the "optimal group" one by one based on a preset step size and a random number for each initial fusion weight set other than the "optimal group" until the adjusted multiple weight sets converge. Through this random weight update strategy, an attempt is made to move the current weight toward a hypothetical optimal weight. At the same time, the exploration space is increased by controlling the step size and adding randomness. This method helps to reduce the occurrence of overfitting.

[0119] It should be noted that in the embodiments of the present disclosure, in the adjusted multiple weight sets, the fusion weight sets of different groups are different. The electronic device can directly determine the target fusion weight set based on the multiple weight sets that converge after the adjustment, or it can recommend materials to the user group again based on the multiple weight sets that converge after the adjustment, and re-determine the recommendation accuracy corresponding to each initial fusion weight set by means such as log data, and determine the target fusion weight set based on the multiple weight sets that converge after the adjustment when the recommendation accuracy is improved compared to before the adjustment. The embodiments of the present disclosure do not limit this.

[0120] Figure 3 is a schematic diagram of adjusting the initial fusion weight set in an embodiment of the present disclosure, such as Figure 3 As shown, after obtaining the log data L31, the electronic device calculates the index calculation value corresponding to each group of users based on the data of user-associated preset behaviors in each user group included in the log data L31 and the observation index L32, that is, calculates the recommendation accuracy of each group of users. Figure 3The flow n-fusion parameter n in the formula indicates that a group of users corresponds to a group of initial fusion weight sets, and the observation index L32 is the preset accuracy detection index. Different groups of users use the same index to calculate the recommendation accuracy. After obtaining the index calculation values ​​of each group of users, the electronic device can sort the index values ​​based on L33 and select the optimal fusion parameter group. The optimal fusion parameter group is the aforementioned benchmark fusion weight set, which can be the initial fusion weight set with the highest recommendation accuracy. Further, the electronic device determines whether all parameter pairs (i.e., multiple groups of initial fusion weight sets) converge based on L34. If converged, the optimal fusion parameter group is updated online based on L35 and all group parameters are reinitialized to enter the next round of material recommendation. If not converged, the remaining three exploration groups are approached to the optimal group parameters with a certain disturbance based on L36. Among them, updating the optimal fusion parameter group online will recommend the initial fusion weight set with the highest accuracy as the target fusion weight set to the target user for recommendation. In addition, the purpose of further reinitializing all group parameters of the electronic device is to update multiple groups of initial fusion weight sets to cope with changes such as material data. Furthermore, if convergence does not occur, the remaining three exploration groups are moved closer to the optimal group parameters, that is, each initial fusion weight set other than the baseline fusion weight set is adjusted until the adjusted multiple fusion weight sets converge. In the disclosed embodiment, after the multiple fusion weight sets are adjusted to converge or after all group parameters are reinitialized, recommendations can be made again based on the updated fusion parameters of each exploration group in L37 to generate new log data.

[0121] As mentioned above, the purpose of reinitializing all group parameters is to perform the next round of material recommendation. In some embodiments, the method further includes:

[0122] After completing the material recommendation for the target user based on the target fusion weight set, updating the multiple groups of initial fusion weight sets based on the target fusion weight set;

[0123] Based on the updated multiple groups of initial fusion weight sets, materials are recommended to the user group, and the recommendation accuracy corresponding to each group of updated initial fusion weight sets is obtained, so as to update the target fusion weight set and then recommend materials to the target user again.

[0124] In the disclosed embodiment, the updated multiple groups of initial fusion weight sets are used to recommend materials to the user group again on the one hand, and to update the target fusion weight set based on the recommendation accuracy of material recommendation to the user group to recommend materials to the target user again on the other hand.

[0125] It can be understood that after the embodiment of the present disclosure completes the material recommendation for the target user based on the target fusion weight set, the target fusion weight set is used to guide the update of multiple groups of initial fusion weight sets. On the one hand, it can respond immediately to changes in material data, etc.; on the other hand, since the target fusion weight set is a better fusion weight set determined in the previous round, the multiple groups of initial fusion weight sets are improved based on the guidance of the better fusion weight set, which can improve the accuracy of the update of the multiple groups of initial fusion weight sets.

[0126] In some embodiments, updating the multiple groups of initial fusion weight sets based on the target fusion weight set includes:

[0127] Based on the weight associated with any preset behavior in the target fusion weight set, determining a weight value range corresponding to any preset behavior;

[0128] For each group of initial fusion weight sets to be updated, the weights corresponding to any preset behavior are randomly generated based on the weight value range corresponding to any preset behavior, and based on the randomly generated weights corresponding to any preset behavior and the preset fitting equation, the weights corresponding to preset behaviors other than any preset behavior are fitted to obtain the updated initial fusion weight set.

