A sequencing optimization method, device, equipment and storage medium

Abstract

Embodiments of the present disclosure provide a ranking optimization method and device, equipment and a storage medium. The method comprises: obtaining predicted scores and real scores of a recommendation object on each target; determining abnormal vertical categories and corresponding abnormal targets according to the predicted scores and real scores of the recommendation object on each target, to form abnormal combination items; generating weight optimization items matched with each abnormal combination item, correcting a multi-target fusion formula according to the weight optimization items; calculating final scores of each recommendation object using the corrected multi-target fusion formula, and ranking each recommendation object according to the final scores. Through the above technical solution, the accuracy of ranking is improved by adding weight optimization items to the multi-target fusion formula to optimize the abnormal vertical categories and corresponding targets of the prediction results.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the technical field of computer, and particularly, to a ranking optimization method and device, equipment and storage medium. BACKGROUND

[0002] In a recommendation system, when several hundred objects to be recommended are roughly screened out from a large number of objects, a specific model of multiple targets is usually used to give a predicted score of the objects to be recommended on each target, and the predicted scores of the objects to be recommended on each target are fused into a total score, based on which all the objects to be recommended are finally ranked and a small number of high-quality recommended objects are selected and recommended to users.

[0003] However, the ranking result obtained by using the ranking method is not accurate, which affects the user experience of the recommendation system. Therefore, it is necessary to optimize the ranking method in the recommendation system. SUMMARY

[0004] Embodiments of the present disclosure provide a ranking optimization method, device, equipment and storage medium, so as to realize optimization of abnormal vertical categories and corresponding targets of abnormal vertical categories by adding a weight optimization term in a multi-target fusion formula, and improve the accuracy of ranking.

[0005] In a first aspect, embodiments of the present disclosure provide a ranking optimization method, comprising:

[0006] obtaining predicted scores and real scores of recommended objects on each target;

[0007] determining abnormal vertical categories and corresponding abnormal targets according to the predicted scores and real scores of the recommended objects on each target, and composing abnormal combination terms;

[0008] generating weight optimization terms matched with each abnormal combination term, and correcting a multi-target fusion formula according to the weight optimization terms;

[0009] calculating final scores of each recommended object using the corrected multi-target fusion formula, and ranking each recommended object according to the final scores.

[0010] In a second aspect, embodiments of the present disclosure also provide a ranking optimization device, comprising:

[0011] a score obtaining module configured to obtain predicted scores and real scores of recommended objects on each target;

[0012] an abnormality detecting module configured to determine abnormal vertical categories and corresponding abnormal targets according to the predicted scores and real scores of the recommended objects on each target, and compose abnormal combination terms;

[0013] The correction module is used to generate weight optimization terms that match each of the abnormal combination terms, and to correct the multi-objective fusion formula according to the weight optimization terms;

[0014] The ranking module is used to calculate the final score of each recommended object using the modified multi-objective fusion formula, and to rank each recommended object according to the final score.

[0015] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising:

[0016] One or more processors;

[0017] Storage device for storing one or more programs.

[0018] When the one or more programs are executed by the one or more processors, the one or more processors implement a sorting optimization method as described in the embodiments of this disclosure.

[0019] Fourthly, embodiments of this disclosure also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a sorting optimization method as described in embodiments of this disclosure.

[0020] This embodiment of the disclosure obtains the predicted and actual scores of recommended objects on each target, and determines abnormal vertical categories and their corresponding abnormal targets based on the predicted and actual scores of the recommended objects on each target, forming abnormal combination items; generates weight optimization terms that match each of the abnormal combination items, and modifies the multi-target fusion formula based on the weight optimization terms; uses the modified multi-target fusion formula to calculate the final score of each recommended object, and sorts the recommended objects based on the final scores. This solves the problem of low accuracy in the ranking method of the recommendation system. By adding weight optimization terms to the multi-target fusion formula, the vertical categories with abnormal prediction results and their corresponding targets are optimized, thereby improving the accuracy of the ranking. Attached Figure Description

[0021] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0022] Figure 1 This is a flowchart illustrating a sorting optimization method provided in an embodiment of this disclosure;

[0023] Figure 2 This is a flowchart illustrating another sorting optimization method provided in an embodiment of this disclosure;

[0024] Figure 3 This is a schematic diagram of a sorting optimization device provided in an embodiment of the present disclosure;

[0025] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0026] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0027] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0028] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0029] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0030] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0031] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0032] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0033] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.

[0034] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0035] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0036] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0037] Figure 1 This is a flowchart illustrating a ranking optimization method provided in this embodiment. This embodiment is applicable to situations where, for vertical categories with poor prediction results and their corresponding targets, a correction factor is added to the multi-objective fusion formula to improve the accuracy of the final score given by the multi-objective fusion formula. This method can be executed by a ranking optimization device, which can be implemented in the form of software and / or hardware, or optionally, by an electronic device, such as a mobile terminal, a PC, or a server.

[0038] like Figure 1 As shown, the method includes:

[0039] S110. Obtain the predicted score and actual score of the recommended object on each target.

[0040] The recommended content can be multimedia data such as videos, music, images, or text to be recommended to users. The target refers to the recommendation feedback metrics that match the recommended content. Taking short videos as an example, the targets for the recommended content could include completion rate, viewing time, number of likes, and profile picture click-through rate.

