Recommended method and device, equipment, storage medium

Abstract

The application provides a recommendation method and device, equipment and a storage medium. The method comprises the following steps: obtaining first behavior data of a target user; determining at least one first item to be recommended according to the first behavior data; determining a first preference of a second item associated with the first item; the first preference is determined based on usage data of at least one other user different from the target user for the second item; determining a first evaluation score of the second item; the first evaluation score is determined based on evaluation information received by the second item; and generating a first recommendation result according to at least the first preference, the first evaluation score of the second item associated with each first item and each first item. Thus, the preference analysis of the target user for the second item associated with the first item to be recommended is more comprehensive, the generated recommendation result is more accurate, and the user experience is improved.

Description

Technical Field

[0001] This application relates to communication technologies, including but not limited to recommended methods, apparatus, devices, and storage media. Background Technology

[0002] With the widespread adoption of 5G and increased network bandwidth, network technologies and application services are experiencing explosive growth. Faced with massive amounts of data, recommendation systems use recommendation algorithms to establish effective and direct connections between users and data information, analyze user demand trends, and organize the results into personalized lists for recommendation to users. While this can effectively alleviate the information overload problem caused by the excessive volume of data exceeding individual capacity, the accuracy of personalized recommendation results remains difficult to guarantee, thus affecting user experience. Summary of the Invention

[0003] In view of this, the recommendation method, apparatus, device, and storage medium provided in this application enable a more comprehensive analysis of the target user's preference for the second item associated with the first item to be recommended, thereby making the generated recommendation results more accurate and improving the user experience.

[0004] According to one aspect of the embodiments of this application, a recommendation method is provided, applied to the edge side, comprising: acquiring first behavioral data of a target user; determining at least one first item to be recommended based on the first behavioral data; determining a first preference of a second item associated with the first item; the first preference being determined based on usage data of the second item by at least one other user different from the target user; determining a first rating score of the second item; the first rating score being determined based on rating information received by the second item; and generating a first recommendation result based at least on the first preference, the first rating score, and each of the first items of the second item associated with each of the first items.

[0005] According to one aspect of the embodiments of this application, a recommendation method is provided, applied on the cloud side, comprising: receiving user behavior data of a target user sent by an edge device; determining at least one item to be recommended based on the user behavior data of the target user and user behavior data of at least one other user different from the target user; determining the usage preferences of associated items related to the item to be recommended; the usage preferences are determined based on usage data of the associated items by at least one other user different from the target user; determining a comprehensive evaluation score for the associated items; the comprehensive evaluation score is determined based on evaluation information received by the associated items; generating item recommendation results based at least on the usage preferences of the associated items, the comprehensive evaluation scores, and each item to be recommended; and sending the item recommendation results to the edge device.

[0006] Understandably, the first recommendation result is generated at least based on the first preference and first rating score of each second item associated with the first item. Thus, when recommending the second item associated with the first item to the target user, not only is the possibility that the target user has developed a preference for the second item because they have used it, but also the rating information received by the second item is taken into account. That is, the possibility that the target user and / or other users may develop a preference for the second item even if they have not used it is considered. This makes the analysis of the target user's preference for the second item more comprehensive, thereby making the generated recommendation result more accurate and improving the user experience.

[0007] According to one aspect of the embodiments of this application, a recommendation device is provided, applied at the edge, the recommendation device comprising: a first acquisition module, configured to acquire first behavioral data of a target user; a first determination module, configured to determine at least one first item to be recommended based on the first behavioral data; a second determination module, configured to determine a first preference of a second item associated with the first item; the first preference being determined based on usage data of the second item by at least one other user different from the target user; a first rating score of the second item being determined based on rating information received by the second item; and a first generation module, configured to generate a first recommendation result based at least on the first preference, the first rating score, and each of the first items of the second item associated with each of the first items.

[0008] According to one aspect of the embodiments of this application, a recommendation device is provided, applied on the cloud side, the recommendation device comprising: a first receiving module, configured to receive user behavior data of a target user sent by an edge device; a third determining module, configured to determine at least one item to be recommended based on the user behavior data of the target user and user behavior data of at least one other user different from the target user; a fourth determining module, configured to determine the usage preferences of associated items related to the item to be recommended; the usage preferences are determined based on usage data of the associated items by at least one other user different from the target user; and determine a comprehensive evaluation score for the associated items; the comprehensive evaluation score is determined based on evaluation information received by the associated items; a second generating module, configured to generate item recommendation results based at least on the usage preferences, comprehensive evaluation scores, and each item to be recommended of the associated items; and a first sending module, configured to send the item recommendation results to the edge device.

[0009] According to one aspect of the present application, an edge device is provided, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the program to implement the method described in the embodiments of the present application.

[0010] According to one aspect of the present application, a cloud device is provided, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the program to implement the method described in the embodiments of the present application.

[0011] According to one aspect of the embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the methods provided in the embodiments of this application.

[0012] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0013] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application. Obviously, the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0014] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0015] Figure 1 A schematic diagram illustrating the implementation flow of the recommended method provided in the embodiments of this application;

[0016] Figure 2 A schematic diagram illustrating the implementation flow of another recommended method provided in this application embodiment;

[0017] Figure 3 A schematic diagram illustrating the implementation flow of another recommended method provided in an embodiment of this application;

[0018] Figure 4 A schematic diagram illustrating the implementation process for determining the recommended score for the fourth item, provided in an embodiment of this application;

[0019] Figure 5 A schematic diagram illustrating the implementation flow of another recommended method provided in the embodiments of this application;

[0020] Figure 6 A flowchart illustrating the hybrid recommendation method provided in this application embodiment;

[0021] Figure 7 This is a schematic diagram of a hybrid recommendation process provided in an embodiment of this application;

[0022] Figure 8 A schematic diagram of the structure of a recommended device provided in an embodiment of this application;

[0023] Figure 9 A schematic diagram of another recommended device provided in the embodiments of this application;

[0024] Figure 10 This is a schematic diagram of the structure of an edge device provided in an embodiment of this application;

[0025] Figure 11 This is a schematic diagram of the structure of a cloud device provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of this application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.

[0027] 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.

[0028] In the following description, references to "some embodiments," "this embodiment," "this application embodiment," and examples, etc., describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subset of all possible embodiments and may be combined with each other without conflict.

[0029] Faced with the explosive growth of network technology and application services resulting in massive amounts of data, deploying recommendation systems in the cloud leverages the centralized computing model of cloud computing to establish an effective and direct connection between users and data information. This allows for the analysis of user demand trends, and the results are compiled into personalized lists and recommended to users in real time. This effectively alleviates a series of problems such as information overload caused by excessive data volume exceeding individual capacity. However, with the widespread adoption of 5G and increased network bandwidth, cloud-deployed systems are facing increasingly severe storage pressure. The centralized computing model of cloud computing also brings increasingly prominent issues such as load, maintenance, and failure risks to cloud systems. Therefore, it is necessary to introduce edge computing architecture, deploying terminal devices in the network edge environment closer to users, providing storage, computing, and network resources, and offloading some critical business applications to the network edge to alleviate data latency issues and improve service efficiency.

[0030] In this application embodiment, the edge side can be understood as an electronic device deployed on the edge side, or it can be described as an edge-side device, edge device, edge node, etc. This application embodiment does not limit it in this way.

[0031] In some embodiments, the edge device may be the target user terminal device or an electronic device close to the target user terminal device; this application does not limit this.

[0032] In this application embodiment, the cloud side can be understood as an electronic device deployed in the cloud or on the cloud side, and can also be described as a cloud-side device, cloud device, cloud, central node, network central node, etc. This application embodiment does not limit it in this way.

[0033] This application provides a recommended method. Figure 1 This is a schematic diagram illustrating the implementation flow of the recommended method provided in the embodiments of this application. The method is applied to the edge side, such as... Figure 1 As shown, the method may include the following steps 101 to 103:

[0034] Step 101: The edge device acquires the first behavioral data of the target user.

[0035] In some embodiments, the first behavioral data may be data generated based on the target user's browsing, clicking, or other behaviors, and the first behavioral data may take multiple forms.<Offset,(User,Item,Score)> The set that constitutes; among which,

[0036] Offset is the offset of the file position, which is used to standardize and correct the first line of data when it is acquired;

[0037] User is a unique identifier for the target user;

[0038] Item is a unique identifier for a particular item;

[0039] The score is the recommendation score for an item determined based on the target user's behavior.

