Data recall method, data recall device, electronic equipment and storage medium

Abstract

The application provides a data recall method, a data recall device, an electronic device and a storage medium, and belongs to the technical field of artificial intelligence. The reference recall list and the to-be-recalled data are acquired, the to-be-recalled data is subjected to first recall processing, a first recall list is obtained, the to-be-recalled data is subjected to second recall processing, a second recall list is obtained, the first recall list is ranked and predicted according to the reference recall list, a first average predicted ranking is obtained, the second recall list is ranked and predicted according to the reference recall list, a second average predicted ranking is obtained, the first average predicted ranking and the second average predicted ranking are compared, if the first average predicted ranking is greater than the second average predicted ranking, the second recall list is screened, a third recall list is obtained, and a target recommendation list is generated according to the first recall list and the third recall list. The target recommendation list is used for item recommendation, and the accuracy of item recommendation can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a data retrieval method, a data retrieval device, an electronic device, and a storage medium. Background Technology

[0002] In related technologies, a target recommendation list is obtained by weighted averaging of multiple recall results. However, the recall results from each of the multiple recall paths can vary significantly. Directly weighting and averaging these results would lead to an inaccurate target recommendation list, affecting the accuracy of item recommendations. Therefore, improving the accuracy of item recommendations has become an urgent technical problem to be solved. Summary of the Invention

[0003] The main objective of this application is to provide a data retrieval method, data retrieval device, electronic device, and storage medium, which aim to improve the accuracy of item recommendations.

[0004] To achieve the above objectives, a first aspect of this application proposes a data retrieval method, the method comprising:

[0005] Obtain the baseline recall list and the data to be recalled;

[0006] The data to be recalled is subjected to a first recall process to obtain a first recall list;

[0007] The data to be recalled is subjected to a second recall process to obtain a second recall list;

[0008] The first recall list is ranked and predicted based on the benchmark recall list to obtain the first average predicted ranking of the first recall list, and the second recall list is ranked and predicted based on the benchmark recall list to obtain the second average predicted ranking of the second recall list.

[0009] Compare the first average predicted ranking with the second average predicted ranking;

[0010] If the first average predicted ranking is greater than the second average predicted ranking, then the second recall list is filtered to obtain the third recall list.

[0011] A target recommendation list is generated based on the first recall list and the third recall list; wherein, the target recommendation list is used for item recommendation.

[0012] In some embodiments, obtaining the baseline recall list and the data to be recalled includes:

[0013] Retrieve initial object behavior data within a preset time period from the preset data tracking event table;

[0014] The initial object behavior data is deduplicated to obtain intermediate object behavior data;

[0015] The intermediate object behavior data is classified and processed to obtain the baseline recall list and the data to be recalled.

[0016] In some embodiments, the preset time period includes a first time point and a second time point, and the classification processing of the intermediate object behavior data to obtain the baseline recall list and the data to be recalled includes:

[0017] Calculate the number of intermediate object behavior data at the first time point;

[0018] If the number of data is less than or equal to a preset threshold, then all intermediate object behavior data at the first time point will be used as the baseline recall list.

[0019] If the number of data is greater than the number threshold, the intermediate object behavior data at the first time point is filtered to obtain the baseline recall list;

[0020] The intermediate object behavior data at the second time point is used as the data to be recalled.

[0021] In some embodiments, the data to be recalled includes object information of multiple target objects, item information of multiple items, and behavior information of the target objects towards the items. The first recall process on the data to be recalled to obtain a first recall list includes:

[0022] Based on the object information, the item information, and the behavior information, item similarity is calculated to obtain the item similarity between different items;

[0023] Based on the item similarity and the behavioral information, the target object's first level of interest in the item is obtained;

[0024] The item information is sorted according to the first degree of interest to obtain the first recall list.

[0025] In some embodiments, the second recall process on the data to be recalled to obtain a second recall list includes:

[0026] Based on the object information, the item information, and the behavior information, the object interest similarity is calculated to obtain the object interest similarity between different target objects.

[0027] Based on the object interest similarity and the behavioral information, the second degree of interest of the target object in the item is obtained;

[0028] The item information is sorted according to the second interest level to obtain the second recall list.

[0029] In some embodiments, the step of ranking the first recall list based on the baseline recall list to obtain a first average predicted ranking of the first recall list includes:

[0030] Data matching is performed between the baseline recall list and the first recall list, and the recall data that appears in both the baseline recall list and the first recall list is used as the matched recall data;

[0031] Obtain the position information of the matched recall data in the first recall list;

[0032] The first average predicted ranking is obtained by calculating the predicted ranking of the first recall list based on the location information.

[0033] In some embodiments, generating a target recommendation list based on the first recall list and the third recall list includes:

[0034] Data matching is performed on the first recall list and the third recall list, and the recall data that appears in both the first recall list and the third recall list is taken as the first target recall data;

[0035] The first recall list is cleaned based on the first target recall data to obtain the second target recall data.