[0041] In one optional implementation, obtaining the predicted score and actual score of the recommended object on each target includes: using the prediction model corresponding to each target to obtain the predicted score of each recommended object on each target; and calculating the actual score of each recommended object on each target based on the historical user feedback data corresponding to each recommended object.

[0042] In this embodiment, each target has a corresponding prediction model. By inputting the recommended object into the prediction model corresponding to each target, the prediction score of each recommended object on each target can be obtained. For example, if video A is input into the prediction model corresponding to the completion rate target, the prediction model predicts that the completion rate of video A after being pushed to users is 0.7, which means that out of 100 users who are pushed video A, 70 users will watch video A completely.

[0043] By analyzing historical user feedback data for each recommended item, the true score for each item on each objective can be calculated. For example, taking video B as an example, by statistically analyzing the historical user feedback data of video B, we can determine the number of users who watched video B in its entirety, the number of users whose viewing time reached 60% of the total duration of video B, and the number of users who liked video B. Based on this information, we can then calculate the true score for video B on each objective.

[0044] S120. Based on the predicted scores and actual scores of the recommended objects on each target, determine the abnormal vertical categories and their corresponding abnormal targets, and form abnormal combination items.

[0045] In this embodiment, there is usually a deviation between the predicted score and the actual score of the recommended object on the target. However, this deviation should be controlled within a reasonable range so that the predicted score of the recommended object on the target is relatively accurate. This ensures the accuracy of the ranking result when the predicted scores of the recommended object on each target are weighted and summed using a multi-target fusion formula to obtain the final score and the recommended objects are ranked based on the final score.

[0046] However, the deviation between the predicted scores and actual scores of each recommended object on each objective is not necessarily within a reasonable range. Furthermore, due to the similarity between recommended objects belonging to the same vertical category, the predicted score given by the prediction model for a specific objective within that vertical category may often deviate significantly from the actual score. In other words, a particular vertical category may exhibit abnormal predicted scores for a particular objective.

[0047] Based on this, we can identify the abnormal vertical categories and their corresponding abnormal targets with large prediction deviations by comparing the predicted scores and actual scores of each recommended object on each target. Each pair of abnormal vertical categories and abnormal targets is treated as an abnormal combination item. Therefore, by compensating for the abnormality of each abnormal combination item in the multi-target fusion formula, we can improve the accuracy of the final ranking of recommended objects.

[0048] A vertical category can be understood as a specific field where users share similar needs or interests. For example, within the short video category, sports short videos are one vertical category, while travel short videos are another.

[0049] S130. Generate weight optimization terms that match each of the aforementioned abnormal combination terms, and modify the multi-objective fusion formula according to the weight optimization terms.

[0050] In this embodiment, a matching weight optimization term is generated for each abnormal combination item. By using the weight optimization term to modify the multi-objective fusion formula, when the predicted scores of the recommended object on each objective are weighted and summed to obtain the final score using the multi-objective fusion formula, the predicted scores of the recommended object belonging to the abnormal vertical category on the abnormal objective can be compensated and adjusted accordingly.

[0051] Optionally, generating a weight optimization term matching each of the abnormal combination items includes: if the abnormal combination item is an abnormally underestimated combination item, generating a weight increase term matching the abnormally underestimated combination item; if the abnormal combination item is an abnormally overestimated combination item, generating a weight decrease term matching the abnormally overestimated combination item.

[0052] In this embodiment, since prediction bias may be caused by an overestimation or underestimation of the predicted score of the recommended object on the target, abnormal combination items can be divided into two types. If the abnormal combination item corresponds to a situation where the predicted score is lower than the actual score, then the abnormal combination item is treated as an abnormal underestimation combination item. In this case, a matching weight increase item should be generated to compensate for the underestimation of the predicted score of the recommended object belonging to the abnormal vertical category on the abnormal target. If the abnormal combination item corresponds to a situation where the predicted score is higher than the actual score, then the abnormal combination item is treated as an abnormal overestimation combination item. In this case, a matching weight decrease item should be generated to correct the overestimation of the predicted score of the recommended object belonging to the abnormal vertical category on the abnormal target.

[0053] Optionally, the step of modifying the multi-objective fusion formula according to the weight optimization terms includes: multiplying the multi-objective fusion formula by each of the weight optimization terms.

[0054] In this embodiment, by multiplying the weight optimization term of the abnormal combination item matching with the multi-objective fusion formula, after calculating the weighted total score of the predicted score of the recommended object on each objective using the multi-objective fusion formula, the abnormal objectives of the abnormal vertical category can be optimized specifically based on the weighted total score through the weight optimization term.

[0055] S140. Calculate the final score of each recommended object using the modified multi-objective fusion formula, and rank each recommended object according to the final score.

[0056] In this embodiment, when calculating the final score of each recommended object using the modified multi-objective fusion formula, the original multi-objective fusion formula is first used to calculate the weighted total score of the predicted scores of the recommended object on each objective. Then, based on the various weight optimization terms in the formula, the scores of recommended objects belonging to the corresponding abnormal vertical category are adjusted based on the prediction deviation of the abnormal objective on the basis of the weighted total score. For recommended objects not belonging to the abnormal vertical category, the weighted total score is not adjusted. The score after adjusting the weighted total score with all weight optimization terms is taken as the final score, and all recommended objects are ranked according to the final score.