[0040] The items involved in the embodiments of this application may be commodities, articles, videos, or music, etc., and the form of the items is not limited in the embodiments of this application.

[0041] Step 102: Based on the first behavior data, determine at least one first item to be recommended.

[0042] In some embodiments, at least one first item to be recommended can be determined based on the recommendation algorithm and the first behavior data of the target user. That is, by using the first behavior data of the target user on the item, a mathematical algorithm is used to filter out items that the target user is unlikely to engage in or dislike from a large number of items, thereby inferring at least one first item that the user may like. In the embodiments of this application, the type of recommendation algorithm is not limited.

[0043] Step 103, the edge device determines a first preference for a second item associated with the first item; the first preference is determined based on usage data of the second item by at least one other user different from the target user.

[0044] In some embodiments, the second project associated with the first project may be a project that appears simultaneously with the first project, or the second project associated with the first project may be a project that the user uses or purchases simultaneously with the first project.

[0045] For example, if a user purchases project A and project B at the same time, then the second project associated with the first project A is project B.

[0046] In some embodiments, user usage data for a project may include the number of times a user uses the project, the duration of use, the average interval between uses, the user's evaluation of the project after using it, and the user's rating of the project after using it.

[0047] In some embodiments, the higher the number of times a user uses an item or the longer the duration of use, the higher the item's usage score; the longer the average interval between user use of an item, the higher the item's usage score; the more positive or praising words (e.g., good-looking, beautiful, and delicious) a user uses in their evaluation of an item after using it, the higher the item's usage score; and the higher the user's rating of an item after using it, the higher the item's usage score.

[0048] In some embodiments, the first preference for the second item can be determined based on the average usage score of the second item by at least one other user different from the target user.

[0049] In some embodiments, the edge device may mine a first frequent itemset for each of the first items in the second row data through the following step 1031; wherein the first frequent itemset includes at least one first frequent association rule for the first items, and a strong association rule (not shown in the figure) is determined in the at least one first frequent association rule through steps 1032 to 1034.

[0050] In some embodiments, the strong association rule can be split to obtain at least one first sub-association rule corresponding to the first item; the first sub-association rule includes a first item and a second item associated with the first item; the second items associated in different first sub-association rules of the same first item are different.

[0051] Step 1031: The edge device mines the first frequent itemsets of each of the first items in the second row data according to the association analysis algorithm; wherein, the first frequent itemsets include at least one first frequent association rule of the first item, and the first frequent association rule is an association rule whose usage frequency is greater than a first threshold.

[0052] In some embodiments, the second behavioral data can be understood as behavioral data stored by the edge device, the second behavioral data including the first behavioral data of the target user; or, the second behavioral data including the first behavioral data of the target user and behavioral data of at least one other user different from the target user. It is understood that if the edge device is a target user terminal device, the edge device may store the first behavioral data of the target user; if the edge device is an electronic device close to the user terminal device, the edge device may store the first behavioral data of the target user and behavioral data of at least one other user different from the target user.

[0053] The so-called mining of the first frequent itemset of the first item can be understood as identifying association rules among the association rules that include the first item and whose usage frequency is greater than a first threshold; wherein, the association rule with a usage frequency greater than the first threshold is called the first frequent association rule, and the association rule can be understood as a set of at least one item, and the association rule that includes the first item can be a set of the first item and other items; other items include items that appear together with or at the same time as the first item.

[0054] In some embodiments, the first threshold may be preset by the edge device according to its own needs.

[0055] Example 1: The first item is product A. When a user buys A, they usually buy B, B and C, D, or B and D together. The association rules containing A are (AB), (ABC), (AD), and (ABD). The number of times a user buys AB together is 1, the number of times they buy ABC together is 3, the number of times they buy AD together is 2, and the number of times they buy ABD together is 3. The first threshold is 1. The first frequent association rules are (ABC), (AD), and (ABD). Accordingly, the first frequent itemset is [(ABC), (AD), (ABD)].

[0056] In some embodiments, at least one strong association rule in a first frequent association rule can be determined through the following steps 1032 to 1034.

[0057] Step 1032: Determine the third number of first frequent association rules in the first frequent itemset that include the j-th first frequent association rule; wherein j is greater than 0 and less than or equal to the total number of first frequent association rules in the first frequent itemset; j is a positive integer.

[0058] Example 2: As described in Example 1 above, the first frequent association rules are (ABC), (AD), and (ABD); the number of times ABC is purchased together is 3, the number of times AD is purchased together is 2, and the number of times ABD is purchased together is 3. Therefore, the third number of the first frequent association rule (ABC) is 3, the third number of the first frequent association rule (AD) is 2, and the third number of the first frequent association rule (ABD) is 3.

[0059] Step 1033: Based on the third number and the total number of first frequent association rules in the first frequent itemset, determine the proportion of the j-th first frequent association rule in the first frequent itemset.

[0060] Example 3: As described in Example 2 above, the total number of the first frequent association rules is 3+2+3=8; the proportion of the first frequent association rules (ABC) is 3 / 8, the proportion of (AD) is 2 / 8, and the proportion of (ABD) is 3 / 8.

[0061] Step 1034: The first frequent association rule with a proportion greater than or equal to the second threshold is taken as the first strong association rule. The first strong association rule is split to obtain at least one first sub-association rule corresponding to the first item. The first sub-association rule includes the first item and the second item associated with the first item. The second items associated in different first sub-association rules of the same first item are different.

[0062] In some embodiments, the second threshold may be preset by the edge device according to its own needs.

[0063] Example 4: As described in Example 3 above, if the second threshold is 3 / 8, then the first strong association rules are (ABC) and (ABD).

[0064] The first strong association rule is split into (ABC) and (ABD), resulting in (AB), (AC), and (AD); where the first item is A, and the second items are B, C, and D.

[0065] Step 104, the edge device determines a first evaluation score for the second item; the first evaluation score is determined based on the evaluation information received by the second item.

[0066] In some embodiments, the evaluation information received by the second item may be determined by the user without using the second item; for example, a user rating a movie without watching it, or a user rating a product without using it.

[0067] In some embodiments, the first rating score may be the average of the ratings given by at least one user to the second item without using the second item.

[0068] In some embodiments, the evaluation information received by the second project may be an estimated score set by the producer of the second project before its launch; for example, a manufacturer is about to launch a product P, and the manufacturer estimates the probability that the product P will be liked by at least one user, and sets a score for the product P based on the estimated probability.

[0069] Step 105: The edge device generates a first recommendation result based at least on the first preference, first evaluation score of the second item associated with each of the first items, and each of the first items.

[0070] In some embodiments, the recommended score of the second item is first determined based at least on the first preference of the second item and the first evaluation score; then the first recommendation result is generated based on the recommended scores of the second items associated with each of the first items and the recommended scores of each of the first items.

[0071] In some embodiments, the recommended score for the second item can be determined according to steps 202 to 207 of the following embodiments, which will not be described in detail here.

[0072] Understandably, the first recommendation result is generated at least based on the first preference and first rating score of each second item associated with the first item. Thus, when recommending the second item associated with the first item to the target user, not only is the case that the target user has developed a preference for the second item because they have used it, but also the rating information received by the second item is taken into account. That is, the case that the target user and / or other users may develop a preference for the second item even if they have not used it is taken into account, making the preference analysis of the second item more comprehensive, thereby making the generated recommendation result more accurate.

[0073] This application provides a recommended method. Figure 2 This is a schematic diagram illustrating the implementation flow of the recommended method provided in the embodiments of this application. The method is applied to the edge side, such as... Figure 2 As shown, the method may include the following steps 201 to 208:

[0074] Step 201: The edge device determines at least one first item to be recommended based on the first behavior data of the target user;

[0075] Step 202: The edge device determines a first preference for a second item associated with the first item;

[0076] Step 203: The edge device determines the first evaluation score for the second item;

[0077] Step 204: The edge device determines the target user's second preference for each of the first items associated with the second item, the second preference being determined based on the target user's usage data for the corresponding first item.

[0078] In some embodiments, the higher the number of times or the longer the duration of the target user's use of the first item, the larger the parameter value of the second preference; the longer the average interval between the user's use of the first item, the larger the parameter value of the second preference; the more words expressing affirmation or praise (e.g., good-looking, beautiful, and delicious) the user uses in their evaluation of the first item after using it, the larger the parameter value of the second preference; the higher the user's rating of the first item after using it, the larger the parameter value of the second preference.