[0036] The first target recall data and the second target recall data are rearranged to generate the target recommendation list.

[0037] To achieve the above objectives, a second aspect of this application provides a data retrieval device, the device comprising:

[0038] The acquisition module is used to acquire the baseline recall list and the data to be recalled;

[0039] The first recall module is used to perform a first recall process on the data to be recalled to obtain a first recall list.

[0040] The second recall module is used to perform a second recall process on the data to be recalled to obtain a second recall list.

[0041] The ranking prediction module is used to predict the ranking of the first recall list based on the benchmark recall list to obtain the first average predicted ranking of the first recall list, and to predict the ranking of the second recall list based on the benchmark recall list to obtain the second average predicted ranking of the second recall list.

[0042] The comparison module is used to compare the first average predicted ranking and the second average predicted ranking;

[0043] The filtering module is used to filter the second recall list to obtain a third recall list if the first average predicted ranking is greater than the second average predicted ranking.

[0044] The target recommendation list generation module is used to generate a target recommendation list based on the first recall list and the third recall list; wherein, the target recommendation list is used for item recommendation.

[0045] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0046] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0047] The data recall method, data recall device, electronic device, and computer-readable storage medium proposed in this application obtain a baseline recall list and data to be recalled, perform a first recall process on the data to be recalled to obtain a first recall list, perform a second recall process on the data to be recalled to obtain a second recall list, perform a ranking prediction on the first recall list based on the baseline recall list to obtain a first average predicted ranking of the first recall list, and perform a ranking prediction on the second recall list based on the baseline recall list to obtain a second average predicted ranking of the second recall list. Through the ranking prediction, the similarity between the first recall list and the baseline recall list, and the similarity between the second recall list and the baseline recall list can be obtained. Furthermore, comparing the first average predicted ranking and the second average predicted ranking, if the first average predicted ranking is greater than the second average predicted ranking, it indicates that the first recall list and the baseline recall list are more similar than the second recall list and the baseline recall list. Since the first recall list and the baseline recall list are more similar, the second recall list is filtered to obtain the third recall list. By comparing the first average predicted ranking and the second average predicted ranking, the weight of the first and second recall lists in the target recommendation list can be dynamically adjusted based on their similarity to the baseline recall list. Compared to the weighted average calculation method, this approach takes into account the differences between different recall lists. Finally, a target recommendation list is generated based on the first and third recall lists. This target recommendation list is used for item recommendations. By considering the differences between different recall lists, the accuracy of the target recommendation list can be improved, thereby increasing the accuracy of item recommendations. Attached Figure Description

[0048] Figure 1 This is a flowchart of the data retrieval method provided in the embodiments of this application;

[0049] Figure 2 yes Figure 1 The flowchart of step S110 in the middle;

[0050] Figure 3 yes Figure 2 The flowchart of step S230 in the middle;

[0051] Figure 4 yes Figure 1 The flowchart of step S120 in the middle;

[0052] Figure 5 yes Figure 1 The flowchart of step S130 in the process;

[0053] Figure 6 yes Figure 1 The flowchart of step S140 in the middle;

[0054] Figure 7 yes Figure 1 The flowchart of step S170 in the process;

[0055] Figure 8 This is a schematic diagram of the data retrieval device provided in the embodiments of this application;

[0056] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0060] First, let's analyze some of the terms used in this application:

[0061] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0062] In recommendation systems, the recall layer is an essential component. Since single-path recall often cannot meet the needs of recommendation business, a complete recommendation system usually adopts multi-path recall. Multi-path recall refers to the strategy of using different strategies, features or simple models to recall a portion of the candidate set, and then mixing these candidate sets together for use by the subsequent ranking model.

[0063] Traditional recommendation systems use a multi-path recall fusion method for item recommendation. However, this method involves a weighted average of the scores from each recall path. Since the recall results from each path can vary significantly, directly averaging multiple paths can lead to situations where an item performs well in one path but poorly in another, resulting in a lower ranking for the item after weighted averaging, even though it actually ranks higher. This can cause inaccurate target recommendation lists and negatively impact item recommendation accuracy. Therefore, improving the accuracy of item recommendations has become a pressing technical problem.

[0064] Based on this, embodiments of this application provide a data retrieval method, a data retrieval device, an electronic device, and a computer-readable storage medium, aiming to improve the accuracy of item recommendations.

[0065] The data retrieval method, data retrieval device, electronic device, and computer-readable storage medium provided in this application are specifically described through the following embodiments. First, the data retrieval method in this application is described.

[0066] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0067] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0068] The data retrieval method provided in this application relates to the field of artificial intelligence technology. The data retrieval method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the data retrieval method, but is not limited to the above forms.

[0069] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0070] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0071] Figure 1 This is an optional flowchart of the data retrieval method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S110 to S170.