[0057] The technical solution of this disclosure, through obtaining the predicted and actual scores of recommended objects on each target, and determining abnormal vertical categories and their corresponding abnormal targets based on the predicted and actual scores of recommended objects on each target, forms abnormal combination items; generating weight optimization terms matching each of the abnormal combination items, and modifying the multi-target fusion formula according to the weight optimization terms; using the modified multi-target fusion formula to calculate the final score of each recommended object, and ranking each recommended object according to the final score, solves the problem of low accuracy of ranking methods in recommendation systems. By adding weight optimization terms to the multi-target fusion formula, the vertical categories with abnormal prediction results and their corresponding targets are optimized, thereby improving the accuracy of ranking.

[0058] Figure 2 This is a flowchart illustrating another ranking optimization method provided in this disclosure. Based on the above embodiments, this embodiment further provides specific steps for determining abnormal vertical categories and their corresponding abnormal targets based on the predicted and actual scores of the recommended object on each target, forming abnormal combination items, and specific steps for generating weight-increasing items that match the abnormal underestimation combination items. Figure 2 As shown, the method includes:

[0059] S210. Obtain the predicted score and actual score of the recommended object on each target.

[0060] S220. Based on the predicted scores and actual scores of the recommended objects on each target, determine the abnormal vertical categories and their corresponding abnormal targets, and form abnormal combination items.

[0061] In one optional implementation, the step of determining abnormal vertical categories and their corresponding abnormal targets based on the predicted scores and actual scores of the recommended objects on each target, and forming abnormal combination items, includes: determining abnormal targets based on the average prediction deviation of all recommended objects on each target; for each abnormal target, determining the abnormal vertical category corresponding to the abnormal target based on the average prediction deviation of each vertical category on the abnormal target; and taking each pair of matched abnormal targets and abnormal vertical categories as an abnormal combination item.

[0062] In this embodiment, for any given target, all recommended objects across all vertical categories have a predicted score for that target. Therefore, if analysis determines that the overall prediction deviation of all recommended objects for that target is large, then that target is considered an anomalous target, and there must be anomalous prediction results for certain vertical categories on that target. At this point, it is necessary to further determine whether the overall prediction performance of each vertical category on that target is within a reasonable range. For each anomalous target, the average prediction deviation of recommended objects for each vertical category on that anomalous target can be calculated. If, based on the average prediction deviation, it is determined that a certain vertical category has a large overall prediction deviation on that anomalous target, then that vertical category is designated as the anomalous vertical category corresponding to that anomalous target, and that anomalous vertical category and the anomalous target are considered as an anomalous combination.

[0063] Optionally, determining abnormal targets based on the average prediction deviation of all recommended objects on each target includes: calculating the difference between the average predicted score and the average actual score of all recommended objects on each target, as the average prediction deviation of each target; and determining targets whose average prediction deviation meets a first threshold condition as abnormal targets.

[0064] In this embodiment, each target is taken as the research object. By analyzing the difference between the average predicted score and the average actual score of each target, the overall prediction accuracy of all recommended targets is determined. A first threshold condition is used to determine whether the overall prediction accuracy of the target meets the standard. If the average prediction deviation of a target meets the first threshold condition, i.e., the overall prediction accuracy does not meet the standard, the target is identified as an abnormal target. If the average prediction deviation of a target does not meet the first threshold condition, i.e., the overall prediction accuracy meets the standard, the target is identified as a normal target. A corresponding first threshold condition can be set for each target, or a single first threshold condition can be shared by all targets.

[0065] Optionally, determining a target whose mean prediction deviation meets the first threshold condition as an abnormal target includes: if the mean prediction deviation is negative and the absolute value of the mean prediction deviation is greater than the first threshold, then the target corresponding to the mean prediction deviation is an abnormally underestimated target; if the mean prediction deviation is positive and the mean prediction deviation is greater than the first threshold, then the target corresponding to the mean prediction deviation is an abnormally overestimated target.

[0066] In this embodiment, since the mean prediction deviation for a target is equal to the difference between the mean predicted score and the mean true score for that target, the mean prediction deviation for a target is negative when the overall predicted score for that target is low. In this case, if the absolute value of the mean prediction deviation is greater than a first threshold, the overall prediction deviation for the target is considered large, and the predicted score of the recommended object for that target is significantly underestimated. Therefore, the target corresponding to the mean prediction deviation is identified as an abnormally underestimated target. Conversely, when the overall predicted score for a target is high, the mean prediction deviation for that target is positive. In this case, if the mean prediction deviation is greater than a first threshold, the overall prediction deviation for the target is considered large, and the predicted score of the recommended object for that target is significantly overestimated. Therefore, the target corresponding to the mean prediction deviation is identified as an abnormally overestimated target.

[0067] Optionally, determining the abnormal vertical category corresponding to each abnormal target based on the average prediction deviation of each vertical category on the abnormal target includes: clustering all recommended objects according to the vertical category to which the recommended object belongs; calculating the difference between the average predicted score and the average actual score of the recommended object corresponding to each vertical category on the abnormal target for each abnormal target, as the average prediction deviation of each vertical category on the abnormal target; and determining the vertical category whose average prediction deviation satisfies the second threshold condition as the abnormal vertical category corresponding to the abnormal target.