[0079] In some embodiments, second items associated with other first items besides the first item are first determined according to steps 1031 to 1034 of the above embodiments; then, it is determined that the second items associated with other first items include each first item corresponding to the second item associated with the first item, thereby determining the target user's second preference for each first item associated with the second item.

[0080] Example 5: As described in Example 4 above, the second items associated with the first item A to be recommended include: B, C, and D. To determine the second preference for each of the first items associated with the second item B, then:

[0081] First, according to steps 1031 to 1034 of the above embodiment, the second items associated with the other first items to be recommended besides the first item A are determined; for example, the first items to be recommended include items A, R, and T; according to steps 1031 to 1034, the first frequent itemsets of R are determined to be (REF), (RG), (RH), and (RF), the first strong association rule of R is (REF), and correspondingly, the first sub-association rules of R are (RE) and (RF); the first frequent itemsets of T are (TBF), (TB), and (TBD), the first strong association rules of T are (TB) and (TBD), and correspondingly, the first sub-association rules of T are (TB) and (TD); then the second items associated with R are E and F, and the second items associated with T are B and D. The second items associated with each first item are shown in the following table:

[0082] Table 1. Second Items Associated with Each First Item

[0083]

[0084] Secondly, it is determined that the second items associated with other first items (R, T) include each first item (T) corresponding to the second items (B, D) associated with the first item (A).

[0085] Understandably, the second items associated with A are B, C, and D, the first item containing B is T, the first item containing D is T, therefore, the second items associated with the other first items besides A include the first item corresponding to the second item associated with A, which is T.

[0086] Finally, the target user's second preference for each of the first items (T and A) associated with the second item (B) is determined.

[0087] Step 205, the edge device determines a first recommendation degree of a first target sub-association rule including the second item; wherein, the first target sub-association rule includes the second item and a first item associated with the second item, and the first recommendation degree characterizes the degree of association between the second item and the first item associated with the second item.

[0088] In some embodiments, before determining the first recommendation degree of the first target sub-association rule including the second item, the first frequent itemsets of each first item are determined according to steps 1031 to 1034 of the above embodiment, and the first strong association rules in each first frequent itemset are split to obtain the first sub-association rule corresponding to each first item.

[0089] In some embodiments, the edge device filters out association rules containing the second item from the first sub-association rules corresponding to each first item, i.e., the first target sub-association rules.

[0090] Example 6: As described in Example 5 above, the first target sub-association rules including the second item (B) are (AB) and (TB), then it is necessary to determine the first recommendation degree of (AB) and (TB).

[0091] In some embodiments, determining the recommendation degree of the i-th first target sub-association rule (e.g., (AB)) including the second item includes the following steps 2051 to 2055 (not shown in the figure):

[0092] Step 2051: The edge device uses a first frequent itemset that includes each of the first items as a first target frequent itemset.

[0093] Example 7: As shown in Table 1 above, the first frequent itemsets of R are (REF), (RG), (RH) and (RF), the first frequent itemsets of T are (TBF), (TB) and (TBD), and the first frequent itemsets of A are (ABC), (AD) and (ABD). Then the first frequent itemsets of the target are (REF), (RG), (RH), (RF), (TBF), (TB) and (TBD), (ABC), (AD) and (ABD).

[0094] Step 2052: The edge device determines the first number of first frequent association rules in the first target frequent item set that include the i-th first target sub-association rule.

[0095] Example 8: As described in Example 7 above, the first frequent association rules containing (AB) are (ABC) and (ABD), so the first number is 2;

[0096] Step 2053: The edge device determines the first confidence level of the i-th first target sub-association rule based on the ratio of the first number to the total number of first frequent association rules in the first target frequent item set;

[0097] In some embodiments, the first confidence level of the i-th first target sub-association rule (e.g., (AB)) can be represented by the following equation (1), where i is a positive integer:

[0098]

[0099] Where conf(A→B) represents the confidence of the first sub-association rule (AB), Sup(A∪B) represents the first number, and Sup(A) represents the total number of the first frequent association rules in the first target frequent itemset.

[0100] Example 9: As described in Example 8 above, the first number is 2, and the total number of the first frequent association rules in the first target frequent itemset is 10. Then the first confidence level of (AB) is 1 / 5.

[0101] Step 2054: The edge device determines a second number of first frequent association rules that include the second item in the first target frequent item set;

[0102] In some embodiments, the second number may be the number of first frequent association rules that include the second item but not the first item.

[0103] As described in Example 7 above, the first frequent association rules that include the second item (B) are: (TBF), (TB), (TBD), (ABC), and (ABD), where the first frequent association rules that do not include A are: (TBF), (TB), and (TBD). Therefore, the second number is 3.

[0104] Step 2055: The edge device determines the first recommendation degree of the i-th first target sub-association rule based on the ratio of the first confidence degree of the i-th first target sub-association rule (e.g., (AB)) to the second number.

[0105] In some embodiments, the recommendation degree of the i-th first target sub-association rule (e.g., (AB)) is represented by the following formula (2):

[0106]

[0107] Where SRec(A→B) is the first recommendation degree of the i-th first target sub-association rule (e.g., (AB)), and Sup(B) represents the second recommendation degree.

[0108] Understandably, if the first confidence level of the first sub-association rule is higher (the higher the frequency of A and B appearing together), and the fewer times B appears alone, the higher the first recommendation level, indicating a strong association between A and B. Therefore, the association between the first item A and the second item B can be determined based on the first recommendation level.

[0109] Step 206: The edge device determines the second evaluation score of the second item based on the first recommendation degree of each of the first target sub-association rules, the second preference of the target user for each of the first items associated with the second item, and the first preference.

[0110] In some embodiments, the second evaluation score of the second item (e.g., B) can be represented by the following formula 3:

[0111]

[0112] Among them, SPre uB 1 is used to represent the second evaluation score for item B; I u R represents the set of all first-choice items to be recommended; uA This indicates the second preference for each of the first items associated with the second item; This indicates the first preference for the second item (B) associated with the first item;

[0113] Step 207: The edge device determines the recommended score for the second item based on the first evaluation score and the second evaluation score.

[0114] In some embodiments, the recommendation score for the second item (e.g., B) can be represented by the following formula 4:

[0115]

[0116] Among them, SPre uB 2 is used to represent the recommended score for item B; This indicates the first evaluation score for the second item (B).

[0117] Understandably, when determining the recommendation score for the second item (B) based on the first and second evaluation scores, not only is the user's preference for the second item (B) due to its use taken into account, but also the evaluation information received by the second item, that is, the user's preference for the second item even without using it is taken into account. This makes the preference analysis for the second item more comprehensive and the generated recommendation results more accurate.

[0118] Step 208: The edge device generates the first recommendation result based on the recommendation scores of the second items associated with each first item and the recommendation scores of each first item.

[0119] In some embodiments, the recommendation scores of each first item and the second items associated with each first item are sorted, and a first recommendation result is generated according to the recommendation scores.

[0120] In some embodiments, the first recommendation result may take the form of: multiple<User,List[(Item1:Prediction Score)]> This is a set consisting of: User, Item1, Prediction Score, and Prediction Score. User represents the displacement identifier of the target user; Item1 represents the name of the first or second item; and Prediction Score represents the recommendation score for the first or second item.

[0121] This application provides a recommended method. Figure 3 The following is a schematic diagram illustrating the implementation flow of the recommended method provided in the embodiments of this application, as shown below. Figure 3 As shown, the method may include the following steps 301 to 316:

[0122] Step 301: The edge device acquires the third behavioral data of the target user; wherein the generation time of the third behavioral data is after the generation time of the first behavioral data.

[0123] The generation time of the third line data is after the generation time of the first line data, which can be understood as: the third line data is updated compared to the first line data.

[0124] Step 302: The edge device updates the first recommendation result based on the third row of data to obtain the second recommendation result.

[0125] In some embodiments, firstly, at least one new first item to be recommended is determined based on the recommendation algorithm and third-row data, and second items associated with each new first item are determined based on steps 1031 to 1034 of the above embodiments; secondly, recommendation scores of the second items associated with each new first item are determined based on the third-row data and steps 202 to 208 of the above embodiments; finally, the order of the first items and / or the second items in the first recommendation list is updated according to the recommendation scores of each new first item and the recommendation scores of the second items associated with the new first items, or the first items and / or the second items in the first recommendation list are added or deleted to obtain a second recommendation result.

