A method, device, electronic device and storage medium for recommending target information

By using proportional coefficients to represent the degree of influence of dimension information on recommendations in information recommendations and determining the target weight subset based on the difference, the problem of mutual constraints between multiple reference factors is solved, and the accuracy and balance of recommendations are improved.

CN113590941BActive Publication Date: 2024-11-29BEIJING SHAREIT INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202110807187.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-07-16
Publication Date
2024-11-29
Estimated Expiration
2041-07-16

AI Technical Summary

Technical Problem

In information recommendation, multiple reference factors restrict and influence each other, resulting in the recommendation result being greatly affected by one or several reference factors, which is prone to falling into local optimization, reducing the accuracy and user experience of recommendations.

Method used

By obtaining the reference parameter set and the alternative parameter set of dimension information of the alternative information to be recommended, the proportional coefficients represent the degree of influence of the dimension information on the recommendation, and according to the difference between the first sum and the second sum, the value from The alternative map value subset determines a subset of target map values, and then determines a subset of target weights, thereby selecting the recommended target information.

Benefits of technology

It improves the recommendation accuracy of target information, reduces the mutual influence and constraints between the information weights in multiple dimensions, and improves the balance and accuracy of the recommendation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113590941B_ABST
    Figure CN113590941B_ABST
Patent Text Reader

Abstract

The present disclosure relates to a method for recommending target information, comprising: obtaining a reference parameter set of at least one dimensional information of alternative information to be recommended and multiple alternative parameter sets of at least one dimensional information; the reference parameter set comprises: a reference weight subset of at least one dimensional information and a reference mapping value subset corresponding to the reference weight subset; the alternative parameter set comprises: an alternative weight subset of at least one dimensional information and an alternative mapping value subset corresponding to each alternative weight subset; determining a target mapping value subset according to the difference between the sum of elements in any alternative mapping value subset and the sum of elements in the reference mapping value subset; each element in the alternative mapping value subset has a proportional coefficient, and the proportional coefficient is used to represent the degree of influence of the dimensional information on the target information determined to be recommended from the alternative information; determining a target weight subset according to the target mapping value subset; and selecting recommended target information from the alternative information according to the target weight subset.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of information processing, and in particular to a method, device, electronic device and storage medium for recommending target information. Background Art

[0002] With the development of science and technology, many new application technologies and application scenarios have emerged. These application technologies can be applied to relevant application scenarios. In different application scenarios, information corresponding to the application scenario will be generated. By processing this information, this information can be better applied in relevant application scenarios and achieve better application results.

[0003] For example, when recommending information, after the relevant information is processed through information processing technology, more suitable information can be recommended to the user. Summary of the invention

[0004] The present disclosure provides a method, device, electronic device and storage medium for recommending target information.

[0005] In a first aspect of an embodiment of the present disclosure, a method for recommending target information is provided, comprising: obtaining a reference parameter set of at least one dimension information of candidate information to be recommended and a plurality of candidate parameter sets of the at least one dimension information; wherein the reference parameter set comprises: a reference weight subset of at least one dimension information and a reference mapping value subset corresponding to the reference weight subset; the candidate parameter set comprises: a candidate weight subset of the at least one dimension information and a candidate mapping value subset corresponding to each of the candidate weight subsets; determining a target mapping value subset from the candidate mapping value subset according to a difference between a first sum value and a second sum value; wherein the first sum value is: the sum of elements in any one of the candidate mapping value subsets; the second sum value is the sum of elements in the reference mapping value subset, each element in the candidate mapping value subset has a proportional coefficient, and the proportional coefficient is used to represent the degree of influence of the dimension information on the target information determined from the candidate information for recommendation; determining a target weight subset according to the target mapping value subset; the target weight subset comprises: a target weight affecting at least one dimension information of the candidate information; and selecting the target information to be recommended from the candidate information according to the target weight subset.

[0006] In one embodiment, the proportionality coefficient is a preset value; determining the target mapping value subset from the alternative mapping value subset based on the difference between the first sum value and the second sum value includes: determining the first sum value based on the proportionality coefficient of each element in the alternative mapping value subset; based on the difference between the first sum value and the second sum value, determining the alternative mapping value subset with the largest difference, or the alternative mapping value subset with a difference greater than a preset difference, as the target mapping value subset.

[0007] In one embodiment, the method further comprises: updating the scaling factor according to the candidate mapping value subset.

[0008] In one embodiment, updating the proportional coefficient according to the subset of alternative mapping values ​​includes: determining, according to the multiple alternative mapping values ​​corresponding to each dimensional information, the changing trend of the multiple alternative mapping values ​​respectively corresponding to each dimensional information; adjusting the proportional coefficient according to each changing trend until each changing trend reaches the target range; wherein the changing trends corresponding to different dimensional information respectively have the corresponding target range.

[0009] In one embodiment, the target range is: a threshold range determined according to the proportion of the dimension information when the target information is selected from the candidate information.

[0010] In one embodiment, the acquisition of multiple alternative parameter sets of at least one dimensional information includes: presetting T initialization parameter sets, the initialization parameter set including: an initialization weight subset of the at least one dimensional information and an initialization mapping value subset corresponding to the initialization weight subset; determining the T+1th parameter set according to the third sum of the elements in the initialization weight subset in each initialization parameter set, and the fourth sum of the elements in each initialization mapping value subset; wherein the T+1th parameter set includes: the T+1th weight subset and the T+1th mapping value subset, T is a positive integer greater than 1; determining the T+n+1th parameter set according to T+n parameter sets; n is a positive integer, and the sum of the elements in the T+nth mapping value subset is a local maximum; determining the T+n+1th parameter set as the alternative parameter set.

[0011] In one embodiment, the preset T initialization parameter sets include: determining T initialization parameter sets matching the alternative information based on the historical parameter set of the alternative information; wherein the historical parameter set includes: a historical weight subset of at least one dimension information and a historical mapping value subset corresponding to the historical weight subset; or, randomly generating T initialization parameter sets based on the alternative information to be recommended.

[0012] In one embodiment, determining the target weight subset according to the target mapping value subset includes: determining the candidate weight subset corresponding to the target mapping value subset as the target weight subset.

[0013] In one embodiment, the alternative information to be recommended includes at least one of the following: video; public account; article; the dimensional information includes at least one of the following: download volume, collection volume, like volume, number of users following the author of the alternative information, average browsing time, number of comments and number of reposts.

[0014] According to a second aspect of an embodiment of the present disclosure, a device for recommending target information is provided, including: an acquisition module, used to acquire a reference parameter set of at least one dimension information of candidate information to be recommended and a plurality of candidate parameter sets of the at least one dimension information; wherein the reference parameter set includes: a reference weight subset of at least one dimension information and a reference mapping value subset corresponding to the reference weight subset; the candidate parameter set includes: a candidate weight subset of the at least one dimension information and a candidate mapping value subset corresponding to each of the candidate weight subsets; a target mapping value subset determination module, used to determine, from the candidate mapping value subsets, a target mapping value subset determination module according to a difference between a first sum value and a second sum value; A target mapping value subset; wherein the first sum value is: the sum of the elements in any one of the alternative mapping value subsets; the second sum value is the sum of the elements in the reference mapping value subset, and each element in the alternative mapping value subset has a proportional coefficient, and the proportional coefficient is used to represent the degree of influence of the dimensional information on the recommended target information determined from the alternative information; a target weight subset determination module, used to determine a target weight subset based on the target mapping value subset; the target weight subset includes: a target weight of at least one dimensional information that affects the alternative information; a recommendation module, used to select the recommended target information from the alternative information based on the target weight.