[0072] Step S110: Obtain the baseline recall list and the data to be recalled;

[0073] Step S120: Perform the first recall process on the data to be recalled to obtain the first recall list;

[0074] Step S130: Perform a second recall process on the data to be recalled to obtain a second recall list;

[0075] Step S140: Rank prediction is performed on the first recall list based on the baseline recall list to obtain the first average predicted ranking of the first recall list; and rank prediction is performed on the second recall list based on the baseline recall list to obtain the second average predicted ranking of the second recall list.

[0076] Step S150: Compare the first average predicted ranking and the second average predicted ranking.

[0077] Step S160: If the first average predicted ranking is greater than the second average predicted ranking, then the second recall list is filtered to obtain the third recall list.

[0078] Step S170: Generate a target recommendation list based on the first recall list and the third recall list; wherein, the target recommendation list is used for item recommendation.

[0079] Steps S110 to S170 as shown in the embodiments of this application involve obtaining a baseline recall list and data to be recalled, performing a first recall process on the data to be recalled to obtain a first recall list, performing a second recall process on the data to be recalled to obtain a second recall list, performing a ranking prediction on the first recall list based on the baseline recall list to obtain a first average predicted ranking of the first recall list, and performing a ranking prediction on the second recall list based on the baseline recall list to obtain a second average predicted ranking of the second recall list. The similarity between the first recall list and the baseline recall list, and the similarity between the second recall list and the baseline recall list can be obtained through the ranking prediction. Furthermore, comparing the first average predicted ranking and the second average predicted ranking, if the first average predicted ranking is greater than the second average predicted ranking, it indicates that the first recall list and the baseline recall list are more similar than the second recall list and the baseline recall list. Since the first recall list and the baseline recall list are more similar, the second recall list is filtered to obtain the third recall list. By comparing the first average predicted ranking and the second average predicted ranking, the weight of the first and second recall lists in the target recommendation list can be dynamically adjusted based on their similarity to the baseline recall list. Compared to the weighted average calculation method, this approach takes into account the differences between different recall lists. Finally, a target recommendation list is generated based on the first and third recall lists. This target recommendation list is used for item recommendations. By considering the differences between different recall lists, the accuracy of the target recommendation list can be improved, thereby increasing the accuracy of item recommendations.

[0080] Please see Figure 2 In some embodiments, step S110 may include, but is not limited to, steps S210 to S230:

[0081] Step S210: Obtain initial object behavior data within a preset time period from the preset data tracking event table;

[0082] Step S220: Perform deduplication on the initial object behavior data to obtain intermediate object behavior data;

[0083] Step S230: Classify the intermediate object behavior data to obtain the baseline recall list and the data to be recalled.

[0084] In step S210 of some embodiments, initial object behavior data within a preset time period is obtained from the data tracking event table. The data tracking event table records interaction events on front-end pages such as websites or applications. The initial object behavior data is the user's historical behavior data on the front-end page, which can include user clicks, favorites, exposures, or purchases. The preset time period can be one week, one month, or one year, for example, the user's historical behavior data for the past month can be obtained from the data tracking event table as the initial object behavior data. When a user triggers a preset behavior rule on the front-end page, the user behavior data is recorded in the backend database, resulting in the data tracking event table. For example, if a user clicks on an article within the application, a click event is generated based on the user's click behavior. In response to the triggering of the click event, user behavior data such as user ID, article ID, article title, and click time are recorded in the backend database, resulting in the data tracking event table.

[0085] In step S220 of some embodiments, the initial object behavior data is cleaned, filtered, and deduplicated to obtain intermediate object behavior data. Specifically, the initial object behavior data is cleaned and filtered to remove invalid data and obtain valid data. Invalid data includes user behavior data with empty user IDs, user behavior data with empty item IDs, or user behavior data that involves time travel. If a piece of user behavior data is uploaded multiple times, it only needs to be retained once. By deduplicating the valid data, intermediate object behavior data is obtained for data analysis.

[0086] In step S230 of some embodiments, the intermediate object behavior data is classified and divided into two parts: one part is used as the real recall data, i.e., the baseline recall list, and the other part is used as the training data, i.e., the data to be recalled.

[0087] Steps S210 to S230 above ensure the accuracy of user behavior data through data cleaning, filtering, and deduplication. By classifying user behavior data, the recall results of the data to be recalled can be evaluated with reference to the baseline recall list, thereby improving the accuracy of item recommendations.

[0088] Please see Figure 3 In some embodiments, the preset time period includes a first time point and a second time point, and step S230 may include, but is not limited to, steps S310 to S330:

[0089] Step S310: Calculate the number of intermediate object behavior data at the first time point;

[0090] Step S320: If the number of data is less than or equal to a preset threshold, then all intermediate object behavior data at the first time point are used as the baseline recall list; if the number of data is greater than the threshold, then the intermediate object behavior data at the first time point are filtered to obtain the baseline recall list.

[0091] Step S330: Use the intermediate object behavior data at the second time point as the data to be recalled.