[0068] In this embodiment, all recommended objects are first clustered according to their respective vertical categories. Then, for each anomalous target, each vertical category is analyzed to determine whether it is an anomalous category. For example, after selecting an anomalous target, the overall prediction accuracy of each vertical category on that anomalous target is determined by analyzing the difference between the average predicted score and the average actual score of each vertical category on that anomalous target. A second threshold condition is used to determine whether the overall prediction accuracy of a vertical category on the anomalous target meets the standard. If the average prediction deviation of a vertical category on the anomalous target meets the second threshold condition, i.e., the overall prediction accuracy does not meet the standard, then the vertical category is determined to be an anomalous category. If the average prediction deviation of a vertical category on the anomalous target does not meet the second threshold condition, i.e., the overall prediction accuracy meets the standard, then the vertical category is determined to be a normal category. Alternatively, a corresponding second threshold condition can be set for each vertical category, or all vertical categories can share a single second threshold condition.

[0069] In this embodiment, abnormal vertical categories can be further divided into abnormally undervalued vertical categories and abnormally overvalued vertical categories. If the mean prediction deviation of a certain vertical category on an abnormal target is negative, and the absolute value of the mean prediction deviation is greater than a second threshold, then the predicted score of the recommended object of that vertical category on the abnormal target is considered significantly underestimated, and the vertical category can be identified as an abnormally undervalued vertical category. If the mean prediction deviation of a certain vertical category on an abnormal target is positive, and the mean prediction deviation is greater than a second threshold, then the predicted score of the recommended object of that vertical category on the abnormal target is considered significantly overvalued, and the vertical category can be identified as an abnormally overvalued vertical category.

[0070] In another optional implementation, the step of determining abnormal vertical categories and their corresponding abnormal targets based on the predicted scores and actual scores of the recommended objects on each target, and forming an abnormal combination item, includes: clustering all recommended objects according to the vertical category to which the recommended objects belong; calculating the difference between the average predicted score and the average actual score of the recommended objects corresponding to each vertical category on each target, as the average prediction deviation of each vertical category on each target; and taking the vertical category and target corresponding to the average prediction deviation that meets the third threshold condition as an abnormal combination item.

[0071] In this embodiment, all recommended objects can first be clustered according to their respective vertical categories. Then, taking each vertical category as the research object, the overall prediction performance of that vertical category on each target is analyzed to determine whether it is within a reasonable range. Specifically, the overall prediction accuracy of that vertical category on each target can be determined by calculating the difference between the average predicted score and the average actual score of that vertical category on each target. A third threshold condition is used to determine whether the overall prediction accuracy of the vertical category on the target meets the standard. If the average prediction deviation of that vertical category on a certain target meets the third threshold condition, the overall prediction accuracy does not meet the standard, and the vertical category is determined to be an abnormal vertical category. The target is determined to be an abnormal target corresponding to the abnormal vertical category, and the two form an abnormal combination item. If the average prediction deviation of that vertical category on a certain target does not meet the third threshold condition, the overall prediction accuracy of that vertical category on that target meets the standard.

[0072] In this embodiment, abnormal combination items can be divided into abnormally underestimated combination items and abnormally overestimated combination items. If the mean prediction deviation of an abnormal vertical category on the corresponding abnormal target is negative, and the absolute value of the mean prediction deviation is greater than a third threshold, then the predicted score of the recommended object of the abnormal vertical category on the abnormal target is considered to be significantly underestimated. Therefore, the abnormal vertical category is determined to be an abnormally underestimated vertical category, the abnormal target is determined to be an abnormally underestimated target, and the abnormal combination item formed by the two is determined to be an abnormally underestimated combination item. If the mean prediction deviation of an abnormal vertical category on the corresponding abnormal target is positive, and the mean prediction deviation is greater than a third threshold, then the predicted score of the recommended object of the abnormal vertical category on the abnormal target is considered to be significantly overestimated. Therefore, the vertical category is determined to be an abnormally overestimated vertical category, the abnormal target is determined to be an abnormally overestimated target, and the abnormal combination item formed by the two is determined to be an abnormally overestimated combination item.

[0073] S230. If the abnormal combination item is an abnormally undervalued combination item, then generate a weight increase item that matches the abnormally undervalued combination item.

[0074] In one optional implementation, generating a weighting term matching the abnormally underestimated combination includes: generating a vertical indicator variable corresponding to the abnormally underestimated vertical category in the abnormally underestimated combination, wherein the vertical indicator variable is used to indicate whether the vertical category to which the recommended object belongs is the abnormally underestimated vertical category; obtaining the weighting coefficient and weighting index of the abnormally underestimated vertical category on the abnormally underestimated target in the abnormally underestimated combination; and generating a weighting term matching the abnormally underestimated combination based on the vertical indicator variable, the weighting coefficient, the weighting index, and the mean prediction deviation of the abnormally underestimated vertical category on the abnormally underestimated target.

[0075] In this embodiment, the formula (1+is_chuilei*weight*model_score)^alpha can be used as the weight increase term for matching the abnormal underestimation combination item; where is_chuilei is a binary variable used to indicate whether the vertical category to which the recommended object belongs is the abnormal underestimation vertical category in the abnormal underestimation combination item, model_score represents the mean prediction deviation of the abnormal underestimation vertical category on the abnormal underestimation target, weight represents the weighting coefficient of the abnormal underestimation vertical category on model_score, and alpha represents the weighting index of the abnormal underestimation vertical category on model_score.