[0015] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: a processor and a memory for storing executable instructions that can be run on the processor, wherein: when the processor is used to run the executable instructions, the executable instructions execute the method described in any one of the above embodiments.

[0016] According to a fourth aspect of the embodiments of the present disclosure, a non-temporary computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the method of any of the above embodiments is implemented.

[0017] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:

[0018] When determining the first sum value, the embodiment of the present disclosure adds a proportional coefficient to each element in the alternative mapping value subset, that is, before each alternative mapping value. Through the method in this embodiment, the recommendation accuracy of the target information can be improved, and the mutual influence and restriction between the weights corresponding to multiple dimensional information can be reduced. The influence of the weight corresponding to one part of the dimensional information on the weight corresponding to another part of the dimensional information is reduced, and the degree of sacrifice of the weight corresponding to another part of the dimensional information on the accuracy of determining the target information is reduced, thereby improving the effect of the weight corresponding to another part of the dimensional information on the accuracy of determining the target information, thereby improving the balance of the weights corresponding to multiple dimensional information when determining the target information, and facilitating improving the accuracy of determining the target information.

[0019] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0021] Figure 1 is a flowchart of a method for recommending target information according to an exemplary embodiment;

[0022] Figure 2 is a schematic diagram of a process of determining a target mapping value subset according to an exemplary embodiment;

[0023] Figure 3 is a schematic diagram of a process of obtaining a candidate parameter set according to an exemplary embodiment;

[0024] Figure 4 is a schematic diagram of a flow chart of updating a proportional coefficient according to an exemplary embodiment;

[0025] Figure 5 is a schematic diagram showing a change trend of candidate mapping values ​​in an application scenario according to an exemplary embodiment;

[0026] Figure 6 is a schematic diagram showing a change trend of candidate mapping values ​​in an application scenario according to an exemplary embodiment;

[0027] Figure 7 is a flowchart of another method for recommending target information according to an exemplary embodiment;

[0028] Figure 8 The invention is a block diagram of a terminal device according to an exemplary embodiment. DETAILED DESCRIPTION

[0029] 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 embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices consistent with some aspects of the present disclosure as detailed in the appended claims.

[0030] Usually, when recommending information, there are many reference factors, that is, many factors that affect the information recommendation results. When recommending information, multiple reference factors will be combined to recommend the final information. However, in general, multiple reference factors will restrict and influence each other. When recommending information, it will be affected by one or several reference factors, which may cause the final recommended information to be greatly affected by one or several reference factors, and it is easy to fall into the local optimum.

[0031] In this case, if there are reference factors among other reference factors that play a key role in the final recommended information, the degree of determination of these key reference factors on the final recommended information will be reduced, resulting in the determination of other reference information on the final recommended information being far less than the determination of one or several reference factors, and thus failing to balance the relationship between multiple reference factors based on multiple reference factors to achieve the overall optimal effect. As a result, the accuracy of the final recommended information is reduced, and the most suitable or most matching recommended information cannot be recommended, thereby reducing the recommendation effect and user experience.

[0032] refer to Figure 1 , is a flow chart of a method for recommending target information provided by an embodiment of the present disclosure, the data processing method comprising the following steps:

[0033] Step S100, obtaining a reference parameter set of at least one dimensional information of the alternative information to be recommended and multiple alternative parameter sets of the at least one dimensional information; wherein the reference parameter set includes: a reference weight subset of at least one dimensional information and a reference mapping value subset corresponding to the reference weight subset; the alternative parameter set includes: an alternative weight subset of the at least one dimensional information and a alternative mapping value subset corresponding to each of the alternative weight subsets.

[0034] Step S200, determining a target mapping value subset from the alternative mapping value subsets according to a difference between a first sum value and a second sum value; wherein the first sum value is: the sum of the elements in any one of the alternative mapping value subsets; the second sum value is the sum of the elements in the reference mapping value subset, and each element in the alternative mapping value subset has a proportional coefficient, and the proportional coefficient is used to represent the degree of influence of the dimensional information on determining the recommended target information from the alternative information.

[0035] Step S300, determining a target weight subset according to the target mapping value subset; the target weight subset includes: a target weight of at least one dimensional information affecting the candidate information.

[0036] Step S400: selecting the recommended target information from the candidate information according to the target weight subset.

[0037] The method can be executed at least in a mobile terminal, that is, the execution subject of the method can at least include a mobile terminal. Mobile terminals can include mobile phones, tablet computers, vehicle-mounted central control devices, wearable devices, smart devices, etc., and smart devices can include smart office equipment and smart home equipment, etc.

[0038] The method of this embodiment can be applied to various recommendation scenarios, and can be applied to recommendation scenarios where the recommendation result is determined by one dimension information, or can be applied to recommendation scenarios where the recommendation result is determined by multiple dimension information. In these recommendation scenarios, the candidate information to be recommended has multiple dimension information that can be referenced, and the final recommendation information can be determined based on these dimension information. Different dimension information has different proportions in determining whether a certain candidate information is the final recommendation information, so different dimension information has different reference values ​​for determining the final recommendation information.

[0039] Of course, this method can also be used in other non-recommended application scenarios. For example, some candidate targets have multiple dimensional information. As long as the scenario is to determine whether the candidate target is the final target based on these different dimensional information, it is within the protection scope of this embodiment.

[0040] This method can be used to determine the optimal target weight corresponding to each dimensional information, thereby determining the recommended target information based on the optimal target weight.

[0041] In this embodiment, the candidate information to be recommended includes at least one of the following: video, public account, article and / or product description information, etc. The products here include but are not limited to: physical goods and / or services. Of course, other types of content can also be included, which will not be listed here one by one. As long as the content includes multiple dimensional information, it can be used as candidate information to be recommended and is within the protection scope of this embodiment.

[0042] In one embodiment, the dimension information of the candidate information to be recommended includes at least one of the following: download volume, collection volume, like volume, number of users following the author of the candidate information, average browsing time, number of comments, and number of reposts, etc. Of course, different candidate information may have at least one dimension information matching the candidate information. In this embodiment, the dimension information of the candidate information to be recommended may be generated based on the historical operations of the user who has received the information on the candidate information.

[0043] In another embodiment, the dimensional information of the alternative information to be recommended includes at least one of the following: the difference between the amount of information of the information to be recommended and the amount of information preferred by the recommended user, the similarity between the information content of the information to be recommended and the content preferred by the recommended user, the strength of recommendation of the information to be recommended, etc.

[0044] In specific application scenarios, for example, when watching or browsing videos, public accounts, and articles, the videos, public accounts, and articles may be downloaded, collected, liked, commented on, and / or forwarded, and the authors of these target messages such as videos, public accounts, and articles may also be followed, etc. Different dimensional information such as the number of downloads, the number of collections, the number of likes, the number of users who follow the author of the candidate information, the average browsing time, the number of comments, and the number of forwardings can reflect the user's preference for the videos, public accounts, and articles that have been watched or browsed.

[0045] Different dimensional information has different weights in the candidate information to be recommended. The weights of these dimensional information in the candidate information can determine whether the candidate information is the target information, that is, the target information can be determined from the candidate information based on the weights of these dimensional information in the candidate information. When recommending subsequent videos, public accounts or articles to users, the target information to be recommended can be determined from the candidate information to be recommended based on the download, collection, like, comment and / or forwarding information of the video, public account, article or other form of content to be recommended.

[0046] Therefore, the weights of information of different dimensions of the candidate information to be recommended play a key role in determining the target information. This embodiment is a description of determining the target information from the candidate information according to the weights of information of different dimensions of the candidate information.