[0092] In step S310 of some embodiments, the intermediate object behavior data is classified according to a first time point and a second time point. The first time point can be multiple time points within a preset time period, and the second time point can be all or part of the other time points within the preset time period except for the first time point. For example, if the preset time period is the past month, then the first time point can be the last day, and the second time point can be all the other time points within the past month except for the last day. The number of intermediate object behavior data points for a certain user at the first time point is calculated. For example, if the first time point includes 10 intermediate object behavior data points for a certain user, then the number of data points is 10.

[0093] In step S320 of some embodiments, if the number of data items is less than or equal to a threshold, all intermediate object behavior data of the user at the first time point are used as the baseline recall list for that user. If the number of data items is greater than the threshold, the intermediate object behavior data of the user at the first time point are sorted in ascending order according to the timestamp, that is, the intermediate object behavior data of the user at the first time point are sorted from smallest to largest timestamp to obtain a sorted list. The top K intermediate object behavior data items are taken from the sorted list as the baseline recall list, where K is an integer greater than or equal to 1. Typically, a user performs operations on no more than 100 items per day, so K can be 100. If it exceeds 100, the top 100 intermediate object behavior data items are sorted from smallest to largest timestamp and used as the baseline recall list for that user.

[0094] In step S330 of some embodiments, the intermediate object behavior data of each user at the second time point is used as the recall data.

[0095] Through the above steps S310 to S330, the intermediate object behavior data of each user can be classified according to the first time point and the second time point to obtain the data to be recalled and the baseline recall list of each user.

[0096] Please see Figure 4 In some embodiments, the data to be recalled includes object information of multiple target objects, item information of multiple items, and behavior information of the target objects towards the items. Step S120 may include, but is not limited to, steps S410 to S430:

[0097] Step S410: Calculate the item similarity based on object information, item information, and behavior information to obtain the item similarity between different items;

[0098] Step S420: Based on item similarity and behavioral information, obtain the target object's first level of interest in the item;

[0099] Step S430: Sort the item information according to the first degree of interest to obtain the first recall list.

[0100] In step S410 of some embodiments, an item-based collaborative filtering algorithm is used to perform a first recall process on the data to be recalled, that is, to recommend similar items to the user's favorite items, thus obtaining the user's first recall list. Item similarity is calculated based on object information, item information, and behavioral information to obtain the item similarity between each pair of items. The object information can be the user ID, represented as userID; the item information can be the item ID, represented as itemID; and the behavioral information includes the user's preference information for the items and the user's interest rating for the items. Preference information is used to characterize whether the user likes the item; a preference score of 1 indicates that the user likes the item, and a preference score of 0 indicates that the user dislikes the item. The interest rating is used to characterize the degree of the user's liking for the item; a higher interest rating indicates a higher degree of liking for the item, and a lower interest rating indicates a lower degree of liking for the item. If it is necessary to calculate the item similarity between item M and item N, the number of the first object and the number of the second object are determined based on the object information, item information, and user preference information for the items. The number of the first object is the number of users who like both item M and item N, and the number of the second object is the number of users contained in the union of the sets of users who like item M and users who like item N. The item similarity between item M and item N is obtained by dividing the number of the first object and the number of the second object.

[0101] In step S420 of some embodiments, the target object's first interest in the item is obtained based on the item similarity and the user's interest rating of the item. Specifically, based on the item similarity between the current item j and other items, the item information of other items is sorted from largest to smallest to obtain a sorting result. The top P item information is selected from the sorting result to obtain the P items most similar to the current item j, where P is an integer greater than or equal to 1. The set of item information of these P items is S(j,P). The method for calculating the user u's first interest in the current item j is shown in formula (1).

[0102] P uj =∑ i∈N(u)∩S(j,P) w ji r ui Formula (1)

[0103] Where N(u) is the set of items liked by user u, S(j,P) is the set of P items most similar to item j, and r ui Rate user u's interest in item i.

[0104] It should be noted that a priority queue or heap can be used to efficiently retrieve the top P most similar items for each item. Since retrieving the top P most similar items requires sorting the entire item set, the time complexity is O(NlogN), while retrieving the top P items using a priority queue or heap has a time complexity of O(NlogP). P is much smaller than N, making it more time-efficient.

[0105] In step S430 of some embodiments, the user u’s first interest level for each item is calculated, items that the user has already liked are removed, and the item information is sorted from largest to smallest according to the first interest level to obtain P items with the highest scores as the first recall list.

[0106] Through the above steps S410 to S430, similar items to the user's favorite items can be recommended, so that the data retrieval results match the current user's preferences.

[0107] Please see Figure 5 In some embodiments, step S130 may include, but is not limited to, steps S510 to S530:

[0108] Step S510: Calculate object interest similarity based on object information, item information, and behavior information to obtain object interest similarity between different target objects;

[0109] Step S520: Based on the object interest similarity and behavioral information, obtain the target object's second interest in the item;

[0110] Step S530: Sort the item information according to the second degree of interest to obtain the second recall list.