[0129] In the embodiment of the present disclosure, when updating multiple groups of initial fusion weight sets, the electronic device determines the weight value range corresponding to any preset behavior based on the weight associated with any preset behavior in the target fusion weight set. For example, the preset behavior is the completion rate. The embodiment of the present disclosure can determine the value range [MIN, MAX] of the completion rate parameter based on the following formula (2):

[0130]

[0131] in, As described above, it is the fusion weight of the completion rate in the initial fusion weight set with the highest recommendation accuracy.

[0132] In the embodiment of the present disclosure, after obtaining the weight value range corresponding to any preset behavior, a weight corresponding to any preset behavior can be randomly generated for each set of initial fusion weights to be updated, and the weights corresponding to other preset behaviors can be determined by fitting to obtain each set of updated fusion weights. In the embodiment of the present disclosure, the weight corresponding to any preset behavior can be randomly generated based on the following formula (3):

[0133]

[0134] Among them, rand represents a random number, That is, the weight of any preset behavior randomly generated in the i-th group, for example, the weight of the completion rate.

[0135] It should be noted that in the embodiments of the present disclosure, when fitting and generating weights corresponding to preset behaviors other than any preset behaviors, it can be based on linear fitting, polynomial fitting, exponential equation fitting, or elliptic equation fitting, etc., and the embodiments of the present disclosure do not limit this. In addition, in the embodiments of the present disclosure, it is also possible to randomly initialize the weights corresponding to any preset behavior for each set of weights based on the aforementioned converged multiple sets of weights, and fit and generate weights corresponding to each preset behavior other than any preset behavior, rather than being limited to randomly generating the weights corresponding to any preset behavior in each set of initial fusion weights to be updated based on the target fusion weight set.

[0136] It can be understood that in the embodiment of the present disclosure, the electronic device randomly generates weights corresponding to any preset behaviors in each other group of initial fusion weight sets to be updated based on the target fusion weight set, and fits the method of generating weights corresponding to preset behaviors other than any preset behaviors. The scheme is simple and can improve the efficiency of updating the fusion weight set.

[0137] In some embodiments, the fitting equation is an ellipse equation, the center of the ellipse represented by the ellipse equation is determined based on the target fusion weight set, the major semi-axis of the represented ellipse is the maximum weight in the target fusion weight set, and the minor semi-axis is the minimum weight in the target fusion weight set.

[0138] In the embodiment of the present disclosure, the fitting equation is the ellipse equation of the following formula (4):

[0139]

[0140] Among them, the ellipse center (x0, y0) can be the target fusion weight set λ1 is the semi-axis length of the x-axis, and λ2 is the semi-axis length of the y-axis.

[0141] In the embodiment of the present disclosure, taking into account that the scales of the two parameters are usually different and the search ranges required are different, a larger value requires a larger search space. Therefore, in the embodiment of the present disclosure, the major semi-axis of the ellipse is set as the maximum weight in the target fusion weight set, and the minor semi-axis is set as the minimum weight in the target fusion weight set. In this way, the update of the fusion weight set can be optimized, which helps to determine a more accurate target fusion weight set, and helps to improve the accuracy of recommendations for the next round of user groups and / or target users.

[0142] For example, in the above ellipse equation Less than So you can set is the radius of the minor axis, is the semi-major axis radius.

[0143] In the embodiment of the present disclosure, when generating the weight corresponding to any preset behavior other than the preset behavior based on the elliptic equation fitting, the following formula (5) may be referred to:

[0144]

[0145] Among them, flag∈{1,-1} is the preset value, is the weight of the preset behavior of the i-th group other than any preset behavior generated based on ellipse fitting, for example, the weight of the duration.

[0146] Figure 4 is an example diagram of an ellipse fitting in an embodiment of the present disclosure, such as Figure 4 As shown, the major semi-axis is The minor semiaxis is The center of the ellipse is the convergence result of the previous round of search, that is, the target fusion weight set. At the beginning of a round of search, the initial points of the four exploration groups are located on the edges of the ellipse, such as Figure 4 As shown in L41, L42, L43 and L44. In each round of recommendation, by calculating the observation indicators of each exploration group, the observation indicators are optimized, and the optimal group is "approached" and updated at each step. The fusion parameters (initial fusion weight set) of all exploration groups gradually move in the same direction (as shown by the arrow in the middle) and finally converge. At this time, the new final optimal fusion parameters (target fusion weight set) are generated. For example, the optimal fusion parameter is the average of the converged multiple groups of fusion parameters.