[0047] For step S100, a reference parameter set of at least one dimension of candidate information to be recommended is obtained. The reference parameter set is used as a basic parameter set for determining target information. The target information is determined by combining information in the candidate parameter set with information in the reference parameter set.

[0048] The reference parameter set includes: a reference weight subset of at least one dimensional information and a reference mapping value subset corresponding to the reference weight subset. The reference weight subset is a set of reference weights of at least one dimensional information, that is, the reference weight subset includes the reference weight of at least one dimensional information. The reference mapping value subset is a set of mapping values ​​corresponding to the reference weight of at least one dimensional information, that is, the reference mapping value subset includes the mapping value of the reference weight of at least one dimensional information. The reference parameter set may be a predetermined set.

[0049] The reference weight subset can be expressed as Xbase={xbase1, xbase2, xbase3, ..., xbasen}, that is, the candidate information to be recommended has n dimensional information, each dimensional information has a corresponding reference weight, and n can be determined according to the candidate information to be recommended. In this embodiment, each weight in the reference weight subset is referred to as an element of the reference weight subset.

[0050] For example, the candidate information to be recommended is a video, which includes seven dimensional information: download volume, favorite volume, like volume, number of users following the author of the candidate information, average browsing time, number of comments, and number of reposts. Each dimensional information has its own reference weight. For example, the reference weight subset is Xbase = {xbase1, xbase2, xbase3, xbase4, xbase5, xbase6, xbase7}. The weight value corresponding to each dimensional information in the reference weight subset, that is, the value of each element, can be predetermined, that is, a reference value that matches the candidate information to be recommended, or a randomly initialized value.

[0051] The reference mapping value subset can be expressed as Ybase={ybase1, ybase2, ybase3..., ybasen}, that is, the mapping value corresponding to the weight of the n dimensional information. Each mapping value in the reference mapping value subset is called an element of the reference mapping value subset, and the number of elements in the subset is equal to the number of elements in the reference weight subset. There is a corresponding mapping relationship between the reference mapping value and the reference weight, which can be determined according to actual business needs, or it can be a preset mapping relationship, etc.

[0052] For example, by using a preset model or function relationship, the corresponding mapping value can be determined according to the reference weight, such as the mapping value ybase1 of xbase1 can be determined according to the preset model or function relationship, the mapping value ybase2 of xbase2 can be determined, and each reference weight corresponds to a reference mapping value, so that the mapping value of each element in the reference weight subset can be determined, thereby determining the reference mapping value subset. For example, Ybase = {ybase1, ybase2, ybase3, ybase4, ybase5, ybase6, ybase7}.

[0053] In step S100, it also includes: obtaining multiple candidate parameter sets of at least one dimensional information, that is, obtaining multiple candidate reference sets, each candidate parameter set includes: an alternative weight subset of at least one dimensional information and an alternative mapping value subset corresponding to each alternative weight subset.

[0054] The number of candidate weights included in the candidate weight subset is the same as the number of reference weights of the dimension information included in the reference weight subset, and each candidate weight in the candidate weight subset is also used as each element of the candidate weight subset. That is, the number of elements included in the candidate weight subset is the same as the number of elements in the reference weight subset.

[0055] In this embodiment, the number of candidate weight subsets can be determined according to actual needs, such as 3 or 5. The candidate weight subsets can be expressed as X1, X2, X3...Xt, that is, including t candidate weight subsets. Each candidate weight subset includes the same number of elements as the number of elements in the reference mapping value subset, for example, 7.

[0056] For example, obtain 5 alternative weight subsets X1, X2, X3, X4 and X5, X1 = {x11, x12, x13, x14, x15, x16, x17}, X2 = {x21, x22, x23, x24, x25, x26, x27}, X3 = {x31, x32, x33, x34, x35, x36, x37}, X4 = {x41, x42, x43, x44, x45, x46, x47}, X5 = {x51, x52, x53, x54, x55, x56, x57}.

[0057] For any two alternative weight subsets, if the alternative weight values ​​corresponding to at least one dimension information in the alternative weight subsets are different, then the two alternative weight subsets are different. For the same dimension information, the weights included in any two different alternative weight subsets may be the same or different. In this embodiment, for the same dimension information, the weights included in any two different alternative weight subsets are different as an example. When different alternative weight subsets are used as target weight subsets, different target information is determined. By determining the target weight subset from these different alternative weight subsets, the optimal target weight subset can be determined, thereby determining the target information with the highest recommendation accuracy, thereby improving the accuracy of the recommendation.

[0058] Each candidate weight in the candidate weight subset may be predetermined, or may be a candidate weight matched from a database and matched with the candidate information to be recommended. Of course, it may also be determined by other methods, which will not be limited here.

[0059] The alternative mapping value subsets corresponding to the alternative weight subsets can be expressed as Y1, Y2, Y3...Yt, and the number of alternative mapping value subsets is the same as the number of alternative weight subsets. For example, if there are 5 alternative weight subsets, then there are 5 alternative mapping value subsets, and there is a mapping relationship between Y1 and X1, a mapping relationship between Y2 and X2, ..., a mapping relationship between Y5 and X5. The number of mapping values ​​included in each alternative mapping value subset is the same as the number of weights in the alternative weight subset, that is, the number of elements in the alternative mapping value subset is the same as the number of elements in the alternative weight subset.

[0060] For example, Y1 = {y11, y12, y13, y14, y15, y16, y17}, Y2 = {y21, y22, y23, y24, y25, y26, y27}, Y3 = {y31, y32, y33, y34, y35, y36, y37}, and so on, and so on for Y4 and Y5. There is a mapping relationship between y11 and x11, a mapping relationship between y12 and x12, a mapping relationship between y13 and x13, ... a mapping relationship between y17 and x17. There is a mapping relationship between y21 and x21, a mapping relationship between y22 and x22, a mapping relationship between y23 and x23, ... a mapping relationship between y27 and x27, and so on.

[0061] In this embodiment, the candidate mapping value subset may be determined based on the candidate weight subset, and may also be determined based on the candidate weight subset using a preset model or functional relationship. The preset model or functional relationship here may be the same as the preset model or functional relationship used to determine the reference mapping value subset based on the reference weight subset.

[0062] The candidate weight subsets and candidate mapping value subsets may be used to further target weight subsets to determine target information.

[0063] As for step S200, a target mapping value subset can be determined through this step.

[0064] The first sum value is: the sum of the elements in any candidate mapping value subset. When determining the first sum value, each element in each candidate mapping value subset has a proportional coefficient, and the proportional coefficient is used to indicate the influence of the dimensional information corresponding to the element on the target information recommended from the candidate information. The first sum value is determined based on each element in the candidate mapping value subset and the proportional coefficient of each element.

[0065] For example, the candidate mapping value subset Y1 = {y11, y12, y13, y14, y15, y16, y17}, when determining the first sum value, each element in the candidate mapping value subset Y1 has a proportional coefficient, such as Y1 = {a*y11, b*y12, c*y13, d*y14, e*y15, f*y16, g*y17}, a, b, c, d, e, f and g are the proportional coefficients. The first sum value corresponding to Y1 can be expressed as sum(a*y11, b*y12, c*y13, d*y14, e*y15, f*y16, g*y17). Similarly, each element in other alternative mapping value subsets also has a proportional coefficient, such as Y2 = {a*y21, b*y22, c*y23, d*y24, e*y25, f*y26, g*y27}, and the first sum corresponding to Y2 is sum(a*y21, b*y22, c*y23, d*y24, e*y25, f*y26, g*y27).

[0066] These proportional coefficients are used to represent the dimensional information corresponding to the weight that has a mapping relationship with the mapping value, and the weight it occupies when determining the target information from the alternative information to be recommended. The larger the proportional coefficient, the greater the weight and the greater the influence.