[0111] In step S510 of some embodiments, a second recall is performed on the data to be recalled using a user-based collaborative filtering algorithm, i.e., recommending items liked by similar users to the user, thus obtaining a second recall list for the user. Object interest similarity is calculated based on object information, item information, and user preferences for items to obtain the object interest similarity between different target objects. If it is necessary to calculate the object interest similarity between target object u and target object v, a first item quantity and a second item quantity are determined based on the object information, item information, and user preferences for items. The first item quantity is the number of items that users u and v both like, and the second item quantity is the number of items contained in the union of the sets of items liked by users u and v. The first item quantity and the second item quantity are divided to obtain the object interest similarity between users u and v. Alternatively, the object interest similarity between users u and v can be calculated based on the log-maximum likelihood estimation (Loglikelihood Ratio). Specifically, the quantities of the first, second, third, and fourth items are determined based on object information, item information, and user preferences for items. The first item quantity is the number of items that users u and v both prefer; the second item quantity is the number of items that user u specifically prefers; the third item quantity is the number of items that user v specifically prefers; and the fourth item quantity is the number of items that users u and v both dislike. For example, if all item information is {a, b, c, d, e, f}, user u's preferred items are {a, b, c}, user v's preferred items are {b, d}, and user u and user v both prefer item b, then the first item quantity is 1. Compared to user v, user u's specific preferences are {a, c}, resulting in a second item quantity of 2. Compared to user u, user v's particular preference is d, resulting in a third item quantity of 1. Items that neither user u nor user v likes are {e, f}, resulting in a fourth item quantity of 2. The row entropy is calculated based on the quantities of the first, second, third, and fourth items, as shown in formula (2). The column entropy is calculated based on the quantities of the first, second, third, and fourth items, as shown in formula (3). The matrix entropy is calculated based on the quantities of the first, second, third, and fourth items, as shown in formula (4). The object interest similarity is calculated based on the row entropy, column entropy, and matrix entropy to obtain the object interest similarity between different target objects, as shown in formula (5).

[0112]

[0113]

[0114]

[0115] UserSimilarity=2×(matrixEntropy-rowEntropy-columnEntropy) Formula (5)

[0116] Where, k 11 k represents the quantity of the first item. 12 k represents the quantity of the second item. 21 k represents the quantity of the third item. 22 denoted as the fourth item quantity, N as the total number of items, rowEntropy as the row entropy, columnEntropy as the column entropy, matrixEntropy as the matrix entropy, and UserSimilarity as the object interest similarity between user u and user v.

[0117] In step S520 of some embodiments, the second interest degree of the target object on the item is obtained based on the object interest similarity and the user's interest rating on the item. Specifically, based on the object interest similarity between the current user u and other users, the user information of other users is sorted from largest to smallest to obtain a sorting result. The top Q user information is selected from the sorting result to obtain the Q users most similar to the current user u, where Q is an integer greater than or equal to 1. The set of user information of these Q users is S(u,Q). The items liked by the user in S are extracted, and the items liked by user u are removed to obtain item j. The calculation method of the second interest degree of user u on the current item j is shown in formula (6).

[0118] P uj =∑ v∈S(u,Q)∩N(j) w uv r vj Formula (6)

[0119] Where N(j) is the set of users who like item j, S(u,Q) is the set of the Q users most similar to user u, and r vj Rate user v's interest in item j.

[0120] In step S530 of some embodiments, a priority queue or heap is used to sort the item information according to the second interest degree, and the top Q items with the highest second interest degree are recommended to user u to obtain a second recall list. Alternatively, the item information is sorted according to the second interest degree to obtain a sorting result, and the top Q items in the sorting result are used as the second recall list to recommend to the user.

[0121] Steps S510 to S530 above, by recommending items that similar users like to the user, can avoid the recommended items being completely dominated by the user's interests, thus improving the breadth of item recommendations.

[0122] Please see Figure 6 In some embodiments, step S140 may include, but is not limited to, steps S610 to S630:

[0123] Step S610: Perform data matching between the baseline recall list and the first recall list, and use the recall data that appears in both the baseline recall list and the first recall list as the matching recall data.

[0124] Step S620: Obtain the position information of the matched recall data in the first recall list;

[0125] Step S630: Calculate the predicted ranking of the first recall list based on the location information to obtain the first average predicted ranking.

[0126] In step S610 of some embodiments, if the baseline recall list is [1,2,3,4,5] and the first recall list is [3,5,6,7,8], data matching is performed on the baseline recall list and the first recall list, and the recall data that appears in both the baseline recall list and the first recall list, namely 3 and 5, are used as the matching recall data.

[0127] In step S620 of some embodiments, the position information of the matched recall data 3 in the first recall list is 1, and the position information of the matched recall data in the first recall list is 2.

[0128] In step S630 of some embodiments, the Mean Reciprocal Rank (MRR) is used as an evaluation function. The location information and the recall quantity of the recalled data in the baseline recall list are input into the evaluation function. The predicted ranking of the first recall list is calculated based on the evaluation function to obtain the first average predicted ranking. For example, based on location information 1, 2 and recall quantity 5, the first average predicted ranking of the first recall list is: MRR = (1 + 1 / 2) / 5 = 3 / 10.