[0126] Step 303: The edge device sends the third line of data to the cloud device;

[0127] Step 304: The cloud device receives the third line of data;

[0128] Step 305: The cloud device determines at least one third item to be recommended based on the third behavioral data and behavioral data of at least one other user different from the target user;

[0129] It should be noted that the cloud device determines the at least one third item to be recommended based on the behavioral data of at least one user stored on the cloud device. The behavioral data of at least one user stored in the cloud device is much larger in volume and involves more users than the behavioral data of at least one user stored in the edge device. The cloud device stores behavioral data of almost all users, while the edge device only stores behavioral data of a portion of users.

[0130] Step 306, the cloud device determines a third preference for the fourth item associated with the third item; the third preference is determined based on usage data of the fourth item by at least one other user different from the target user.

[0131] In some embodiments, the third preference for the fourth item can be determined based on the average usage score of at least one other user different from the target user for the fourth item.

[0132] In some embodiments, the cloud device determines the second sub-association rule (not shown in the figure) for each third item through the following steps 3061 to 3064:

[0133] Step 3061: The cloud device mines a second frequent itemset for each of the third items in the third behavioral data and the behavioral data of at least one other user different from the target user, according to the association analysis algorithm; wherein, the second frequent itemset includes at least one second frequent association rule for the third item, and the second frequent association rule is an association rule whose usage frequency is greater than a third threshold.

[0134] In some embodiments, the third threshold may be preset by the cloud device according to its own needs.

[0135] It should be noted that the behavioral data of at least one user used to mine the second frequent itemset of each third item is the behavioral data of at least one user stored on a cloud device.

[0136] Step 3062: The cloud device determines the sixth number of second frequent association rules in the second frequent itemset that includes the m-th second frequent association rule; where m is greater than 0 and less than or equal to the total number of second frequent association rules in the second frequent itemset, and m is a positive integer;

[0137] Step 3063: The cloud device determines the proportion of the m-th second frequent association rule in the second frequent itemset based on the sixth number and the total number of second frequent association rules in the second frequent itemset;

[0138] Step 3064: The cloud device takes the second frequent association rule with a proportion greater than or equal to the fourth threshold as the second strong association rule, and splits the second strong association rule to obtain at least one second sub-association rule corresponding to the third item; the second sub-association rule includes the third item and a fourth item associated with the third item, and the fourth item associated in different second sub-association rules of the same third item is different.

[0139] In some embodiments, the third threshold may be preset by the cloud device according to its own needs.

[0140] Step 307: The cloud device determines the fifth evaluation score for the fourth item; the fifth evaluation score is determined based on the evaluation information received for the fourth item.

[0141] Step 308: The cloud device generates a third recommendation result based at least on the third preference, fifth evaluation score of the fourth item associated with each of the third items, and each of the third items.

[0142] In some embodiments, the recommendation score of the fourth item is first determined based at least on the third preference of the fourth item and the fifth evaluation score; then the third recommendation result is generated based on the recommendation scores of the fourth items associated with each of the third items and the recommendation scores of each of the third items.

[0143] In some embodiments, the recommended scores of the fourth items associated with each of the third items can be determined through steps 401 to 404 of the following embodiments, which will not be elaborated here.

[0144] Step 309: The cloud device sends the third recommendation result to the edge device;

[0145] Step 310: The edge device receives the third recommendation result returned by the cloud device; wherein the third recommendation result includes the recommendation scores of the third item and the fourth item associated with the third item;

[0146] Step 311: The edge device determines the interaction duration based on the sending time of the third line data and the receiving time of the third recommendation result.

[0147] In some embodiments, the interaction duration can be determined based on the timestamp when the cloud device sends the third action data and the timestamp when the edge device receives the third recommendation result.

[0148] Step 312: The edge device determines the time decay factor based on the interaction duration.

[0149] Understandably, edge devices send the latest third-party behavioral data of the target user to the cloud device so that the cloud device can generate a third-party recommendation result based on the third-party behavioral data and the full data stored in the cloud. However, because data transmission between the cloud device and the edge device takes time, the target user will still generate updated behavioral data after the edge device sends the third-party behavioral data to the cloud device. The cloud device cannot determine the third-party recommendation result based on the updated behavioral data, thus preventing it from sensing changes in user interest in real time. For example, after the edge device sends the third-party behavioral data to the cloud device, the cloud device returns the third-party recommendation result generated based on the third-party behavioral data and the full data after a period of time. During this period, the target user's interest in the fourth item in the third-party recommendation result may decrease. Since the recommendation score for the fourth item in the third-party recommendation result is determined based on the third-party behavioral data, not the updated behavioral data generated by the user in real time, it is necessary to decay the recommendation score of the fourth item according to the interaction time between the edge device and the cloud device.

[0150] In some embodiments, the time decay factor can be determined according to the following formula (5):

[0151]

[0152] Among them, |t ei -t ci | represents the interaction duration; α represents the weighting coefficient.

[0153] Step 313: The edge device re-determines the recommended score for the fourth item based on the time decay factor;

[0154] In some embodiments, the edge device first attenuates the third evaluation score of the fourth item according to the time decay factor to obtain a fourth evaluation score; then, based on the fourth evaluation score and the fifth evaluation score of the fourth item, it re-determines the recommended score of the fourth item.

[0155] In some embodiments, a third evaluation score for the fourth item can be determined first based on the second recommendation degree of each of the second target sub-association rules, the fourth preference of the target user for each of the third items associated with the fourth item, and the third preference. Then, the third evaluation score is attenuated to obtain a fourth evaluation score. Finally, the recommendation score for the fourth item is re-determined based on the fourth evaluation score and the fifth evaluation score of the fourth item.

[0156] In some embodiments, before determining the second recommendation degree, the cloud device can determine the second frequent itemset for each third item according to steps 3061 to 3064 above, and split the second strong association rule in each second frequent itemset to obtain the second sub-association rule corresponding to each third item. The edge device filters out the association rule containing the fourth item from the second sub-association rule corresponding to each third item, i.e., the second target sub-association rule.

[0157] In some embodiments, the third evaluation score can be attenuated according to the following formula (6) to obtain the fourth evaluation score:

[0158]

[0159] in, The third evaluation score for the fourth item, SPre u1B1new2 The fourth evaluation score used for the attenuated fourth item; I u1 R represents the set of all third items to be recommended; u1A1 This indicates a fourth preference for the third item associated with the fourth item; The third preference for the fourth item associated with the third item is represented by SRec(A1→B1), which is the second recommendation degree of the k-th second target sub-association rule.

[0160] In some embodiments, the fifth evaluation score of the fourth item can be determined according to the following formula (7):

[0161]

[0162] Among them, SPre u1B1new Used to represent the recommended score for the fourth item after decay; This represents the fifth evaluation score for the fourth item.

[0163] Understandably, due to the fifth evaluation score of the fourth item... It is unrelated to user behavior and does not change over time or with user behavior. The third preference, used to determine the third rating score, is different. and fourth preference All of these factors are related to user behavior. Therefore, when determining the recommended score for the fourth item, only the score for the third item was attenuated, while the score for the fifth item was not attenuated, thus ensuring the accuracy of the recommended score for the fourth item.

[0164] Step 314: The edge device updates the third recommendation result based on the re-determined recommendation score of the fourth item to obtain the fourth recommendation result.

[0165] Understandably, when determining the fourth recommendation result, the potential for changes in target user behavior during interaction time on cloud and edge devices, which could lead to a decrease in target user interest in the fourth item, was taken into account. Therefore, the third evaluation score of the fourth item was reduced, and the fourth recommendation result was determined based on the reduced recommendation score of the fourth item, making the determined fourth recommendation result more in line with the user's actual interest needs.

[0166] In some embodiments, updating the third recommendation result may involve reordering the third and / or fourth items in the third recommendation result according to the recommendation score, or adding or deleting the third and / or fourth items in the third recommendation result.

[0167] Step 315: The edge device generates a target recommendation result based on the second recommendation result and the fourth recommendation result.

[0168] In some embodiments, the first and fourth items in the second recommendation result can be mixed with the third and fourth items in the fourth recommendation result and then reordered according to the recommendation score. Alternatively, the mixed recommendation items can be added or deleted to obtain the target recommendation result. The fourth recommendation result generated in this way takes into account the situation that the target user's behavior changes over time, which may cause the user's interest in the fourth item to change. This ensures the real-time perception of user interest while determining the recommendation result.