[0067] The second sum is: the sum of the elements in the reference mapping value subset. When determining the first sum, simply add the elements in the reference mapping value subset. For example, sum(ybase1, ybase2, ybase3, ybase4, ybase5, ybase6, ybase7).

[0068] Then, according to the difference between the first sum and the second sum, the target mapping value subset is determined from the candidate mapping value subsets. The specific determination method is not limited, for example, reference can be made to the subsequent embodiments, for example, the candidate mapping value subset with the largest difference can be used as the target mapping value subset, etc.

[0069] For example, based on the difference between the first sum value sum(a*y51, b*y52, c*y53, d*y54, e*y55, f*y56, g*y57) corresponding to the alternative mapping value subset Y5={y51, y52, y53, y54, y55, y56, y57} and the second sum value sum(ybase1, ybase2, ybase3, ybase4, ybase5, ybase6, ybase7), the target mapping value subset is determined to be Y5={y51, y52, y53, y54, y55, y56, y57}.

[0070] For step S300,

[0071] After determining the target mapping value subset, determine the target weight subset according to the determined target mapping value subset. Since the target mapping value subset is selected from the candidate mapping value subset, each candidate mapping value subset corresponds to an alternative weight subset with a mapping relationship, so the alternative weight subset with a mapping relationship corresponding to the target mapping value subset is used as the target weight subset. That is, the alternative weight subset corresponding to the target mapping value subset is determined as the target weight subset.

[0072] The target weight subset includes: target weights that affect at least one dimension of candidate information. The target weight subset includes target weights corresponding to at least one dimension of information, which can affect the determination of target information from candidate information, that is, affect whether a certain candidate information is target information.

[0073] For example, the target mapping value subset is Y5 = {y51, y52, y53, y54, y55, y56, y57}. Since the alternative weight subset that has a mapping relationship with Y5 is X5 = {x51, x52, x53, x54, x55, x56, x57}, X5 is the target weight subset.

[0074] When determining the first sum value, by adding a proportional coefficient to each element in the alternative mapping value subset, that is, before each alternative mapping value, on the one hand, the ability of the Bayesian algorithm to solve the waste of computing time and space is retained, and on the other hand, the problem that the Bayesian algorithm is prone to fall into local optimality and the weights of multiple dimensional information restrict and influence each other, resulting in inaccurate target information, is reduced, thereby improving the accuracy of determining the target information.

[0075] For step S400,

[0076] After determining the target weight subset, the recommended target information can be selected from the candidate information according to the target weight subset. Since the target weight subset includes the weight of at least one dimension information, the target information can be determined from the candidate information to be recommended according to the weight of at least one dimension information.

[0077] For example, the target weight subset is X5 = {x51, x52, x53, x54, x55, x56, x57}, then the target video is determined from the candidate videos to be recommended according to each element in the target weight subset, that is, according to the weights corresponding to the seven dimensional information.

[0078] Through the method in this embodiment, the recommendation accuracy of the target information can be improved, the mutual influence and restriction between the weights corresponding to multiple dimensional information can be reduced, the influence of the weights corresponding to one part of the dimensional information on the weights corresponding to another part of the dimensional information can be reduced, and the degree of sacrifice of the weights corresponding to another part of the dimensional information on the accuracy of determining the target information can be reduced, thereby improving the effect of the weights corresponding to another part of the dimensional information on the accuracy of determining the target information, thereby improving the balance of the weights corresponding to multiple dimensional information when determining the target information, and facilitating improving the accuracy of determining the coding information.

[0079] In another embodiment, when determining the first sum value, the proportional coefficient of each element in the candidate mapping value subset may be a preset value, which may be fixed. The specific value of the proportional coefficient is determined according to the actual business needs, that is, according to the candidate information to be recommended. For example, according to the recommendation needs of the candidate information to be recommended, where the influence of a part of the dimensional information on the determination of the target information is greater than the influence of another part of the dimensional information on the determination of the target information, the weight corresponding to a part of the dimensional information is set to a proportional coefficient of the mapping value corresponding to a part of the dimensional information, which is greater than the proportional coefficient of the mapping value corresponding to another part of the dimensional information.

[0080] refer to Figure 2 , which is a flow chart of determining a target mapping value subset, step S200, determining a target mapping value subset from the candidate mapping value subsets according to a difference between the first sum value and the second sum value, comprising:

[0081] Step S201: determining a first sum value according to a proportional coefficient of each element in the candidate mapping value subset.

[0082] The first sum value is: the sum of the elements in any candidate mapping value subset. When determining the first sum value, each element in each candidate mapping value subset has a proportional coefficient, and the proportional coefficient is used to indicate the influence of the dimensional information corresponding to the element on the target information recommended from the candidate information. The first sum value is determined based on each element in the candidate mapping value subset and the proportional coefficient of each element.

[0083] For example, the candidate mapping value subset Y1 = {y11, y12, y13, y14, y15, y16, y17}, when determining the first sum value, each element in the candidate mapping value subset Y1 has a proportional coefficient, such as Y1 = {a*y11, b*y12, c*y13, d*y14, e*y15, f*y16, g*y17}, a, b, c, d, e, f and g are the proportional coefficients. The first sum value corresponding to Y1 can be expressed as sum(a*y11, b*y12, c*y13, d*y14, e*y15, f*y16, g*y17). Similarly, each element in other alternative mapping value subsets also has a proportional coefficient, such as Y2 = {a*y21, b*y22, c*y23, d*y24, e*y25, f*y26, g*y27}, and the first sum corresponding to Y2 is sum(a*y21, b*y22, c*y23, d*y24, e*y25, f*y26, g*y27).

[0084] These proportional coefficients are used to represent the dimensional information corresponding to the weight that has a mapping relationship with the mapping value, and the weight it occupies when determining the target information from the alternative information to be recommended. The larger the proportional coefficient, the greater the weight and the greater the influence.

[0085] Step S202: According to the difference between the first sum and the second sum, a candidate mapping value subset with the largest difference, or a candidate mapping value subset with a difference greater than a preset difference, is determined as a target mapping value subset.

[0086] After determining the difference between the first sum and the second sum, since the selection of the optimal target mapping subset is determined by maximizing the difference, in this embodiment, the alternative mapping value subset with the largest difference is determined as the target mapping value subset. In this way, the optimal mapping value subset among all the alternative mapping value subsets can be determined, and the optimal mapping value subset is used as the target mapping value subset.

[0087] A subset of alternative mapping values ​​whose difference is greater than a preset difference can also be used as a target mapping value subset. The preset difference can be a preset difference determined based on the alternative information and matching the alternative information. When the difference between the first sum and the second sum is greater than the preset difference, an optimal mapping value subset can be determined among the alternative mapping value subsets, and the optimal mapping value subset can be determined as the target mapping value subset.

[0088] In this embodiment, the target information may be determined according to the target weight subset corresponding to the optimal mapping value subset.

[0089] In another embodiment, reference Figure 3 , is a flow chart of obtaining a candidate parameter set, wherein the step S100 of obtaining a plurality of candidate parameter sets of at least one dimension information includes:

[0090] Step S101 , preset T initialization parameter sets, the initialization parameter sets including: an initialization weight subset of at least one dimension information and an initialization mapping value subset corresponding to the initialization weight subset.

[0091] In this step, T initialization parameter sets can be preset in a preset manner, each of which includes an initialization weight subset of at least one dimension of information and an initialization mapping value subset corresponding to the initialization weight subset, that is, there is a mapping relationship between the initialization weight subset and the initialization mapping value subset.