[0129] Specifically, each recalled item in the baseline recall list is matched against the first recall list. If the i-th recalled item is the same as a recalled item in the baseline recall list, the position information P of the i-th recalled item in the first recall list is recorded. i The matching score of the i-th recall item is obtained as follows: If the i-th recall item is different from all the recall items in the baseline recall list, the matching score of the i-th recall item is 0. The matching scores of each recall item are added together and the scores are averaged to obtain the first average predicted ranking.

[0130] The calculation method for the first average predicted ranking is shown in formula (7).

[0131]

[0132] Where S represents the number of recalls, and Pi represents the location information of the matched recall data.

[0133] It should be noted that the calculation method for the second average predicted ranking is the same as that for the first average predicted ranking, and will not be repeated here.

[0134] The first average prediction ranking can be used to obtain the position of the recalled data in the baseline recall list, i.e., the user's demand item, in the first recall list. The higher the first average prediction ranking, the higher the position of the recalled data in the baseline recall list in the first recall list. Conversely, the higher the first average prediction ranking, the lower the position of the recalled data in the baseline recall list in the first recall list.

[0135] Through steps S610 to S630, the similarity between the first recall list and the baseline recall list can be quantitatively assessed based on the first average predicted ranking, thereby evaluating the recall results of the first recall list.

[0136] In step S150 of some embodiments, in order to evaluate the recall effect of the first recall list and the second recall list, the first average predicted ranking and the second average predicted ranking are compared.

[0137] In step S160 of some embodiments, if the first average predicted ranking is greater than the second average predicted ranking, it indicates that the similarity between the first recall list and the baseline recall list is greater than the similarity between the second recall list and the baseline recall list. The first recall list and the baseline recall list are more similar, and the earlier the recalled data in the baseline recall list appears in the first recall list, the better the recall result of the first recall list is compared to the second recall list. Therefore, the second recall list is filtered to obtain a third recall list. For example, the top K% of the recalled data in the second recall list are taken as the third recall list, where K can be 20. If the first average predicted ranking is lower than the second average predicted ranking, it means that the similarity between the first recall list and the baseline recall list is less than the similarity between the second recall list and the baseline recall list. The second recall list and the baseline recall list are more similar, and the more recall data in the baseline recall list appears earlier in the second recall list, the better the recall results of the second recall list are compared to the first recall list. Therefore, the first recall list is filtered, for example, the top K percent of the first recall list is taken as the third recall list, where K can be 20. If the first average predicted ranking equals the second average predicted ranking, then a target recommendation list is generated based on the first and second recall lists for item recommendation. The first and second average predicted rankings can dynamically adjust the weights of multiple recall results in the target recommendation list generation process, which improves the accuracy of item recommendation compared to a weighted average method.

[0138] Please see Figure 7 In some embodiments, step S170 may include, but is not limited to, steps S710 to S730:

[0139] Step S710: Perform data matching on the first recall list and the third recall list, and use the recall data that appears in both the first recall list and the third recall list as the first target recall data;

[0140] Step S720: Clean the first recall list based on the first target recall data to obtain the second target recall data;

[0141] Step S730: Rearrange the first target recall data and the second target recall data to generate a target recommendation list.

[0142] In step S710 of some embodiments, if the first average predicted ranking is greater than the second average predicted ranking, data matching is performed on the first recall list and the third recall list, and the recall data appearing in both the first and third recall lists is used as the first target recall data. If the first average predicted ranking is less than the second average predicted ranking, data matching is performed on the second and third recall lists, and the recall data appearing in both the second and third recall lists is used as the first target recall data. Specifically, if the first average predicted ranking is greater than the second average predicted ranking, the first recall list is traversed, and data matching is performed on the first and third recall lists. If a recall item in the first recall list appears in the third recall list, the recall item is added to the set to which the previous recall item that appeared in both the first and third recall lists belongs, thus obtaining the first target recall data. This avoids the problem that an item has good recall results in one recall list but poor recall results in another, resulting in a poor recall result after weighted averaging. It should be noted that the method for obtaining the first target recall data when the first average predicted ranking is lower than the second average predicted ranking is the same as the method for obtaining the first target recall data when the first average predicted ranking is higher than the second average predicted ranking, and will not be repeated here.

[0143] In step S720 of some embodiments, if the first average predicted ranking is greater than the second average predicted ranking, the first target recall data is removed from the first recall list, and the remaining recall data in the first recall list is used as the second target recall data. If the first average predicted ranking is less than the second average predicted ranking, the first target recall data is removed from the second recall list, and the remaining recall data in the second recall list is used as the second target recall data.

[0144] In step S730 of some embodiments, the first target recall data and the second target recall data are concatenated to generate a target recommendation list.