[0169] Understandably, in determining the final target recommendation result, both the latest behavioral data of the edge device itself and the full data of the cloud device, as well as the recommendation score of the fourth item after decay, are taken into account, making the generated target recommendation result more comprehensive and accurate.

[0170] Step 316: The edge device recommends items based on the target recommendation results.

[0171] In some embodiments, it is first determined whether the number of all items included in the target recommendation result is greater than N. If it is greater than N, then the project recommendation is performed based on the target recommendation result; otherwise, at least one item with the highest usage frequency among all users in the cloud device is obtained and the project recommendation is performed.

[0172] This application provides a method for determining a fourth item's recommendation score, the method being applied on the cloud side. Figure 4 A schematic diagram illustrating the implementation process for determining the recommended score for the fourth item provided in this application embodiment is shown below. Figure 4 As shown, the method includes steps 401 to 404:

[0173] Step 401: The cloud device determines the target user's fourth preference for each of the third items associated with the fourth item; the fourth preference is determined based on the target user's usage data for the corresponding third item.

[0174] In some embodiments, the higher the number of times or the longer the duration of the target user's use of the third item, the larger the parameter value of the fourth preference; the longer the average interval between the user's use of the third item, the larger the parameter value of the fourth preference; the more words expressing affirmation or praise (e.g., good-looking, beautiful, and delicious) the user uses in their evaluation of the third item after using it, the larger the parameter value of the fourth preference; the higher the user's rating of the third item after using it, the larger the parameter value of the fourth preference.

[0175] Step 402: The cloud device determines a second recommendation degree for a second target sub-association rule that includes the fourth item; wherein the second target sub-association rule includes the fourth item and a third item associated with the fourth item, and the second recommendation degree characterizes the degree of association between the fourth item and the third item associated with the fourth item.

[0176] In some embodiments, the second recommendation degree of the k-th second target sub-association rule including the fourth item can be determined through steps 4021 to 4025, where k is a positive integer:

[0177] Step 4021: Determine a second target frequent itemset that includes each of the third items;

[0178] Step 4022: Determine the fourth number of second frequent association rules in the second target frequent item set that include the k-th second target sub-association rule;

[0179] Step 4023: Determine the second confidence level of the kth second target sub-association rule based on the ratio of the fourth number to the total number of second frequent association rules in the second target frequent item set;

[0180] Step 4024: Determine the fifth number of second frequent association rules that include the fourth item in the second target frequent item set;

[0181] Step 4025: Determine the second recommendation degree of the k-th second target sub-association rule based on the ratio of the second confidence degree of the k-th second target sub-association rule to the fifth number.

[0182] Step 403: Determine the third evaluation score of the fourth item based on the second recommendation degree of each of the target sub-association rules, the fourth preference of the target user for each of the third items associated with the fourth item, and the third preference.

[0183] Step 404: Determine the recommended score for the fourth item based on the fifth evaluation score and the third evaluation score.

[0184] In some embodiments, cloud devices are in multiple<Item i,List[(Item j,Score)]> The recommendation scores of the fourth items associated with each third item are recorded in the form of a set, and a recommendation score matrix as shown in Table 2 is generated. The recommendation score matrix is ​​sent to the edge device so that the edge device can generate the target recommendation result based on the recommendation scores of the fourth items associated with each third item.

[0185] Table 2. Recommendation rating matrix of each third item related to the fourth item.

[0186]

[0187] Where I1~I2 represent the third item, I a ~I b The fourth item is represented by "Score", which represents the recommended score for the fourth item in relation to the third item.

[0188] This application provides a recommended method. Figure 5 This is a schematic diagram illustrating the implementation flow of the recommended method provided in the embodiments of this application. The method is applied to the cloud side, such as... Figure 5 As shown, the method may include the following steps 501 to 506:

[0189] Step 501: Receive user behavior data of the target user sent by the edge device.

[0190] In some embodiments, the user behavior data of the target user may be the third behavior data.

[0191] Step 502: Based on the user behavior data of the target user and the user behavior data of at least one other user different from the target user, determine at least one item to be recommended.

[0192] In some embodiments, the item to be recommended may be the third item.

[0193] Step 503: Determine the usage preferences of associated projects related to the project to be recommended; the usage preferences are determined based on usage data of the associated projects by at least one other user different from the target user.

[0194] In some embodiments, the associated item may be a fourth item, and the usage preference may be a third preference.

[0195] Step 504: Determine the comprehensive evaluation score of the associated project; the comprehensive evaluation score is determined based on the evaluation information received by the associated project.

[0196] In some embodiments, the comprehensive evaluation score may be a fifth evaluation score.

[0197] Step 505: Generate project recommendation results based at least on the usage preferences, comprehensive evaluation scores, and individual projects associated with each project to be recommended.

[0198] In some embodiments, the project recommendation result may be a third recommendation result.

[0199] Step 506: Send the project recommendation results to the edge device.

[0200] With the development of cloud services, the scale of network technology and application services has experienced explosive growth. Faced with massive amounts of data, recommendation systems use recommendation algorithms to establish effective and direct connections between users and data information, analyze user demand trends, and organize the results into personalized lists for real-time recommendation to users. This can effectively alleviate a series of problems such as information overload caused by excessive data volume exceeding individual capacity. However, the popularization of 5G and the increase in network bandwidth are putting increasing pressure on the storage of cloud-deployed systems, and the centralized computing model of cloud computing is increasingly highlighting the problems of load, operation and maintenance, and failure risks brought to cloud systems. In particular, recommendation systems based on traditional cloud computing scenarios, that is, recommendation systems deployed in the cloud and generating recommendation results through centralized computing by network central nodes, suffer from problems such as network bandwidth limitations or latency between network central nodes and user terminal devices. Therefore, the real-time performance and accuracy of personalized recommendation results are difficult to guarantee, thus affecting user interaction and experience. Edge computing decomposes large services that were originally handled entirely by network central nodes. Through edge cloud platforms, computing and storage capabilities and business service capabilities can be migrated to edge nodes, alleviating data latency problems and improving service operation efficiency.

[0201] To address the challenges faced by recommendation systems in traditional cloud computing scenarios, researchers have adopted edge computing as a starting point. They propose building a real-time user behavior feature system on user terminal devices using user behavior data. This includes user exposure behavior features on items, such as exposure duration and scrolling speed; and user behavior features on the details page after clicking on an item, such as dwell time. A real-time user perception model is constructed on the device to determine changes in user preferences in real time based on user behavior features and to promptly reorder the information stream of items to be recommended on the user's terminal device. This reduces the asynchronous nature of user preference changes caused by the delay in acquiring user behavior data from the cloud and the timing of the recommendation system's perception of users and adjustments to recommended content, improving the accuracy of personalized recommendations and increasing user browsing and clicking interest.

[0202] The main drawbacks of related technologies are:

[0203] (1) Personalized recommendation technology based on edge cloud scenarios obtains an initial list of recommended items based on user historical behavior data stored in the cloud. However, it does not take into account the existing relationships between items, resulting in inaccurate initial data obtained by edge nodes, which in turn affects the recommendation results. Furthermore, there is a cold start problem for new users without behavioral data.

[0204] (2) Personalized recommendation technology based on real-time computing of edge nodes is limited by the resources contained in the edge nodes. The data used for recommendation algorithms is relatively scarce, and the personalized recommendation results generated in real time based on sparse data are also relatively simple, with a small coverage, and cannot meet user preferences in breadth.

[0205] (3) Parallel personalized recommendation technology based on cloud-edge collaboration generates recommendation results by combining user data returned from the cloud with the computing power of edge nodes. However, it does not take into account the changes in user interest caused by data transmission and latency issues when the cloud returns the results. This results in a certain error between the final recommendation results and the user's real-time preferences, which affects the user experience.

[0206] Based on this, the following will describe an exemplary application of the embodiments of this application in a practical application scenario.

[0207] This application provides a recommendation method based on an edge cloud platform, which can solve the technical problems and shortcomings that cannot be solved in related technical solutions. The technical problems solved include:

[0208] (1) Addressing the issues of inaccurate initial data and cold start, a strong association rule-based recommendation scoring method is proposed based on association rules. This method leverages the existing relationships between items to improve the accuracy of recommended items. Furthermore, for new users without behavioral data, initial recommendation results are generated by mixing data from popular items, thus enhancing the user experience.

[0209] (2) Addressing the issues of limited resources and data sparsity at edge nodes. Based on cloud-edge parallelism, the latest user behavior data for projects is transmitted to the cloud in real time. Recommendation results are generated using the full data stored in the cloud and combined with recommendation results generated from local data at edge nodes to create a richer personalized hybrid recommendation list for target users, thereby improving recommendation quality.