[0147] Figure 5 A material recommendation device diagram is provided for an embodiment of the present disclosure, such as Figure 5 As shown, the device 500 includes:

[0148] The acquisition module 501 is configured to acquire the recommendation accuracy corresponding to each initial fusion weight set after recommending materials to the user group based on multiple groups of initial fusion weight sets; wherein the recommendation accuracy is determined based on the preset behaviors of the users in the group on the recommended materials; each weight in a group of initial fusion weight sets is associated with a preset behavior, and the weights in the initial fusion weight sets of different groups are different; the initial fusion weight set is used to perform weighted fusion on the predicted probability of the users in the group to predict the occurrence of each preset behavior based on the material prediction, so as to sort the materials and make recommendations;

[0149] A determination module 502 is configured to determine a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets;

[0150] The first prediction module 503 is configured to recommend materials to target users outside the user group based on the target fusion weight set; wherein the target fusion weight set is used to perform weighted fusion on the predicted probability of the target user performing each preset behavior on the material to sort the materials and make recommendations.

[0151] In some embodiments, the determination module 502 is configured to determine whether each group of initial fusion weight sets meets a preset convergence condition; wherein, meeting the preset convergence condition means that the difference between each group of initial fusion weight sets is less than a preset difference threshold; when each group of initial fusion weight sets meets the preset convergence condition, the target fusion weight set is determined based on the converged multiple groups of weight sets; wherein, the converged multiple groups of weight sets refer to the multiple groups of initial fusion weight sets; when each group of initial fusion weight sets does not meet the preset convergence condition, a baseline fusion weight set is selected from the multiple groups of initial fusion weight sets based on the recommendation accuracy corresponding to each group of initial fusion weight sets, and the target fusion weight set is determined based on the baseline fusion weight set.

[0152] In some embodiments, the determination module 502 is configured to adjust each group of initial fusion weight sets other than the baseline fusion weight set based on the baseline fusion weight set, until the multiple groups of adjusted fusion weight sets obtained after adjustment meet the preset convergence condition; based on the converged multiple groups of weight sets, determine the target fusion weight set; wherein the converged multiple groups of weight sets refer to the multiple groups of adjusted fusion weight sets obtained by adjustment.

[0153] In some embodiments, the determination module 502 is configured to approach the benchmark fusion weight set one by one based on a preset step size and a random number for each group of initial fusion weight sets other than the benchmark fusion weight set, until the adjusted multiple groups of adjusted fusion weight sets meet the preset convergence condition.

[0154] In some embodiments, the determination module 502 is configured to determine the weight set with the highest recommendation accuracy among the converged groups of weight sets as the target fusion weight set; or, based on the weight associated with each preset behavior in the converged groups of weight sets, determine the weight statistics corresponding to each preset behavior, and generate the target fusion weight set based on the weight statistics corresponding to each preset behavior.

[0155] In some embodiments, the apparatus further comprises:

[0156] An updating module, configured to update the multiple groups of initial fusion weight sets based on the target fusion weight set after completing the material recommendation for the target user based on the target fusion weight set;

[0157] The second recommendation module is configured to recommend materials to the user group based on the updated multiple groups of initial fusion weight sets, and obtain the recommendation accuracy corresponding to each group of updated initial fusion weight sets, so as to update the target fusion weight set and recommend materials to the target user again.

[0158] In some embodiments, the update module is configured to determine the weight value range corresponding to any preset behavior based on the weight associated with any preset behavior in the target fusion weight set; for each group of initial fusion weight sets to be updated, the weight corresponding to any preset behavior is randomly generated based on the weight value range corresponding to any preset behavior, and based on the randomly generated weight corresponding to any preset behavior and a preset fitting equation, the weight corresponding to the preset behavior other than any preset behavior is fitted to obtain an updated initial fusion weight set.

[0159] In some embodiments, the fitting equation is an ellipse equation, the center of the ellipse represented by the ellipse equation is determined based on the target fusion weight set, the major semi-axis of the represented ellipse is the maximum weight in the target fusion weight set, and the minor semi-axis is the minimum weight in the target fusion weight set.

[0160] In some embodiments, the user group includes multiple groups, the multiple user groups correspond one-to-one to the multiple initial fusion weight sets, and the users in the multiple user groups are different; the acquisition module 501 is configured to obtain log data of recommended materials for each user group after material recommendations are made to each user group based on the corresponding initial fusion weight set; based on the data associated with each preset behavior in the log data of each user group, and the preset accuracy detection index, determine the score value of each user group under the preset accuracy detection index, and use the score value of each user group as the recommendation accuracy corresponding to each initial fusion weight set.