[0145] Taking the case where the first average predicted ranking is greater than the second average predicted ranking as an example, the process of generating the target recommendation list is described. Let the first recall list be denoted as array 'a', and the third recall list as array 'b'. Iterate through array 'a'. If an element is found to appear in both array 'b', it is considered to have higher importance, and its position is moved forward to after the previous element that appears in both arrays 'a' and 'b'. Create two arrays, 'one' and 'two'. Array 'one' records items that appear only in array 'a', and array 'two' records items that appear in both arrays 'a' and 'b'. Finally, concatenate arrays 'two' and 'one', with 'two' appearing first and 'one' appearing second, to obtain the target recommendation list.

[0146] It's important to note that because elements are continuously added to the end of arrays one and two as array a is traversed, the relative positions of items with the same frequency in each array remain unchanged. For example, if array a is [1,2,3,4,5,6] and array b is [2,5], the elements 2 and 5 appear twice during the traversal of array a from left to right, while the other elements appear only once. These once-appearing elements are added to array one, resulting in two = [2,5] and one = [1,3,4,6]. The relative order of elements in one and two is consistent with the original order of arrays a and b. The elements in array one remain unchanged relative to array a; only the element with a frequency of 2 is moved forward.

[0147] The data recall method in this application does not require reordering; it only involves left-shifting collinear elements, resulting in a time complexity of O(N). In contrast, the weighted average recall method requires reordering after calculation, leading to a time complexity of O(NlogN). Therefore, the data recall method in this application has reduced time complexity and can recommend items more efficiently. It should be noted that collinear elements refer to recalled data that appear in both recall lists simultaneously.

[0148] Through steps S710 to S730 above, a recommendation list can be generated based on the differences in the multi-path recall results, which improves the accuracy of item recommendations compared to the weighted average method.

[0149] Please see Figure 8 This application also provides a data retrieval device that can implement the above-described data retrieval method. The device includes:

[0150] The acquisition module 810 is used to acquire the baseline recall list and the data to be recalled;

[0151] The first recall module 820 is used to perform the first recall processing on the data to be recalled and obtain the first recall list.

[0152] The second recall module 830 is used to perform a second recall process on the data to be recalled and obtain a second recall list.

[0153] The ranking prediction module 840 is used to predict the ranking of the first recall list based on the benchmark recall list to obtain the first average predicted ranking of the first recall list, and to predict the ranking of the second recall list based on the benchmark recall list to obtain the second average predicted ranking of the second recall list.

[0154] Comparison module 850 is used to compare the first average predicted ranking and the second average predicted ranking.

[0155] The filtering module 860 is used to filter the second recall list to obtain a third recall list if the first average predicted ranking is greater than the second average predicted ranking.

[0156] The target recommendation list generation module 870 is used to generate a target recommendation list based on the first recall list and the third recall list; wherein, the target recommendation list is used for item recommendation.

[0157] The specific implementation of this data retrieval device is basically the same as the specific implementation of the data retrieval method described above, and will not be repeated here.

[0158] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described data retrieval method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0159] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0160] The processor 910 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0161] The memory 920 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 920 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 920 and is invoked by the processor 910 to execute the data retrieval method of the embodiments of this application.

[0162] The input / output interface 930 is used to implement information input and output;

[0163] The communication interface 940 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0164] Bus 950 transmits information between various components of the device (e.g., processor 910, memory 920, input / output interface 930, and communication interface 940);

[0165] The processor 910, memory 920, input / output interface 930 and communication interface 940 are connected to each other within the device via bus 950.

[0166] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described data retrieval method.

[0167] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0168] The data recall method, data recall device, electronic device, and computer-readable storage medium provided in this application embodiment obtain a baseline recall list and data to be recalled, perform a first recall process on the data to be recalled to obtain a first recall list, perform a second recall process on the data to be recalled to obtain a second recall list, perform a ranking prediction on the first recall list based on the baseline recall list to obtain a first average predicted ranking of the first recall list, and perform a ranking prediction on the second recall list based on the baseline recall list to obtain a second average predicted ranking of the second recall list. The similarity between the first recall list and the baseline recall list, and the similarity between the second recall list and the baseline recall list can be obtained through the ranking prediction. Furthermore, comparing the first average predicted ranking and the second average predicted ranking, if the first average predicted ranking is greater than the second average predicted ranking, it indicates that the first recall list and the baseline recall list are more similar than the second recall list and the baseline recall list. Since the first recall list and the baseline recall list are more similar, the second recall list is filtered to obtain the third recall list. By comparing the first average predicted ranking and the second average predicted ranking, the weight of the first and second recall lists in the target recommendation list can be dynamically adjusted based on their similarity to the baseline recall list. Compared to the weighted average calculation method, this approach takes into account the differences between different recall lists. Finally, a target recommendation list is generated based on the first and third recall lists. This target recommendation list is used for item recommendations. By considering the differences between different recall lists, the accuracy of the target recommendation list can be improved, thereby increasing the accuracy of item recommendations.