[0076] S240. If the abnormal combination item is an abnormally overestimated combination item, then generate a weight reduction item that matches the abnormally overestimated combination item.

[0077] Correspondingly, the weight reduction term matched with the abnormally overestimated combination can be represented by the formula (1-is_chuilei2*weight2*model_score2)^alpha2, where is_chuilei2 is a binary variable used to indicate whether the vertical category to which the recommended object belongs is the abnormally overestimated vertical category in the abnormally overestimated combination, model_score2 represents the mean prediction deviation of the abnormally overestimated vertical category on the abnormally overestimated target, weight2 represents the weight coefficient of the abnormally overestimated vertical category on model_score2, and alpha2 represents the weight index of the abnormally overestimated vertical category on model_score2.

[0078] Of course, other methods can be used to represent the weight reduction item. This embodiment does not limit this, as long as it is ensured that the weight reduction item matched with the abnormal overestimation combination item can correct the predicted score of the recommended object belonging to the abnormal vertical category that is overestimated on the abnormal target.

[0079] S250. Multiply each of the weight optimization terms with the multi-objective fusion formula to modify the multi-objective fusion formula.

[0080] In this embodiment, by multiplying the weight optimization term matched by each abnormal combination item with the multi-objective fusion formula, after calculating the weighted total score of the predicted score of the recommended object on each objective using the multi-objective fusion formula, the abnormal objectives of the abnormal vertical category can be optimized specifically based on the weighted total score through the weight optimization term.

[0081] S260. Calculate the final score of each recommended object using the modified multi-objective fusion formula, and rank each recommended object according to the final score.

[0082] The technical solution of this disclosure, through obtaining the predicted and actual scores of recommended objects on each target, and determining abnormal vertical categories and their corresponding abnormal targets based on the predicted and actual scores of recommended objects on each target, forms abnormal combination items; generating weight optimization terms matching each of the abnormal combination items, and modifying the multi-target fusion formula according to the weight optimization terms; using the modified multi-target fusion formula to calculate the final score of each recommended object, and ranking each recommended object according to the final score, solves the problem of low accuracy of ranking methods in recommendation systems. By adding weight optimization terms to the multi-target fusion formula, the vertical categories with abnormal prediction results and their corresponding targets are optimized, thereby improving the accuracy of ranking.

[0083] Figure 3 This is a schematic diagram of a sorting optimization device provided in an embodiment of this disclosure, as shown below. Figure 3As shown, the device includes: a score acquisition module 310, an anomaly detection module 320, a correction module 330, and a sorting module 340.

[0084] The score acquisition module 310 is used to acquire the predicted score and the actual score of the recommended object on each target.

[0085] Anomaly detection module 320 is used to determine abnormal vertical categories and their corresponding abnormal targets based on the predicted scores and actual scores of the recommended objects on each target, and to form anomaly combination items;

[0086] The correction module 330 is used to generate a weight optimization term that matches each of the abnormal combination terms, and to correct the multi-objective fusion formula according to the weight optimization term;

[0087] The sorting module 340 is used to calculate the final score of each recommended object using the modified multi-objective fusion formula, and to sort the recommended objects according to the final score.

[0088] The technical solution of this disclosure, through obtaining the predicted and actual scores of recommended objects on each target, and determining abnormal vertical categories and their corresponding abnormal targets based on the predicted and actual scores of recommended objects on each target, forms abnormal combination items; generating weight optimization terms matching each of the abnormal combination items, and modifying the multi-target fusion formula according to the weight optimization terms; using the modified multi-target fusion formula to calculate the final score of each recommended object, and ranking each recommended object according to the final score, solves the problem of low accuracy of ranking methods in recommendation systems. By adding weight optimization terms to the multi-target fusion formula, the vertical categories with abnormal prediction results and their corresponding targets are optimized, thereby improving the accuracy of ranking.

[0089] Optionally, the score acquisition module 310 is used for:

[0090] Use the prediction model corresponding to each objective to obtain the prediction score of each recommended object on each objective.

[0091] Based on the historical user feedback data corresponding to each recommended object, calculate the true score of each recommended object on each objective.

[0092] Optionally, the anomaly detection module 320 includes:

[0093] The first determining unit is used to determine the abnormal targets based on the average prediction deviation of all recommended objects on each target;

[0094] The second determining unit is used to determine the abnormal vertical category corresponding to each abnormal target based on the average prediction deviation of each vertical category on the abnormal target.

[0095] The combination unit is used to combine each pair of matched abnormal targets and abnormal verticals into an abnormal combination item.

[0096] Optionally, the first determining unit includes:

[0097] The calculation subunit is used to calculate the difference between the average predicted score and the average actual score of all recommended objects on each target, which is used as the average prediction bias of each target.

[0098] A sub-unit is defined to identify targets whose average prediction deviation meets the first threshold condition as abnormal targets.