[0210] (3) Addressing the issue of user interest changes caused by data transmission and latency. A time decay factor is introduced to improve the strong association rule recommendation scoring calculation method proposed above, generating personalized recommendation results that better match the user's current interests over time, improving the prediction accuracy of the real-time user perception model on the device for user preferences, and reducing the impact of data transmission and latency issues on the recommendation results.

[0211] Based on this, the following will describe an exemplary application of the embodiments of this application in a practical application scenario.

[0212] This application provides a hybrid recommendation method based on an edge cloud platform. By utilizing the edge cloud platform, the processing of applications, data, and services, which was originally handled entirely by the central network node, is now collaboratively processed with logically edge nodes, improving operational efficiency. Simultaneously, through a cloud-edge parallel approach, association rules are introduced to perform association analysis on items based on traditional recommendation methods. Combined with a time decay factor, the recommendation results generated at both the cloud and edge ends are mixed in real time, improving the accuracy of personalized recommendation technology in edge cloud scenarios. The steps are as follows:

[0213] Step 1: Edge Device Acquisition<Offset,(User,Item,Score)> The first row of data is in the form of Offset, where Offset is the file position offset, User is the user's unique identifier, Item is the item's unique identifier, and Score is the rating score generated for the item based on user behavior. A recommendation algorithm is used to generate a list of items to be recommended, which includes at least one first item. A strong association rule-based recommendation scoring calculation method is used to perform association analysis on the at least one first item in the list of items to be recommended, and recommendation scores for the second items associated with each first item are calculated.<User,List[(Item:Prediction score)]> The first recommendation result is generated in the form of [format].

[0214] Step Two: Cloud-Edge Parallelism. At edge nodes, typically user terminal devices, a user behavior feature system is established to construct a real-time user perception model. Based on real-time user behavior features, changes in user preferences are determined. The initial recommendation results are reordered in real-time on the device using local data stored at the edge nodes, and the latest third-party behavior data is synchronously transmitted to the cloud. In the cloud, based on the third-party behavior data provided by the edge nodes, a third recommendation list is generated using the full dataset and a strong association rule recommendation scoring method incorporating a time decay factor.

[0215] Step 3: Hybrid Recommendation. The hybrid recommendation process includes: attenuating the third recommendation results returned from the cloud, mixing them with the second recommendation results generated in real time by edge nodes, and re-sorting them to generate more accurate target recommendation results, which are then recommended to the target user in a Top-N manner. For new users and cases where the number of items in the target recommendation results does not meet N, popular items are calculated to fill the recommendation item set in the target recommendation results, ensuring the quality of personalized recommendation results.

[0216] Before elaborating on the process in detail, let's first introduce three basic concepts related to the process:

[0217] Edge cloud platform: The edge cloud platform adopts a distributed computing architecture, offloading the processing of applications, data, and services previously handled by the central network node to logically located edge nodes. The closer proximity of edge nodes to users improves data processing and transmission speed, further reducing latency.

[0218] Recommendation Algorithm: By using the target user's historical behavior data on items, mathematical algorithms are used to filter out items that the user is unlikely to engage with from massive amounts of data, thereby inferring items that the user may like and generating a personalized list of items to recommend to the target user.

[0219] Association Rules: Association analysis algorithms are introduced to uncover the correlations between items, specifically calculating the probability that a user will also choose item B if they select item A. Based on this, a method for calculating the recommendation strength of strong association rules is constructed by combining itemset support and itemset confidence.

[0220] Figure 6 A flowchart illustrating the hybrid recommendation method provided in this application embodiment is shown below. Figure 6 As shown, the hybrid recommendation method includes steps 601 to 608:

[0221] Step 601: The edge device acquires the first behavioral data of the target user;

[0222] Step 602: Generate a list of items to be recommended based on the recommendation algorithm.

[0223] In some embodiments, the edge device generates a first list of items to be recommended based on a recommendation algorithm, wherein the list includes at least one first item to be recommended;

[0224] Step 603: The cloud-based system reorders and generates an initial list of recommended projects.

[0225] In some embodiments, the edge device determines the second items associated with each first item according to the association analysis algorithm and generates a first recommendation result; the cloud device rearranges each first item and the second items associated with each first item in the first recommendation result according to the full data and generates an initial recommendation list.

[0226] Step 604: The cloud device builds a real-time user perception model on the device.

[0227] In some embodiments, the cloud device builds a real-time user perception model on the end to perceive whether the user generates new behavioral data.

[0228] Step 605: The cloud device determines whether the user has generated third-party action data. If yes, proceed to step 606; otherwise, proceed to step 608.

[0229] Step 606: Real-time sorting of edge nodes to synchronize user data to the cloud.

[0230] In some embodiments, the edge node updates the first recommendation result based on the third row of data to obtain the second recommendation result, and then transmits the third row of data to the cloud.

[0231] Step 607: Generate hybrid recommendation results in parallel from cloud to edge.

[0232] In some embodiments, the cloud generates a third recommendation result based on the third row of data, performs a decay process on the recommendation score of the fourth item in the third recommendation result to obtain a fourth recommendation result, and then mixes it with the second recommendation result to obtain the target recommendation result.

[0233] Step 608: Popular items population, Top-N recommendations.

[0234] In some embodiments, the first recommendation result or the target recommendation result is populated with the most popular items based on the highest user interest, and the number of items in the populated recommendation result is greater than N.

[0235] In this embodiment, the cloud-edge parallelism is implemented, and personalized hybrid recommendation results are generated through correlation analysis and on-device real-time user perception model. The specific process is as follows:

[0236] First, the edge device generates a first recommendation result based on the first action data. After incorporating association analysis, the first recommendation result is reordered in the cloud to generate an initial recommendation list, which is then cached on the user's terminal device. Next, leveraging the computing power of the edge nodes, the recommendation system deployed on the user's terminal device acquires the user's third action data in real time, performs association analysis on the latest items generated by the user's action, and synchronizes the third action data to the cloud. Then, the edge nodes use the results of the association analysis to reorder the first recommendation result on the device in real time, obtaining a second recommendation result. The cloud then uses the third action data uploaded by the edge nodes to generate a third recommendation result and transmits it to the edge nodes. Finally, the edge nodes perform time decay processing on the third recommendation result returned from the cloud to obtain a fourth recommendation result, and combine it with the latest second recommendation result on the device to generate a final hybrid recommendation result (an example of a target recommendation result) to sort the items of interest to the target user on the device in real time. The cloud-edge parallel recommendation process can be divided into the following stages:

[0237] (1) In the first stage, the recommendation system of edge devices uses the first set of data collected to...<Offset,(User,Item,Score)> Recommendation algorithms are used to determine at least one first item. Then, association analysis algorithms are used to mine frequent itemsets of all items within the same category for each first item, and strong association rules that meet preset conditions are selected based on a pre-set first threshold.<List[(Item i,Item j,...)...]> Output in the form of. Through analysis, this application embodiment proposes a new recommendation score calculation method based on the introduction of correlation analysis between items for the recommendation method implemented on edge devices. Strong correlation rules can effectively represent the hidden relationships between items. Correlation analysis is performed through algorithms to obtain effective strong correlation rules. Then, the effective strong correlation rules are split into many-to-one or one-to-one row expressions. The specific splitting formula is shown in the following formula (8):

[0238]

[0239] According to the definition of association rules, the number of itemset support for a strong association rule is denoted as Sup, and the itemset confidence is denoted as Conf. The formula for calculating the recommendation degree of the i-th target sub-association rule derived from the algorithmic association analysis is shown in equation (2). It represents the ratio of the probability that the i-th target sub-association rule contains item A and also contains item B, to the probability that it contains item B but does not contain item A.

[0240] The recommendation score Rec of the i-th target sub-association rule obtained from the above association analysis is introduced, and based on this recommendation score, the user's preference for the second item is predicted, thereby improving the accuracy of personalized recommendations and enhancing the user experience to a certain extent. The calculation method for the score of the second item is shown in Equation (4):

[0241] Finally, the second items associated with each first item predicted by the strong association rule are mixed with the first items generated by the recommendation algorithm and reordered to...<User,List[(Item:Score)]> The output is displayed in the form of Top-N to the user terminal device.