[0161] In some embodiments, the first recommendation module 503 is configured to, in response to the recommendation request of the target user, recall materials for the target user, and obtain a recalled material candidate set and multiple predicted probabilities corresponding to each material in the material candidate set; wherein the predicted probability is the predicted probability of the target user performing a preset behavior on the material, and one predicted probability corresponds to one preset behavior; for each material in the material candidate set, the associated predicted probabilities are weighted using the weights corresponding to the preset behaviors in the target fusion weight set, and the weighted probability values ​​are added together to obtain a recall score value for each material; based on the recall score value of each material in the material candidate set, all materials in the material candidate set are screened, and material recommendations are made to the target user based on the screened and retained materials.

[0162] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0163] Figure 6 is a schematic diagram of a hardware entity of an electronic device in an embodiment of the present disclosure, such as Figure 6 As shown, the hardware entity of the computer device 600 includes: a processor 601, a communication interface 602 and a memory 603, wherein:

[0164] Processor 601 generally controls the overall operation of computer device 600 .

[0165] The communication interface 602 enables the computer device to communicate with other terminals or servers through a network.

[0166] The memory 603 is configured to store instructions and applications executable by the processor 601, and can also cache data to be processed or processed by the processor 601 and each module in the computer device 600 (for example, image data, audio data, voice communication data, and video communication data), which can be implemented by flash memory (FLASH) or random access memory (Random Access Memory, RAM). Data can be transmitted between the processor 601, the communication interface 602 and the memory 603 through the bus 604. Among them, the processor 601 is used to execute some or all of the steps in the above method.

[0167] Correspondingly, an embodiment of the present disclosure provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, part or all of the steps in the above method are implemented.

[0168] The embodiment of the present disclosure provides a computer program product, which includes: a computer program or executable instructions, which are stored in a computer-readable storage medium. The processor of the computer device reads the computer program or executable instructions from the computer-readable storage medium, and the processor executes the computer program or executable instructions, so that the computer device executes any one of the above-mentioned material recommendation methods of the embodiment of the present disclosure.

[0169] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of the present disclosure, please refer to the description of the method embodiments of the present disclosure for understanding.

[0170] It should be understood that in various embodiments of the present disclosure, the size of the sequence number of each process does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure. The sequence numbers of the embodiments of the present disclosure are only for description and do not represent the advantages and disadvantages of the embodiments.

[0171] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0172] In the several embodiments provided in the present disclosure, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0173] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0174] In addition, all functional units in the embodiments of the present disclosure may be integrated into one processing unit, or each unit may be separately configured as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0175] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0176] Alternatively, if the above-mentioned integrated unit of the present disclosure is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure can essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0177] The above description is only an embodiment of the present disclosure and is not intended to limit the protection scope of the present disclosure. Any modification, equivalent replacement and improvement made within the spirit and scope of the present disclosure are included in the protection scope of the present disclosure.

Claims

1. A material recommendation method, characterized in that: include: After recommending materials to user groups based on multiple groups of initial fusion weight sets, the recommendation accuracy corresponding to each group of initial fusion weight sets is obtained; wherein the recommendation accuracy is determined based on the preset behaviors of users in the group on the recommended materials; each weight in a group of initial fusion weight sets is associated with a preset behavior, and the weights in the initial fusion weight sets of different groups are different; the initial fusion weight sets are used to perform weighted fusion on the predicted probabilities of users in the group on the materials to perform sorting and recommendation; Based on the recommendation accuracy corresponding to each set of initial fusion weight sets, a set of target fusion weight sets is determined; Based on the target fusion weight set, material recommendations are made to target users outside the user group; wherein the target fusion weight set is used to perform weighted fusion on the predicted probabilities of the target users performing various preset behaviors on the materials so as to sort the materials and make recommendations.

2. The method according to claim 1, characterized in that The step of determining a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets includes: Determine whether each group of initial fusion weight sets meets a preset convergence condition; wherein satisfying the preset convergence condition means that the difference between each group of initial fusion weight sets is less than a preset difference threshold; When each group of initial fusion weight sets meets the preset convergence condition, the target fusion weight set is determined based on the converged multiple groups of weight sets; wherein the converged multiple groups of weight sets refer to the multiple groups of initial fusion weight sets; when each group of initial fusion weight sets does not meet the preset convergence condition, a baseline fusion weight set is selected from the multiple groups of initial fusion weight sets based on the recommendation accuracy corresponding to each group of initial fusion weight sets, and the target fusion weight set is determined based on the baseline fusion weight set.