[0169] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0170] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0171] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0172] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0173] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0174] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0175] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0176] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0177] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0178] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0179] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A data retrieval method, characterized in that, The method includes: Obtain the baseline recall list and the data to be recalled; The data to be recalled is subjected to a first recall process to obtain a first recall list; The data to be recalled is subjected to a second recall process to obtain a second recall list; The first recall list is ranked and predicted based on the benchmark recall list to obtain the first average predicted ranking of the first recall list, and the second recall list is ranked and predicted based on the benchmark recall list to obtain the second average predicted ranking of the second recall list. Compare the first average predicted ranking with the second average predicted ranking; If the first average predicted ranking is greater than the second average predicted ranking, then the second recall list is filtered to obtain the third recall list. A target recommendation list is generated based on the first recall list and the third recall list; wherein, the target recommendation list is used for item recommendation; The data to be recalled includes object information of multiple target objects, item information of multiple items, and behavior information of the target objects towards the items. The first recall process is performed on the data to be recalled to obtain a first recall list, including: Based on the object information, the item information, and the behavior information, item similarity is calculated to obtain the item similarity between different items; based on the item similarity and the behavior information, the first interest level of the target object towards the item is obtained; based on the first interest level, the item information is sorted to obtain the first recall list; The second recall process is performed on the data to be recalled to obtain a second recall list, including: Based on the object information, the item information, and the behavior information, object interest similarity is calculated to obtain object interest similarity between different target objects; based on the object interest similarity and the behavior information, a second interest degree of the target object on the item is obtained; the item information is sorted according to the second interest degree to obtain the second recall list.

2. The data retrieval method according to claim 1, characterized in that, The process of obtaining the baseline recall list and the data to be recalled includes: Retrieve initial object behavior data within a preset time period from the preset data tracking event table; The initial object behavior data is deduplicated to obtain intermediate object behavior data; The intermediate object behavior data is classified and processed to obtain the baseline recall list and the data to be recalled.

3. The data retrieval method according to claim 2, characterized in that, The preset time period includes a first time point and a second time point. The process of classifying the intermediate object behavior data to obtain the baseline recall list and the data to be recalled includes: Calculate the number of intermediate object behavior data at the first time point; If the number of data is less than or equal to a preset threshold, then all intermediate object behavior data at the first time point will be used as the baseline recall list. If the number of data is greater than the number threshold, the intermediate object behavior data at the first time point is filtered to obtain the baseline recall list; The intermediate object behavior data at the second time point is used as the data to be recalled.

4. The data retrieval method according to any one of claims 1 to 3, characterized in that, The step of ranking and predicting the first recall list based on the baseline recall list to obtain the first average predicted ranking of the first recall list includes: Data matching is performed between the baseline recall list and the first recall list, and the recall data that appears in both the baseline recall list and the first recall list is used as the matched recall data; Obtain the position information of the matched recall data in the first recall list; The first average predicted ranking is obtained by calculating the predicted ranking of the first recall list based on the location information.

5. The data retrieval method according to any one of claims 1 to 3, characterized in that, The step of generating a target recommendation list based on the first recall list and the third recall list includes: Data matching is performed on the first recall list and the third recall list, and the recall data that appears in both the first recall list and the third recall list is taken as the first target recall data; The first recall list is cleaned based on the first target recall data to obtain the second target recall data. The first target recall data and the second target recall data are rearranged to generate the target recommendation list.

6. A data retrieval device, characterized in that, The device includes: The acquisition module is used to acquire the baseline recall list and the data to be recalled; the data to be recalled includes object information of multiple target objects, item information of multiple items, and behavior information of the target objects on the items; The first recall module is used to perform a first recall process on the data to be recalled to obtain a first recall list. The second recall module is used to perform a second recall process on the data to be recalled to obtain a second recall list. The ranking prediction module is used to predict the ranking of the first recall list based on the benchmark recall list to obtain the first average predicted ranking of the first recall list, and to predict the ranking of the second recall list based on the benchmark recall list to obtain the second average predicted ranking of the second recall list. The comparison module is used to compare the first average predicted ranking and the second average predicted ranking; The filtering module is used to filter the second recall list to obtain a third recall list if the first average predicted ranking is greater than the second average predicted ranking. The target recommendation list generation module is used to generate a target recommendation list based on the first recall list and the third recall list; wherein, the target recommendation list is used for item recommendation; The device is also used for: Based on the object information, the item information, and the behavior information, item similarity is calculated to obtain the item similarity between different items; based on the item similarity and the behavior information, the first interest level of the target object towards the item is obtained; based on the first interest level, the item information is sorted to obtain the first recall list; Based on the object information, the item information, and the behavior information, object interest similarity is calculated to obtain object interest similarity between different target objects; based on the object interest similarity and the behavior information, a second interest degree of the target object on the item is obtained; the item information is sorted according to the second interest degree to obtain the second recall list.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the data retrieval method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the data retrieval method according to any one of claims 1 to 5.

Images