[0242] The second phase involves building a real-time user perception model on the client side.<User,Item,Action]> The system acquires third-line data in real time. Using a cloud-edge parallel approach, edge nodes analyze the first recommendation result on the user's terminal device in real time based on cached local data and the user's latest third-line data, and then reorder the first recommendation result to obtain the second recommendation result.

[0243] Simultaneously, the third-row data is synchronously transmitted to the cloud, and the full amount of data stored in the cloud is used to continue using recommendation algorithms and association analysis to generate third-row recommendation results.<Item i,List[(Item j,Score)]> The recommendation rating matrix for each second item associated with the first item is generated in the form of [formula / method].

[0244] (3) In the third stage, a recommendation score calculation method based on a time decay factor is introduced to generate the fourth recommendation result. The time decay factor function is constructed based on the timestamp when the third recommendation result is returned from the cloud and the timestamp when the third recommendation result is transmitted to the edge node. The specific function is shown in Equation (5).

[0245] Analysis revealed that data transmission and latency issues between the cloud and edge nodes could delay the perception of user interests. Furthermore, items with similar time intervals better reflect user preferences than items with larger time intervals. Therefore, a time decay factor was introduced to attenuate the third recommendation result, resulting in the fourth recommendation result.

[0246] Finally, the fourth recommendation result, after undergoing time decay processing, is mixed and reordered with the latest second recommendation result from the edge nodes to generate the final target recommendation result.<Item i,List[(Item j,Score)]> In a hybrid format, recommendations are made to the target users.

[0247] In some embodiments, it is determined whether the number of items in the target recommendation result is greater than N. If it is greater than N, a Top-N recommendation strategy is used to generate a final recommendation list and recommend it to the target user; otherwise, the N items with the most frequent behavior in the user's historical behavior data, i.e., the most popular items among all items, are obtained.<Item,Score> The target recommendation results are filled in the form of .

[0248] Figure 7This is a schematic diagram of the hybrid recommendation process provided in the embodiments of this application, such as... Figure 7 As shown, hybrid recommendations can be implemented according to steps 701 to 709.

[0249] Step 701: Obtain the target recommendation results;

[0250] Step 702: Determine whether the number of target items in the target recommendation results is greater than N; if yes, proceed to step 703; otherwise, proceed to step 705.

[0251] In some embodiments, N is greater than 0;

[0252] Step 703: Determine the top N target items based on the recommended scores;

[0253] Step 704: Generate the final personalized recommendation list.

[0254] In some embodiments, a final personalized recommendation list is generated based on the top N target items.

[0255] Step 705, Obtain the first row of data<Offset,(User,Item,Score)> ;

[0256] Step 706: Obtain the recommended ratings for all items in the first row of data.<Item,List[Value]> ;

[0257] Step 707: Determine popular projects based on the recommended ratings of all projects.<Item,Score> The number of the popular projects is less than or equal to N;

[0258] Step 708: Populate the target recommendation results with popular items;

[0259] Step 709: Generate the final personalized recommendation list based on the target recommendation results and popular items.

[0260] This application proposes a hybrid recommendation method based on an edge cloud platform. It addresses the low recommendation accuracy issue in existing edge cloud platform personalized recommendation technologies caused by neglecting the relationships between items. By introducing association analysis, a strong association rule-based recommendation degree calculation method is constructed, thereby calculating predicted rating data that better matches the target user's preferences, improving the accuracy of personalized recommendation results and user experience.

[0261] (2) A method for calculating strong association rules with an improved time decay factor is proposed. Based on a cloud-edge parallel recommendation model, it generates richer and more personalized recommendation results that better match the user's current interests. Simultaneously, a hybrid recommendation method is used to apply different recommendation strategies to target users based on the number of items in the set of items to be recommended. This effectively ensures the quality of recommendation results when recommending users and reduces the impact of sparsity of user historical behavior data and cold start problems in the recommendation system.

[0262] This application proposes a recommendation scoring method based on association rules, constructed through association analysis. This method leverages the existing relationships between items to improve the accuracy of recommendations. Furthermore, for new users without behavioral data, initial recommendation results are generated by mixing data from popular items, thus enhancing the user experience.

[0263] This application's embodiment is based on a cloud-edge parallel recommendation model, which transmits the latest user behavior data for items to the cloud in real time. Recommendation results are generated using the full data stored in the cloud and combined with recommendation results generated from local data at edge nodes to create a richer, more personalized hybrid recommendation list for target users, thereby improving recommendation quality.

[0264] This application proposes to introduce a time decay factor to improve the strong association rule recommendation score calculation method mentioned above, generate personalized recommendation results that are more in line with the user's current interests over time, improve the prediction accuracy of the real-time user perception model on the device for user preferences, and reduce the impact of data transmission and latency issues on the recommendation results.

[0265] It is understood that in the embodiments of this application, user information, user behavior data and other related data are involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws and standards of the relevant countries and regions.

[0266] It should be noted that although the steps of the method in this application are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps; or steps from different embodiments may be combined into a new technical solution.

[0267] Based on the foregoing embodiments, this application provides a recommended device, which includes the included modules and the units included in each module, which can be implemented by a processor; of course, it can also be implemented by specific logic circuits; in the implementation process, the processor can be a central processing unit (CPU), microprocessor (MPU), digital signal processor (DSP) or field programmable gate array (FPGA), etc.

[0268] Figure 8 A schematic diagram of the structure of the recommended device provided in the embodiments of this application is shown below. Figure 8 As shown, the recommendation device 80 includes a first acquisition module 81, a first determination module 82, a second determination module 83, and a first generation module 84, wherein:

[0269] The first acquisition module 81 is used to acquire the first behavior data of the target user;

[0270] The first determining module 82 is used to determine at least one first item to be recommended based on the first behavior data;

[0271] The second determining module 83 is configured to determine a first preference for a second item associated with the first item; the first preference is determined based on usage data of the second item by at least one other user different from the target user; and to determine a first rating score for the second item; the first rating score is determined based on rating information received by the second item.

[0272] The first generation module 84 is configured to generate a first recommendation result based at least on the first preference, first evaluation score of the second item associated with each of the first items, and each of the first items.

[0273] In some embodiments, the recommendation device 80 further includes a fifth determining module, the fifth determining module being configured to determine the target user's second preference for each of the first items associated with the second item; the second preference being determined based on the target user's usage data of the corresponding first item; determine a first recommendation degree including a first target sub-association rule for the second item; wherein the first target sub-association rule includes the second item and the first items associated with the second item, and the first recommendation degree characterizes the degree of association between the second item and the first items associated with the second item; determine a second evaluation score for the second item based on the first recommendation degree of each of the first target sub-association rules, the target user's second preference for each of the first items associated with the second item, and the first preference; determine a recommendation score for the second item based on the first evaluation score and the second evaluation score; and the first generating module being configured to generate a first recommendation result based on the recommendation scores of the second items associated with each of the first items and the recommendation scores of each of the first items.

[0274] In some embodiments, the recommendation device 80 further includes a first mining module, which is configured to mine a first frequent itemset for each of the first items in the second behavioral data according to an association analysis algorithm; wherein the first frequent itemset includes at least one first frequent association rule for the first items.

[0275] In some embodiments, the recommendation device 80 further includes a first splitting module, which is used to determine strong association rules in the at least one first frequent association rules, split the strong association rules to obtain at least one first sub-association rule corresponding to the first item; the second items associated in different first sub-association rules of the same first item are different.

[0276] In some embodiments, the recommendation device 80 further includes a first update module, which is configured to acquire third behavioral data of the target user; update the first recommendation result based on the third behavioral data to obtain a second recommendation result; wherein the generation time of the third behavioral data is after the generation time of the first behavioral data.

[0277] In some embodiments, the recommendation device 80 further includes a second sending module and a second receiving module. The second sending module is used to send the third behavioral data to a cloud device, so that the cloud device determines at least one third item to be recommended based on the third behavioral data and behavioral data of at least one other user different from the target user. The second receiving module is used to receive a third recommendation result returned by the cloud device. The third recommendation result includes the third item and a recommendation score for a fourth item associated with the third item. The recommendation device 80 further includes a sixth determining module and a second updating module. The sixth determining module is used to determine the interaction duration based on the sending time of the third behavioral data and the receiving time of the third recommendation result; determine a time decay factor based on the interaction duration; and re-determine the recommendation score of the fourth item based on the time decay factor. The second updating module is used to update the third recommendation result based on the re-determined recommendation score of the fourth item to obtain a fourth recommendation result.