3. The method according to claim 2, characterized in that The benchmark fusion weight set is an initial fusion weight set with the highest recommendation accuracy, and determining the target fusion weight set based on the benchmark fusion weight set includes: Based on the benchmark fusion weight set, adjusting each group of initial fusion weight sets other than the benchmark fusion weight set until the multiple groups of adjusted fusion weight sets obtained by adjustment meet the preset convergence condition; Based on the converged multiple groups of weight sets, the target fusion weight set is determined; wherein the converged multiple groups of weight sets refer to the multiple groups of adjusted fusion weight sets obtained by adjustment.

4. The method according to claim 2 or 3, characterized in that: The step of determining the target fusion weight set based on the converged multiple weight sets includes: Determine the weight set with the highest recommendation accuracy among the converged weight sets as the target fusion weight set; or, Based on the weight associated with each preset behavior in each converged group of weight sets, a weight statistic corresponding to each preset behavior is determined, and the target fusion weight set is generated based on the weight statistic corresponding to each preset behavior.

5. The method according to claim 1, characterized in that The method further comprises: After completing the material recommendation for the target user based on the target fusion weight set, updating the multiple groups of initial fusion weight sets based on the target fusion weight set; Based on the updated multiple groups of initial fusion weight sets, materials are recommended to the user group, and the recommendation accuracy corresponding to each group of updated initial fusion weight sets is obtained, so as to update the target fusion weight set and then recommend materials to the target user again.

6. The method according to claim 5, characterized in that The updating of the multiple groups of initial fusion weight sets based on the target fusion weight set includes: Based on the weight associated with any preset behavior in the target fusion weight set, determining a weight value range corresponding to any preset behavior; For each group of initial fusion weight sets to be updated, the weights corresponding to any preset behavior are randomly generated based on the weight value range corresponding to any preset behavior, and based on the randomly generated weights corresponding to any preset behavior and the preset fitting equation, the weights corresponding to preset behaviors other than any preset behavior are fitted to obtain the updated initial fusion weight set.

7. The method according to claim 6, characterized in that The fitting equation is an ellipse equation, the center of the ellipse represented by the ellipse equation is determined based on the target fusion weight set, the major semi-axis of the represented ellipse is the maximum weight in the target fusion weight set, and the minor semi-axis is the minimum weight in the target fusion weight set.

8. The method according to claim 1 or 5, characterized in that: The user group includes multiple groups, the multiple user groups correspond to the multiple initial fusion weight sets one by one, and the users in the multiple user groups are different; The acquisition of the recommendation accuracy corresponding to each set of initial fusion weights after material recommendations are made to the user group based on multiple sets of initial fusion weights includes: Obtain log data of recommended materials for each user group after material recommendations are made for each user group based on the corresponding initial fusion weight set; Based on the data associated with each preset behavior in the log data of each user group and the preset accuracy detection index, the score value of each user group under the preset accuracy detection index is determined, and the score value of each user group is used as the recommendation accuracy corresponding to each initial fusion weight set.

9. The method according to claim 1 or 5, characterized in that: The recommending materials to target users outside the user group based on the target fusion weight set includes: In response to the recommendation request of the target user, materials are recalled for the target user to obtain a recalled material candidate set and a plurality of prediction probabilities corresponding to each material in the material candidate set; wherein the prediction probability is a prediction probability of a preset behavior of the target user on the material, and one prediction probability corresponds to one preset behavior; For each material in the material candidate set, weight the associated prediction probability using the weight corresponding to each preset behavior in the target fusion weight set, and add the weighted probability values ​​to obtain the recall score value of each material; Based on the recall score value of each material in the material candidate set, all materials in the material candidate set are screened, and materials are recommended to the target user based on the screened and retained materials.

10. A material recommendation device, characterized in that: include: An acquisition module is configured to acquire the recommendation accuracy corresponding to each initial fusion weight set after recommending materials to a user group based on multiple initial fusion weight sets; wherein the recommendation accuracy is determined based on the preset behaviors of the users in the group on the recommended materials; each weight in a set of initial fusion weight sets is associated with a preset behavior, and the weights in the initial fusion weight sets of different groups are different; the initial fusion weight set is used to perform weighted fusion on the predicted probability of the users in the group on the materials to sort the materials and make recommendations; A determination module configured to determine a set of target fusion weight sets based on the recommendation accuracy corresponding to each set of initial fusion weight sets; The recommendation module is configured to recommend materials to target users outside the user group based on the target fusion weight set; wherein the target fusion weight set is used to perform weighted fusion on the predicted probability of the target user's occurrence of each preset behavior on the material to sort the materials and then make recommendations.

Images