[0099] And, the second determining unit, used for:

[0100] Cluster all recommended objects based on the vertical category to which they belong;

[0101] For each anomalous target, the difference between the mean predicted score and the mean actual score of the recommended object corresponding to each vertical category on the anomalous target is calculated as the mean prediction bias of each vertical category on the anomalous target.

[0102] The vertical category whose mean prediction deviation meets the second threshold condition is determined as the abnormal vertical category corresponding to the abnormal target.

[0103] Optionally, determine the sub-unit for:

[0104] If the mean of the prediction deviation is negative and the absolute value of the mean of the prediction deviation is greater than the first threshold, then the target corresponding to the mean of the prediction deviation is an abnormally underestimated target.

[0105] If the mean of the prediction deviation is positive and the mean of the prediction deviation is greater than the first threshold, then the target corresponding to the mean of the prediction deviation is an abnormally overestimated target.

[0106] Optional, the anomaly detection module 320 is used for:

[0107] Cluster all recommended objects based on the vertical category to which they belong;

[0108] Calculate the difference between the mean predicted score and the mean actual score of the recommended object for each vertical category on each target, and use it as the mean prediction bias of each vertical category on each target.

[0109] The vertical category and target corresponding to the mean prediction deviation that meets the third threshold condition are treated as an anomaly combination.

[0110] Optionally, the correction module 330 includes: an optimization term generation unit, used for...

[0111] If the abnormal combination item is an abnormally undervalued combination item, then a weight increase item matching the abnormally undervalued combination item is generated;

[0112] If the abnormal combination is an abnormally overestimated combination, then a weight reduction term matching the abnormally overestimated combination is generated.

[0113] Optionally, the optimized item generation unit includes: adding an item generation subunit for...

[0114] Generate a vertical category indicator variable corresponding to the abnormal underestimation vertical category in the abnormal underestimation combination item. The vertical category indicator variable is used to indicate whether the vertical category to which the recommended object belongs is the abnormal underestimation vertical category.

[0115] Obtain the weighting coefficient and weighting index of the abnormally undervalued vertical category on the abnormally undervalued target in the abnormally undervalued combination item;

[0116] Based on the vertical category indicator variable, the weighting coefficient, the weighting index, and the mean prediction deviation of the abnormally undervalued vertical category on the abnormally undervalued target, a weighting increase term matching the abnormally undervalued combination term is generated.

[0117] Optionally, the correction module 330 includes:

[0118] The formula correction unit is used to multiply the multi-objective fusion formula with each of the weight optimization terms.

[0119] The sorting optimization apparatus provided in this disclosure can execute the sorting optimization method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the method execution.

[0120] It is worth noting that the various units and modules included in the above-mentioned device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this disclosure.

[0121] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Reference is made below. Figure 4 It illustrates an electronic device suitable for implementing embodiments of the present disclosure (e.g., Figure 4 The diagram below shows the structure of the terminal device or server 500. The terminal device in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 4The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0122] like Figure 4 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An edit / output (I / O) interface 505 is also connected to the bus 504.

[0123] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0124] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, it performs the functions defined in the methods of embodiments of this disclosure.

[0125] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0126] The electronic device provided in this embodiment and the sorting optimization method provided in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0127] This disclosure provides a computer storage medium storing a computer program that, when executed by a processor, implements the sorting optimization method provided in the above embodiments.

[0128] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0129] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol, such as HTTP (Hypertext Transfer Protocol), and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0130] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0131] The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to:

[0132] Obtain the predicted and actual scores of the recommended objects on each objective; based on the predicted and actual scores of the recommended objects on each objective, determine the abnormal vertical categories and their corresponding abnormal objectives, forming abnormal combination items; generate weight optimization terms that match each of the abnormal combination items, and modify the multi-objective fusion formula according to the weight optimization terms; use the modified multi-objective fusion formula to calculate the final score of each recommended object, and rank the recommended objects according to the final scores.

[0133] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0134] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0135] The units described in the embodiments of this disclosure can be implemented in software or in hardware. The name of a unit does not necessarily limit the unit itself; for example, the first acquisition unit can also be described as "a unit that acquires at least two Internet Protocol addresses".

[0136] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0137] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0138] According to one or more embodiments of this disclosure, Example 1 provides a sorting optimization method, including:

[0139] Obtain the predicted and actual scores of the recommended object for each objective;

[0140] Based on the predicted and actual scores of the recommended objects on each target, abnormal vertical categories and their corresponding abnormal targets are determined to form abnormal combination items.

[0141] Generate weight optimization terms that match each of the aforementioned abnormal combination terms, and modify the multi-objective fusion formula based on the weight optimization terms;

[0142] The final score of each recommended object is calculated using the modified multi-objective fusion formula, and the recommended objects are ranked according to the final score.

[0143] According to one or more embodiments of this disclosure, Example 2, based on the method described in Example 1, includes obtaining the predicted scores and actual scores of the recommended object on each target, comprising:

[0144] Use the prediction model corresponding to each objective to obtain the prediction score of each recommended object on each objective.

[0145] Based on the historical user feedback data corresponding to each recommended object, calculate the true score of each recommended object on each objective.