[0092] Each initialization weight subset includes weights of at least one dimension of information, and the number of initialization weights in the subset is the same as the number of weights in the reference weight subset. Each initialization mapping value subset includes mapping values ​​corresponding to weights of at least one dimension of information, and the number of mapping values ​​in the subset is the same as the number of initialization weights in the initialization weight subset.

[0093] The number of T can be set according to actual needs, for example, it can be a positive integer such as 3 or 5.

[0094] For example, the weight subsets are initialized as X6, X7, X8, X9 and X10, and the mapping value subsets are initialized as Y6 to Y10. X6 and Y6 have a mapping relationship, X7 and Y7 have a mapping relationship, and so on. X6 = {x61, x62, x63, x64, x65, x66, x67}, X7 = {x71, x72, x73, x74, x75, x76, x77}, and X3, X4 and X5 can be deduced by analogy. Y6 = {y61, y62, y63, y64, y65, y66, y67}, Y7 = {y71, y72, y73, y74, y75, y76, y77}, and so on, and Y8, Y9 and Y10 can be deduced by analogy. There is a mapping relationship between Y61 and x61, a mapping relationship between y62 and x62, a mapping relationship between y63 and x73, ... a mapping relationship between y67 and x67. There is a mapping relationship between Y71 and x71, a mapping relationship between y72 and x72, a mapping relationship between y73 and x73, ... a mapping relationship between y77 and x77, and so on.

[0095] Step S102, determining the T+1th parameter set according to the third sum of the elements in the initialized weight subsets in each initialized parameter set and the fourth sum of the elements in each initialized mapping value subset; the T+1th parameter set includes: the T+1th weight subset and the T+1th mapping value subset, where T is a positive integer greater than 1.

[0096] In this step, according to each element in each initialization weight subset, i.e., the initialization weight, the sum of all the initialization weights in each initialization weight subset is determined as the third sum value. According to each element in each initialization mapping value subset, i.e., the initialization mapping value, the sum of all the initialization mapping values ​​in each initialization mapping value subset is determined as the fourth sum value.

[0097] Since there is a mapping relationship between the initialization weight subset and the initialization mapping value subset, each initialization weight subset corresponds to an initialization mapping value subset, and the initialization weight subset and the initialization weight subset with a mapping relationship can be used as a set of parameters. Different initialization parameter sets include an initialization weight subset and an initialization mapping value subset, so the third sum value corresponding to an initialization weight subset and the fourth sum value corresponding to an initialization mapping value subset can constitute the coordinates of an initialization parameter in the coordinate system. The third sum value corresponds to the X-axis, and the fourth sum value corresponds to the Y-axis.

[0098] According to the development trend of the coordinate positions of the first initial parameter set to the Tth initial parameter set in the known T initialization parameter sets, the coordinates of the T+1th parameter set are predicted. The T+1th parameter set includes the T+1th weight subset and the T+1th mapping value subset, and there is a mapping relationship between the T+1th mapping value subset and the T+1th weight subset.

[0099] The specific prediction method can be determined based on the Gaussian distribution and the conditional probability distribution of each mapping value subset. For example:

[0100] By adjusting x in X, the optimal target information can be determined from the candidate information. The target information can be evaluated by Y corresponding to X, that is, the mapping value subset, to see whether it is optimal, such as when Y reaches an extreme value (maximum or minimum value within a certain interval). The target information recommendation method can be implemented through a recommendation model. In the recommendation scenario, auc and gauc are generally used to evaluate the accuracy of the target information recommendation model. Through the equation f of auc and gauc, f(X) = Y, the recommendation accuracy of the model can be evaluated. Therefore, when f is determined, it is necessary to determine a set of X that maximizes Y.

[0101] By constructing a surrogate function of the objective function f, the surrogate function can obtain the optimal solution by evaluating the score of the surrogate function on X. Assuming that X and Y obey Gaussian distribution, the posterior probability distribution of (X, Y) can be fitted through Gaussian distribution, then we can estimate the value of Y based on X. Each time Y is obtained based on X, the estimate of the posterior probability distribution can be corrected. Of course, the more samples are taken, the closer the distribution is to the truth, but the computational cost is huge. If there is any infinite sampling, the time consumption is close to the time consumption of traversing all X to find Y.

[0102] Therefore, in this embodiment, a sampling limit is set, which is called the harvest function. , which can be max(mean+var). mean represents the mean. The higher the mean, the more likely the sampling of the area where the global optimal solution is likely to appear. var represents the variance. The higher the variance, the more likely the global optimal solution is in the unsampled area, that is, the sampling points are obtained in the unsampled area. If the global optimal solution is pursued blindly, the optimal X will linger around the points around X, which is not conducive to the rapid fitting of the real distribution. Therefore, it is necessary to add the variance to expand the variance and find some unsampled areas. According to the harvest function max(mean+var) plus the constraint relationship between mean and var, the optimal X under the current distribution can be determined, and Y can be obtained through f. (X, Y) is the predicted information, which is used to update the data distribution and continue to iterate.

[0103] Similarly, through the function Y=f(X), y and x also satisfy the function, which is equivalent to optimizing y1+y2+y3…yn=f(x1)+f(x2)+f(x3)…f(xn). The function f can be used to determine the x included in X. After x is determined, y can be determined according to the function f. The specific process will not be described in detail here, please refer to the Bayesian optimization algorithm.

[0104] The sum of all weights in the T+1th weight subset of the T+1th parameter set is a local maximum, and the sum of all mapping values ​​in the T+1th mapping value subset is a local maximum.

[0105] Step S103, determining the T+n+1th parameter set according to the T+n parameter sets; n is a positive integer, and the sum of the elements in the T+nth mapping value subset is a local maximum.

[0106] The T+1th parameter set can be determined based on the known T initialization parameter sets, that is, based on the third sum corresponding to the initialization weight subset and the fourth sum corresponding to the initialization mapping value subset in the known T initialization parameter sets, the positions of the third sum and the fourth sum corresponding to the first initialization parameter set to the Tth initialization parameter set in the coordinate system can be determined. According to the Gaussian distribution, the positions of the sum corresponding to the weight subset and the sum corresponding to the mapping value subset in the T+1th parameter set in the coordinate system can be predicted, thereby determining the sum corresponding to the weight subset and the sum corresponding to the mapping value subset in the T+1th parameter set, and then determining the weight subset and the mapping value subset corresponding to the T+1th parameter set.

[0107] According to the third sum values ​​and the fourth sum values ​​corresponding to the T initialization parameter sets and the T+1th parameter set respectively, the sum value of the weight subset and the sum value of the mapping value subset in the T+2th parameter set are determined, thereby determining the weight subset and the mapping value subset corresponding to the T+2th parameter set.

[0108] By analogy, the sum value corresponding to the weight subset and the sum value corresponding to the mapping value subset in the T+nth parameter set can be determined based on the sum value corresponding to the weight subset and the sum value corresponding to the mapping value subset in the first T+n-1 parameter sets. The sum value corresponding to the weight subset and the sum value corresponding to the mapping value subset in the T+n+1th parameter set are determined based on the sum value corresponding to the weight subset and the sum value corresponding to the mapping value subset in the first T+n parameter sets, thereby determining the T+n+1th parameter set.

[0109] In this step, n can be determined according to actual needs, for example, it can be a positive integer from 1 to 10000. When determining the T+nth parameter set, the sum of the weights of the weight subsets in each parameter set is a local maximum, and the sum of the mapping values ​​in the mapping value subset is also a local maximum.

[0110] Step S104: determine T+n+1 parameter sets as candidate parameter sets.