[0146] According to one or more embodiments of this disclosure, Example 3, based on the method described in Example 1, involves determining abnormal vertical categories and their corresponding abnormal targets based on the predicted and actual scores of the recommended object on each target, and forming abnormal combination items, including:

[0147] Identify outlier targets based on the average prediction deviation of all recommended targets across all objectives;

[0148] For each abnormal target, the abnormal vertical category corresponding to the abnormal target is determined based on the average prediction deviation of each vertical category on the abnormal target;

[0149] Each pair of matching abnormal targets and abnormal verticals is treated as an abnormal combination item.

[0150] According to one or more embodiments of this disclosure, Example 4 describes the method described in Example 3, wherein determining the anomalous target based on the mean prediction deviation of all recommended objects on each target includes:

[0151] Calculate the difference between the mean predicted score and the mean actual score of all recommended objects on each target, and use it as the mean prediction bias for each target.

[0152] Targets whose mean prediction deviation meets the first threshold condition are identified as abnormal targets;

[0153] as well as,

[0154] For each anomalous target, determining the anomalous vertical category corresponding to the anomalous target based on the average prediction deviation of each vertical category on the anomalous target includes:

[0155] Cluster all recommended objects based on the vertical category to which they belong;

[0156] For each anomalous target, the difference between the mean predicted score and the mean actual score of the recommended object corresponding to each vertical category on the anomalous target is calculated as the mean prediction bias of each vertical category on the anomalous target.

[0157] The vertical category whose mean prediction deviation meets the second threshold condition is determined as the abnormal vertical category corresponding to the abnormal target.

[0158] According to one or more embodiments of this disclosure, Example 5, based on the method described in Example 4, identifies targets whose mean prediction deviation meets a first threshold condition as anomalous targets, including:

[0159] If the mean of the prediction deviation is negative and the absolute value of the mean of the prediction deviation is greater than the first threshold, then the target corresponding to the mean of the prediction deviation is an abnormally underestimated target.

[0160] If the mean of the prediction deviation is positive and the mean of the prediction deviation is greater than the first threshold, then the target corresponding to the mean of the prediction deviation is an abnormally overestimated target.

[0161] According to one or more embodiments of this disclosure, Example 6 describes the method described in Example 1, wherein determining the abnormal vertical category and its corresponding abnormal target based on the predicted score and actual score of the recommended object on each target, and forming an abnormal combination item, includes:

[0162] Cluster all recommended objects based on the vertical category to which they belong;

[0163] Calculate the difference between the mean predicted score and the mean actual score of the recommended object for each vertical category on each target, and use it as the mean prediction bias of each vertical category on each target.

[0164] The vertical category and target corresponding to the mean prediction deviation that meets the third threshold condition are treated as an anomaly combination.

[0165] According to one or more embodiments of this disclosure, Example 7 describes the method according to any one of Examples 1-6, wherein generating a weight optimization term matching each of the anomalous combination terms includes:

[0166] If the abnormal combination item is an abnormally undervalued combination item, then a weight increase item matching the abnormally undervalued combination item is generated;

[0167] If the abnormal combination is an abnormally overestimated combination, then a weight reduction term matching the abnormally overestimated combination is generated.

[0168] According to one or more embodiments of this disclosure, Example 8, based on the method of Example 7, the generation of a weighting term matching the abnormally undervalued combination term includes:

[0169] Generate a vertical category indicator variable corresponding to the abnormal underestimation vertical category in the abnormal underestimation combination item. The vertical category indicator variable is used to indicate whether the vertical category to which the recommended object belongs is the abnormal underestimation vertical category.

[0170] Obtain the weighting coefficient and weighting index of the abnormally undervalued vertical category on the abnormally undervalued target in the abnormally undervalued combination item;

[0171] Based on the vertical category indicator variable, the weighting coefficient, the weighting index, and the mean prediction deviation of the abnormally undervalued vertical category on the abnormally undervalued target, a weighting increase term matching the abnormally undervalued combination term is generated.

[0172] According to one or more embodiments of this disclosure, Example 9 describes the method described in Example 1, wherein the modification of the multi-objective fusion formula based on the weight optimization term includes:

[0173] Multiply the multi-objective fusion formula with each of the weight optimization terms.

[0174] According to one or more embodiments of this disclosure, Example 10 provides a sorting optimization apparatus, comprising:

[0175] The score acquisition module is used to obtain the predicted score and actual score of the recommended object on each target;

[0176] Anomaly detection module is used to determine abnormal vertical categories and their corresponding abnormal targets based on the predicted scores and actual scores of the recommended objects on each target, and to form anomaly combination items;

[0177] The correction module is used to generate weight optimization terms that match each of the abnormal combination terms, and to correct the multi-objective fusion formula according to the weight optimization terms;

[0178] The ranking module is used to calculate the final score of each recommended object using the modified multi-objective fusion formula, and to rank each recommended object according to the final score.

[0179] According to one or more embodiments of this disclosure, Example 11 provides an electronic device, the electronic device comprising:

[0180] One or more processors;

[0181] Storage device for storing one or more programs.

[0182] When the one or more programs are executed by the one or more processors, the one or more processors implement a sorting optimization method as described in any one of Examples 1-9.

[0183] According to one or more embodiments of the present disclosure, Example 12 provides a storage medium containing computer-executable instructions that, when executed by a computer processor, are used to perform a sorting optimization method as described in any one of Examples 1-9.