[0111] After T+n+1 parameter sets are determined, the T+n+1 parameter sets are determined as candidate parameter sets, and a target parameter set is determined from the T+n+1 parameter sets, thereby determining a target mapping value subset.

[0112] In another embodiment, T+n parameter sets may also be determined as candidate parameter sets.

[0113] In another embodiment, step S101, presetting T initialization parameter sets includes:

[0114] According to the historical parameter set of the candidate information, T initialization parameter sets matching the candidate information are determined; wherein the historical parameter set includes: a historical weight subset of at least one dimension of information and a historical mapping value subset corresponding to the historical weight subset.

[0115] The number of elements in the historical weight subset and the historical mapping value subset included in the historical parameter set may be equal to the number of elements in the reference weight subset and the reference mapping value subset.

[0116] By determining T initialization parameter sets from the historical parameter set, the matching degree between the initialization parameter set and the candidate information can be improved, thereby improving the accuracy of the determined target information.

[0117] or,

[0118] According to the candidate information to be recommended, T initialization parameter sets are randomly generated.

[0119] In another embodiment, the method further comprises: updating the scaling factor according to the candidate mapping value subset.

[0120] In this embodiment, the proportional coefficient can be changed dynamically. By dynamically adjusting the proportional coefficient, the proportional coefficient of each element in the alternative mapping value subset can be dynamically updated according to the change of the sum of the mapping values ​​in the alternative mapping value subset, thereby balancing the influence of at least one dimensional information on the target information, reducing the mutual constraints between the target weights corresponding to multiple dimensional information, and thus reducing the low accuracy of the determined target information, thereby improving the accuracy of the determined target information.

[0121] In another embodiment, reference Figure 4 , which is a flow chart of updating the proportional coefficient. Updating the proportional coefficient according to the candidate mapping value subset includes:

[0122] Step S10: determining a change trend of the multiple candidate mapping values ​​corresponding to each dimensional information according to the multiple candidate mapping values ​​corresponding to each dimensional information.

[0123] After determining the candidate mapping value subsets in the multiple candidate parameter sets corresponding to at least one dimension information, each mapping value in the candidate mapping value subsets in each candidate parameter set can be determined, and each candidate mapping value subset includes the mapping value corresponding to each dimension information, so that the multiple mapping values ​​corresponding to each dimension information can be determined, and thus the change trend of each mapping value corresponding to each dimension information can be obtained. That is, each dimension information corresponds to a change trend of multiple mapping values ​​corresponding to the dimension information.

[0124] Step S20, according to each change trend, adjust the proportional coefficient until each change trend reaches the target range; wherein, the change trends corresponding to different dimensional information have corresponding target ranges. The target range is: a threshold range determined according to the proportion of the dimensional information when selecting the target information from the candidate information. That is, each change trend corresponds to a threshold range, and the threshold range is determined by the proportion of different dimensional information when selecting the target information from the candidate information.

[0125] After determining the change trend corresponding to each dimensional information, the proportional coefficient of the candidate mapping value corresponding to each dimensional information can be adjusted according to the change trend, and the different change trends can be adjusted to the corresponding target range. In this way, the influence of at least one dimensional information on the target information can be balanced, and the mutual constraints between the target weights corresponding to multiple dimensional information can be reduced, resulting in a low accuracy of the determined target information, thereby improving the accuracy of the determined target information.

[0126] For example, the seven dimensions of information, namely, the number of downloads, the number of favorites, the number of likes, the number of users who follow the author of the alternative information, the average browsing time, the number of comments, and the number of reposts, have corresponding threshold ranges of 0 to 0.02%, 0.1% to 0.3%, 0.2% to 0.5%, 0.3% to 0.7%, 0.1% to 0.6%, -0.9% to 0, and 0.05% to 0.08%, etc. When the change rate corresponding to the change trend reaches within the corresponding threshold range, the proportional coefficient can be adjusted. When the change rate is outside the corresponding threshold range, the corresponding change rate is adjusted to within the corresponding threshold range.

[0127] In another embodiment, it may also be: according to the change trend, when multiple alternative mapping values ​​corresponding to each dimensional information exceed the corresponding threshold range, the proportional coefficient of the alternative mapping value corresponding to each dimensional information is adjusted to adjust the alternative mapping value corresponding to the dimensional information within the threshold range. The threshold range in this embodiment is the numerical range of the alternative mapping value.

[0128] In another embodiment, reference Figure 5 , is a schematic diagram of a device for recommending target information, the device comprising:

[0129] Acquisition module 1, used to acquire a reference parameter set of at least one dimension of candidate information to be recommended and multiple candidate parameter sets of the at least one dimension of information; wherein the reference parameter set includes: a reference weight subset of at least one dimension of information and a reference mapping value subset corresponding to the reference weight subset; the candidate parameter set includes: a candidate weight subset of the at least one dimension of information and a candidate mapping value subset corresponding to each of the candidate weight subsets;

[0130] A target mapping value subset determination module 2 is used to determine a target mapping value subset from the candidate mapping value subset according to a difference between a first sum value and a second sum value; wherein the first sum value is: the sum of elements in any one of the candidate mapping value subsets; the second sum value is the sum of elements in the reference mapping value subset, and each element in the candidate mapping value subset has a proportional coefficient, and the proportional coefficient is used to indicate the influence of the dimensional information on the determination of the recommended target information from the candidate information;

[0131] A target weight subset determination module 3 is used to determine a target weight subset according to the target mapping value subset; the target weight subset includes: a target weight of at least one dimensional information affecting the candidate information;

[0132] The recommendation module 4 is used to select the recommended target information from the candidate information according to the target weight.

[0133] In another embodiment, the proportionality factor is a preset value;

[0134] The target mapping value subset determination module 2 comprises:

[0135] A first sum value determining unit, configured to determine the first sum value according to a proportional coefficient of each element in the candidate mapping value subset;

[0136] A target mapping value subset determination unit is used to determine the alternative mapping value subset with the largest difference, or the alternative mapping value subset with a difference greater than a preset difference, as the target mapping value subset based on the difference between the first sum value and the second sum value.

[0137] In another embodiment, the apparatus further comprises:

[0138] A proportionality coefficient updating module is used to update the proportionality coefficient according to the candidate mapping value subset.

[0139] The scale factor update module includes:

[0140] a change trend determining unit, configured to determine, according to the multiple candidate mapping values ​​corresponding to the respective dimensional information, the change trend of the multiple candidate mapping values ​​corresponding to the respective dimensional information;

[0141] The adjustment unit is used to adjust the proportional coefficient according to each of the change trends until each of the change trends reaches a target range; wherein the change trends corresponding to different dimensional information have corresponding target ranges respectively.

[0142] In another embodiment, the target range is: a threshold range determined according to the proportion of the dimensional information when the target information is selected from the candidate information.

[0143] Acquisition module 1 includes:

[0144] An initialization parameter set acquisition unit, configured to preset T initialization parameter sets, wherein the initialization parameter sets include: an initialization weight subset of the at least one dimension information and an initialization mapping value subset corresponding to the initialization weight subset;

[0145] A T+1th parameter set determination unit is used to determine a T+1th parameter set according to the third sum of each element in the initialization weight subset in each initialization parameter set and the fourth sum of each element in each initialization mapping value subset; wherein the T+1th parameter set includes: a T+1th weight subset and a T+1th mapping value subset, and T is a positive integer greater than 1;

[0146] A T+n+1th parameter set determination unit, used to determine a T+n+1th parameter set according to the T+n parameter sets; n is a positive integer, and the sum of the elements in the T+nth mapping value subset is a local maximum;

[0147] The candidate parameter set determining unit is used to determine the T+n+1 parameter sets as the candidate parameter sets.