[0184] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0185] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0186] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for sorting optimization, the method comprising: The method comprises the following steps: obtaining the predicted scores and the real scores of the recommendation objects on each target; determining the abnormal vertical categories and the corresponding abnormal targets according to the predicted scores and the real scores of the recommendation objects on each target, and composing abnormal combination items, wherein the abnormal targets are determined according to the mean predicted deviation of all the recommendation objects on each target; for each abnormal target, the abnormal vertical category corresponding to the abnormal target is determined according to the mean predicted deviation of each vertical category on the abnormal target; each matched abnormal target and abnormal vertical category is taken as an abnormal combination item; generating a weight optimization item matched with each abnormal combination item, and correcting the multi-target fusion formula according to the weight optimization item; calculating the final scores of each recommendation object by using the corrected multi-target fusion formula, and ranking the recommendation objects according to the final scores.

2. The method of claim 1, wherein, The method comprises the following steps: obtaining the predicted scores and the real scores of the recommendation objects on each target; calculating the real scores of the recommendation objects on each target according to the historical user feedback data corresponding to each recommendation object.

3. The method of claim 1, wherein, The method comprises the following steps: calculating the difference between the mean predicted score and the mean real score of all the recommendation objects on each target as the mean predicted deviation of each target; determining the target whose mean predicted deviation meets the first threshold condition as an abnormal target; and The method comprises the following steps: clustering all the recommendation objects according to the vertical categories to which the recommendation objects belong; calculating the difference between the mean predicted score and the mean real score of the recommendation objects corresponding to each vertical category on each abnormal target as the mean predicted deviation of each vertical category on the abnormal target; determining the vertical category whose mean predicted deviation meets the second threshold condition as the abnormal vertical category corresponding to the abnormal target. The method comprises the following steps:

4. The method of claim 3, wherein, if the mean predicted deviation is negative and the absolute value of the mean predicted deviation is greater than the first threshold value, the target corresponding to the mean predicted deviation is an abnormal underestimation target; if the mean predicted deviation is positive and the mean predicted deviation is greater than the first threshold value, the target corresponding to the mean predicted deviation is an abnormal overestimation target. The method comprises the following steps:

5. The method of claim 1, wherein, clustering all the recommendation objects according to the vertical categories to which the recommendation objects belong; calculating the difference between the mean predicted score and the mean real score of the recommendation objects corresponding to each vertical category on each target as the mean predicted deviation of each vertical category on each target; determining the vertical category and the target corresponding to the mean predicted deviation meeting the third threshold condition as an abnormal combination item. The method comprises the following steps:

6. The method according to any one of claims 1 to 5, characterized in that, ​ if the abnormal combination item is an abnormal underestimation combination item, generating a weight increase item matched with the abnormal underestimation combination item; if the abnormal combination item is an abnormal overestimation combination item, generating a weight decrease item matched with the abnormal overestimation combination item.

7. The method of claim 6, wherein, The generating of the weight increase item matched with the abnormal underestimation combination item comprises: generating a vertical class indication variable corresponding to an abnormal underestimation vertical class in the abnormal underestimation combination item, the vertical class indication variable being used to indicate whether a vertical class to which a recommendation object belongs is the abnormal underestimation vertical class; obtain a weight increasing coefficient and a weight increasing index of the abnormal underestimation vertical class on the abnormal underestimation target in the abnormal underestimation combination item, wherein the weight increasing item matched with the abnormal underestimation combination item is: where is_chuilei is a binary variable indicating whether the vertical class to which the recommendation object belongs is the abnormal underestimation vertical class in the abnormal underestimation combination item, model_score represents a predicted bias mean of the abnormal underestimation vertical class on the abnormal underestimation target, weight represents a weight increasing coefficient of the abnormal underestimation vertical class on model_score, and alpha represents a weight increasing index of the abnormal underestimation vertical class on model_score. generating the weight increase item matched with the abnormal underestimation combination item based on the vertical class indication variable, the weight increase coefficient, the weight increase index and a predicted deviation mean value of the abnormal underestimation vertical class on the abnormal underestimation target.

8. The method of claim 1, wherein, The modifying of the multi-target fusion formula according to the weight optimization items comprises: multiplying the multi-target fusion formula by each weight optimization item.

9. An ordering optimization apparatus, characterized by, The method comprises: a score acquisition module, configured to acquire predicted scores and real scores of recommendation objects on each target; an abnormality detection module, configured to determine abnormal vertical classes and corresponding abnormal targets according to the predicted scores and real scores of the recommendation objects on each target, to form abnormal combination items, wherein abnormal targets are determined according to predicted deviation mean values of all recommendation objects on each target; for each abnormal target, abnormal vertical classes corresponding to the abnormal target are determined according to predicted deviation mean values of each vertical class on the abnormal target; each pair of matched abnormal target and abnormal vertical class is taken as an abnormal combination item; a modifying module, configured to generate weight optimization items matched with each abnormal combination item, and to modify a multi-target fusion formula according to the weight optimization items; a sorting module, configured to calculate final scores of each recommendation object using the modified multi-target fusion formula, and to sort the recommendation objects according to the final scores.

10. An electronic device, comprising: The electronic device comprises: one or more processors; a storage device, configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement a sorting optimization method as claimed in any one of claims 1-8.

11. A storage medium containing computer executable instructions for performing a sorting optimization method as claimed in any one of claims 1-8 when executed by a computer processor.

Images