[0148] Initialize parameter set acquisition unit, also used for:

[0149] Determine T initialization parameter sets matching the candidate information according to the historical parameter set of the candidate information; wherein the historical parameter set includes: a historical weight subset of at least one dimension information and a historical mapping value subset corresponding to the historical weight subset;

[0150] or,

[0151] According to the candidate information to be recommended, T initialization parameter sets are randomly generated.

[0152] In another embodiment, the target weight subset determination module 3 is further used for:

[0153] The candidate weight subset corresponding to the target mapping value subset is determined as the target weight subset.

[0154] In another embodiment, the candidate information to be recommended includes at least one of the following:

[0155] video;

[0156] Official account;

[0157] article;

[0158] The dimension information includes at least one of the following:

[0159] Number of downloads, number of favorites, number of likes, number of users following the author of the alternative information, average browsing time, number of comments, and number of reposts.

[0160] In another embodiment, an electronic device is provided, including:

[0161] A processor and a memory for storing executable instructions capable of running on the processor, wherein:

[0162] When the processor is used to run the executable instructions, the executable instructions execute the method described in any one of the above embodiments.

[0163] In another embodiment, a non-transitory computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the method described in any of the above embodiments is implemented.

[0164] In another embodiment, another method for recommending target information is provided.

[0165] With the rapid development of big data technology and machine learning technology, the application of machine learning has become more abundant and complex, resulting in a large number of parameter adjustments that require not only more machine learning background and business knowledge of algorithms, especially in the current rapid model updates and rapid product iterations. The demand for automated parameter adjustment is also becoming stronger, but traditional automated parameter adjustment cannot completely solve the problem of adjusting multiple parameters at the same time, and the adjustment results of multiple parameters that interact with each other are often not ideal. Therefore, the present invention was born to solve the problem of effectively selecting reasonable hyperparameters under the influence of multiple parameters.

[0166] The automatic parameter adjustment methods usually used include grid search, random search, Bayesian optimization, etc. However, these methods have various problems: grid search wastes a lot of time and space, and is often not suitable for rapid product iteration; random search is likely to miss the optimal point; traditional Bayesian optimization, although providing a black box of hyperparameters, is easy to fall into local optimality under multiple interacting parameters, or it is difficult to find a balance to achieve the best overall effect, especially in multi-parameter scenarios. If multiple parameters restrict and influence each other, the optimal hyperparameters selected by the Bayesian algorithm are often not effective.

[0167] In this process, Bayesian optimization depends on our objective function y=f(x), which is Max(sum(y1,y2…yt)-sum(ybase1,ybase2,…ybaset)). However, the traditional objective function often seeks the optimal point in the adjustment of multiple parameters, which is not the optimal result in many scenarios. It can be seen that the above expression seeks to maximize the total value. In the scenario where multiple parameters are independent of each other, the ideal effect can be achieved. However, in many scenarios, the parameters restrict each other. Under the influence of this goal, the effects brought by certain parameters are often sacrificed. Sometimes, the effects brought by these sacrificed parameters play a vital role in the overall improvement. Therefore, we define a strong constraint algorithm combined with Bayesian optimization to improve the blindness of our objective equation and avoid obtaining local optimality or effect deviation. In this process, we can strongly limit the interaction between parameters and effectively transform business experience into solvable equations for repeated use.

[0168] First, let's look at the optimization of a single parameter. We generally define the target equation Ysingle = f(Xsingle), and by finding the optimal parameter xt, we can make the result yt the optimal solution, that is, we can obtain max(yn)n = 1, 2, 3...t. For multiple parameters, we will find the maximum sum of multiple Ysingles, that is, we will find a set of parameters Xt = {x1, x2...xn} to maximize the result and find Max(Yt). Max(Yt) = Max(y1+y2+y3...yt). This exposes the two problems we mentioned earlier. First, the rate at which each parameter yt changes with the change of xt is different, which will cause the faster changing yt to be assigned more weight, that is, for a faster changing (larger slope equation g(x)) g(x+1) will increase more than the value of f(x+1) with a smaller slope, so Bayes will be more willing to adjust the g(x) with a larger slope. Second, our business experience is difficult to fall into the Bayesian equation, so no matter what kind of automatic parameter adjustment, we cannot learn business experience.

[0169] This embodiment provides a combination method based on the Bayesian principle and a strong constraint algorithm. On the one hand, it retains the ability of Bayesian to solve the waste of computing time and space. On the other hand, the strong constraint algorithm of parameters solves the problem that Bayesian is prone to fall into local optimality and finally cannot achieve ideal benefits. It solves the problem that the overall effect of the optimized hyperparameters decreases instead of increases due to the mutual constraints and influences of multiple parameters. At the same time, we convert the usable business experience into solvable equations for repeated use.

[0170] In order to achieve the above object, the present invention adopts the following technical solution, and the specific steps are as follows:

[0171] Step 1: Select a set of hyperparameters X by artificial experience, or randomly initialize a set of hyperparameters X, and calculate the score Ybase = {ybase1, ybase2, ybase3, ..., ybasen} for each target corresponding to the hyperparameter set Xbase = {xbase1, xbase2, xbase3, ..., xbasen}. This constitutes our basic group (Xbase, Ybase), which is used to compare with the experimental group.

[0172] Step 2: Define our optimization goal. Each time we select the hyperparameter X during the optimization process, we will get our target score Y = {y1, y2, y3…, yn}. We need to expand the distance between the sum of our target score Y and the sum of the basic group, that is, Max(sum(y1, y2…yn)-sum(ybase1, ybase2,…ybasen)), select the one greater than 0 (select the experimental group). This is the form of selecting the target equation in the traditional business scenario. We mainly make improvements here. For details, please see the detailed explanation in the notes of the subsequent steps. (Not necessary, it can be used to determine step 5 and stop iteration)

[0173] Step 3: Bayesian will randomly generate several sets of hyperparameters X1, X2, X3…Xt, and their corresponding target score results combined with Y1, Y2, Y3, Y4…Yt in the initialization stage. Bayesian believes that it obeys Gaussian distribution, so we can construct the data set D = {(X1, Y1), (X2, Y2)…(Xt, Yt)} and the conditional probability distribution of Y after updating D.

[0174] Step 4: Bayes uses the Gaussian distribution initialized in step 3 and the objective equation defined in step 2 to find a balance between development and exploration, find the next set of parameters Dt = (Xt, Yt) that maximizes Y (the sum of y), and then add Dt = (Xt, Yt) to the total set D. You can stop when the number of iterations t reaches a certain number.

[0175] Step 5: The maximum parameter aggregation Dt = (Xt, Yt) found in each iteration will update the original data set D. The change of the original data set brings a new conditional probability distribution Y, and the new distribution can be used to find the next optimal solution. So iterate step 4 repeatedly until the final convergence, so that the posterior distribution is close to the true distribution. Get the optimal hyperparameter.

[0176] In another embodiment, reference Figure 6 , which is a schematic diagram of the changing trend of alternative mapping values ​​in an application scenario.

[0177] Among them, p1, p2, p3, p4, p5, and p6 are 6 different dimensional information. The horizontal axis represents the number of candidate mapping values, and the vertical axis represents the rate of change of the candidate mapping values ​​as the number of candidate mapping values ​​increases. As the number of horizontal axis parameter values ​​increases, the change of the candidate mapping value corresponding to the vertical axis coordinate includes the change of the parameter auc.

[0178] The traditional Bayesian iterative calculation will solve Y = MAX (X * AUC), which will eventually lead to p6 being too large because of its rapid growth rate, and p3 may gradually disappear because it is decreasing rapidly. But this is actually not what we want. Because the online benefits brought by a 1% increase in p1 may be greater than the online benefits brought by a 10% increase in p6.

[0179] Therefore, our constraint condition becomes f(X)=Max(a*y1+b*y2+c*y3+d*y4+e*y5).

[0180] We can increase the weight of p1 by 10 times that of p6 to balance the benefits. In this way, we can reasonably combine the business and turn all the restrictions and constraints into solvable equations.

[0181] refer to Figure 7 , which is a flowchart of another target information recommendation method.

[0182] The business experience analyzer in the figure is used to determine the candidate information to be recommended in combination with the actual business needs, and the equation builder in the target equation builder can be used to determine the first sum value, the second sum value, and the difference between the first sum value and the second sum value, etc. The value converter can be used to convert between weights and mapping values.

[0183] It should be noted that the “first” and “second” in the embodiments of the present disclosure are only for the convenience of description and distinction and have no other specific meanings.

[0184] Figure 8 1 is a block diagram of a terminal device according to an exemplary embodiment. For example, the terminal device may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0185] Reference Figure 8 The terminal device may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .

[0186] The processing component 802 generally controls the overall operation of the terminal device, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0187] The memory 804 is configured to store various types of data to support operations on the terminal device. Examples of such data include instructions for any application or method operating on the terminal device, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0188] The power component 806 provides power to various components of the terminal device. The power component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the terminal device.

[0189] The multimedia component 808 includes a screen that provides an output interface between the terminal device and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the terminal device is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0190] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the terminal device is in an operation mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0191] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.

[0192] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the terminal device. For example, the sensor assembly 814 can detect the open / closed state of the terminal device, the relative positioning of components, such as the display and keypad of the terminal device, and the sensor assembly 814 can also detect the position change of the terminal device or a component of the terminal device, the presence or absence of user contact with the terminal device, the orientation or acceleration / deceleration of the terminal device, and the temperature change of the terminal device. The sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0193] The communication component 816 is configured to facilitate wired or wireless communication between the terminal device and other devices. The terminal device can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0194] In an exemplary embodiment, the terminal device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0195] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0196] It should be understood that the present disclosure is not limited to the exact 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 disclosure is limited only by the appended claims.

Claims

1. A method for recommending target information, characterized in that: include: Obtaining a reference parameter set of at least one dimension of candidate information to be recommended and a plurality of candidate parameter sets of the at least one dimension of information; wherein the reference parameter set includes: a reference weight subset of at least one dimension of information and a reference mapping value subset corresponding to the reference weight subset; the candidate parameter set includes: a candidate weight subset of the at least one dimension of information and a candidate mapping value subset corresponding to each of the candidate weight subsets; Determine a target mapping value subset from the candidate mapping value subsets according to a difference between the first sum value and the second sum value; wherein the first sum value is: the sum of elements in any one of the candidate mapping value subsets; the second sum value is the sum of elements in the reference mapping value subset, and each element in the candidate mapping value subset has a proportional coefficient, and the proportional coefficient is used to indicate the influence of the dimensional information on determining the recommended target information from the candidate information; Determining a target weight subset according to the target mapping value subset; the target weight subset includes: a target weight of at least one dimensional information affecting the candidate information; The recommended target information is selected from the candidate information according to the target weight subset.

2. The method according to claim 1, characterized in that The proportionality coefficient is a preset value; The step of determining the target mapping value subset from the candidate mapping value subset according to the difference between the first sum value and the second sum value comprises: determining the first sum value according to a proportional coefficient of each element in the candidate mapping value subset; According to the difference between the first sum value and the second sum value, the candidate mapping value subset with the largest difference, or the candidate mapping value subset with the difference greater than a preset difference, is determined as the target mapping value subset.

3. The method according to claim 1, characterized in that The method further comprises: The scaling factor is updated according to the candidate mapping value subset.

4. The method according to claim 3, characterized in that The updating of the proportionality coefficient according to the candidate mapping value subset comprises: Determine, according to the multiple candidate mapping values ​​respectively corresponding to the respective dimensional information, the change trend of the multiple candidate mapping values ​​respectively corresponding to the respective dimensional information; According to each of the change trends, the proportional coefficient is adjusted until each of the change trends reaches the target range; wherein the change trends corresponding to different dimensional information respectively have corresponding target ranges.

5. The method according to claim 4, characterized in that The target range is: a threshold range determined according to the proportion of the dimension information when the target information is selected from the candidate information.

6. The method according to claim 1, characterized in that The acquiring of a plurality of candidate parameter sets of the at least one dimension information includes: Preset T initialization parameter sets, wherein the initialization parameter sets include: an initialization weight subset of the at least one dimension information and an initialization mapping value subset corresponding to the initialization weight subset; Determine a T+1th parameter set according to the third sum of the elements in the initialization weight subsets in each initialization parameter set and the fourth sum of the elements in each initialization mapping value subset; wherein the T+1th parameter set includes: the T+1th weight subset and the T+1th mapping value subset, and T is a positive integer greater than 1; Determine the T+n+1th parameter set according to the T+n parameter sets; n is a positive integer, and the sum of the elements in the T+nth mapping value subset is a local maximum; The T+n+1 parameter sets are determined as the candidate parameter sets.

7. The method according to claim 6, characterized in that The preset T initialization parameter sets include: Determine T initialization parameter sets matching the candidate information according to the historical parameter set of the candidate information; wherein the historical parameter set includes: a historical weight subset of at least one dimension information and a historical mapping value subset corresponding to the historical weight subset; or, According to the candidate information to be recommended, T initialization parameter sets are randomly generated.

8. The method according to claim 1, characterized in that Determining a target weight subset according to the target mapping value subset includes: The candidate weight subset corresponding to the target mapping value subset is determined as the target weight subset.

9. The method according to any one of claims 1 to 8, characterized in that: The candidate information to be recommended includes at least one of the following: video; Official account; article; The dimension information includes at least one of the following: Number of downloads, number of favorites, number of likes, number of users following the author of the alternative information, average browsing time, number of comments, and number of reposts.

10. A device for recommending target information, characterized in that: include: An acquisition module, used to acquire a reference parameter set of at least one dimension of candidate information to be recommended and multiple candidate parameter sets of the at least one dimension of information; wherein the reference parameter set includes: a reference weight subset of at least one dimension of information and a reference mapping value subset corresponding to the reference weight subset; the candidate parameter set includes: a candidate weight subset of the at least one dimension of information and a candidate mapping value subset corresponding to each of the candidate weight subsets; a target mapping value subset determination module, configured to determine a target mapping value subset from the candidate mapping value subset according to a difference between a first sum value and a second sum value; wherein the first sum value is: the sum of elements in any one of the candidate mapping value subsets; the second sum value is the sum of elements in the reference mapping value subset, and each element in the candidate mapping value subset has a proportional coefficient, and the proportional coefficient is used to indicate the degree of influence of the dimensional information on the determination of the recommended target information from the candidate information; A target weight subset determination module is used to determine a target weight subset according to the target mapping value subset; the target weight subset includes: a target weight of at least one dimensional information affecting the candidate information; A recommendation module is used to select the recommended target information from the candidate information according to the target weight.

11. An electronic device, characterized in that: include: A processor and a memory for storing executable instructions capable of running on the processor, wherein: When the processor is used to run the executable instructions, the executable instructions execute the method described in any one of claims 1 to 9 above.

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

Citation Information

Patent Citations

  • Information recommendation method and apparatus, and computer device

    CN108304429A

  • Resource recommendation method, device, apparatus, and storage medium

    CN109241425A