[0278] In some embodiments, the sixth determining module is further configured to attenuate the third evaluation score of the fourth item according to the time decay factor to obtain a fourth evaluation score; and to re-determine the recommended score of the fourth item according to the fourth evaluation score and the fifth evaluation score of the fourth item.

[0279] In some embodiments, the recommendation device 80 further includes a third generation module and a recommendation module. The third generation module is used to generate a target recommendation result based on the second recommendation result and the fourth recommendation result. The recommendation module is used to recommend items based on the target recommendation result.

[0280] Figure 9 This is a schematic diagram of the structure of the recommended device in the embodiments of this application, such as... Figure 9 As shown, the recommendation device 90 includes a first receiving module 91, a third determining module 92, a fourth determining module 93, a second generating module 94, and a first sending module 95, wherein:

[0281] The first receiving module 91 is used to receive user behavior data of the target user sent by the edge device;

[0282] The third determining module 92 is used to determine at least one item to be recommended based on the user behavior data of the target user and the user behavior data of at least one other user different from the target user;

[0283] The fourth determining module 93 is used to determine the usage preferences of associated projects related to the project to be recommended; the usage preferences are determined based on usage data of the associated projects by at least one other user different from the target user; and to determine the comprehensive evaluation score of the associated projects; the comprehensive evaluation score is determined based on the evaluation information received by the associated projects.

[0284] The second generation module 94 is used to generate project recommendation results based at least on the usage preferences, comprehensive evaluation scores, and each of the projects to be recommended.

[0285] The first sending module 95 is used to send the project recommendation results to the edge device.

[0286] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0287] It should be noted that, in the embodiments of this application... Figure 8 and Figure 9 The module division shown in the recommended device is illustrative and represents only one logical functional division; in actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, exist as separate physical units, or be integrated into one unit with two or more units. The integrated units can be implemented in hardware, as software functional units, or a combination of both.

[0288] It should be noted that, in the embodiments of this application, if the above-described methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause edge devices and cloud devices to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0289] This application provides an edge device. Figure 10 This is a schematic diagram of the hardware entity of the edge device according to an embodiment of this application, such as... Figure 10As shown, the edge device 100 includes a memory 1001 and a processor 1002. The memory 1001 stores a computer program that can run on the processor 1002. When the processor 1002 executes the program, it implements the steps in the method provided in the above embodiments.

[0290] It should be noted that the memory 1001 is configured to store instructions and applications executable by the processor 1002, and can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data and video communication data) in the processor 1002 and various modules in the edge device 100. It can be implemented by flash memory or random access memory (RAM).

[0291] This application provides a cloud device. Figure 11 This is a schematic diagram of the hardware entity of the cloud device according to an embodiment of this application, such as... Figure 11 As shown, the cloud device 110 includes a memory 1101 and a processor 1102. The memory 1101 stores a computer program that can run on the processor 1102. When the processor 1102 executes the program, it implements the steps in the method provided in the above embodiments.

[0292] It should be noted that the memory 1101 is configured to store instructions and applications executable by the processor 1102, and can also cache data to be processed or already processed in the various modules of the processor 1102 and the cloud device 110 (e.g., image data, audio data, voice communication data and video communication data), which can be implemented by flash memory or random access memory (RAM).

[0293] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the recommended method provided in the above embodiments.

[0294] This application provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the steps in the recommended method provided in the above-described method embodiments.

[0295] It should be noted that the descriptions of the storage medium and device embodiments above are similar to the descriptions of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium, storage medium, and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0296] It should be understood that the phrases "one embodiment," "an embodiment," or "some embodiments" mentioned throughout the specification mean that a specific feature, structure, or characteristic related to an embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment," "in one embodiment," or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The descriptions of the various embodiments above tend to emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, they will not be repeated here.

[0297] It is understood that in the embodiments of this application, user information, user data and other related data are involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0298] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three kinds of relationships. For example, object A and / or object B can represent three situations: object A exists alone, object A and object B exist simultaneously, and object B exists alone.

[0299] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0300] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or modules can be electrical, mechanical, or other forms.

[0301] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0302] In addition, each functional module in the various embodiments of this application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the integrated modules can be implemented in hardware or in the form of hardware plus software functional units.

[0303] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0304] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause edge devices and cloud devices to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0305] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0306] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0307] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0308] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A recommendation method, characterized in that, The method is applied to the edge side, and the method includes: Obtain the target user's initial behavioral data; Based on the first behavioral data, at least one first item to be recommended is determined; Determine a first preference for a second item associated with the first item; the first preference is determined based on usage data of the second item from at least one other user different from the target user. A first evaluation score is determined for the second item; the first evaluation score is determined based on the evaluation information received for the second item. Determine the target user's second preference for each of the first items associated with the second item; the second preference is determined based on the target user's usage data for the corresponding first item. A first recommendation degree is determined for a first target sub-association rule that includes the second item; wherein the first target sub-association rule includes the second item and a first item associated with the second item, and the first recommendation degree characterizes the degree of association between the second item and the first item associated with the second item; Based on the first recommendation degree of each of the first target sub-association rules, the second preference of the target user for each of the first items associated with the second item, and the first preference, the second evaluation score of the second item is determined. Based on the first evaluation score and the second evaluation score, determine the recommended score for the second item; A first recommendation result is generated based on the recommendation scores of the second items associated with each of the first items and the recommendation scores of each of the first items.

2. The method according to claim 1, characterized in that, Before determining the first recommendation degree of the first target sub-association rule that includes the second item, the method further includes: According to the association analysis algorithm, the first frequent itemsets of each of the first items are mined from the second row of data; wherein, the first frequent itemsets include at least one first frequent association rule of the first item; Determine the strong association rule among the at least one first frequent association rule, and split the strong association rule to obtain at least one first sub-association rule corresponding to the first item; the second items associated with different first sub-association rules of the same first item are different.

3. The method according to claim 2, characterized in that, After acquiring the first behavioral data, the method further includes: Obtain the third behavioral data of the target user; The first recommendation result is updated based on the third row of data to obtain the second recommendation result; wherein the third row of data was generated after the first row of data was generated. The third behavioral data is sent to a cloud device so that the cloud device can determine at least one third item to be recommended based on the third behavioral data and behavioral data of at least one other user different from the target user. Receive a third recommendation result returned by the cloud device; wherein the third recommendation result includes the recommendation scores of the third item and the fourth item associated with the third item; The interaction duration is determined based on the sending time of the third row of data and the receiving time of the third recommendation result; Determine the time decay factor based on the interaction duration; The recommended score for the fourth item is recalculated based on the time decay factor. The third recommendation result is updated based on the re-determined recommendation score of the fourth item to obtain the fourth recommendation result; Based on the second recommendation result and the fourth recommendation result, a target recommendation result is generated; Based on the target recommendation results, project recommendations are made.

4. The method according to claim 3, characterized in that, The step of re-determining the recommended score for the fourth item based on the time decay factor includes: The third evaluation score of the fourth item is attenuated according to the time decay factor to obtain the fourth evaluation score; Based on the fourth evaluation score and the fifth evaluation score of the fourth item, the recommended score for the fourth item is re-determined.

5. A recommended method, characterized in that, The method is applied to the cloud side, and the method includes: Receive third-party behavior data of the target user sent by the edge device; Based on the third behavioral data of the target user and the behavioral data of at least one other user different from the target user, determine at least one third item to be recommended; A third preference for a fourth item associated with the third item is determined; the third preference is determined based on usage data of the fourth item by at least one other user different from the target user. A fifth evaluation score is determined for the fourth item; the fifth evaluation score is determined based on the evaluation information received for the fourth item. Determine the target user's fourth preference for each of the third items associated with the fourth item; the fourth preference is determined based on the target user's usage data for the corresponding third item. A second recommendation degree is determined for the second target sub-association rule that includes the fourth item; wherein the second target sub-association rule includes the fourth item and a third item associated with the fourth item, and the second recommendation degree characterizes the degree of association between the fourth item and the third item associated with the fourth item; Based on the second recommendation degree of each of the second target sub-association rules, the target user's fourth preference for each of the third items associated with the fourth item, and the third preference, the third evaluation score of the fourth item is determined; The recommended score for the fourth item is determined based on the fifth evaluation score and the third evaluation score. A third recommendation result is generated based on the recommendation scores of the fourth items associated with each of the third items and the recommendation scores of each of the third items; The third recommendation result is sent to the edge device.

6. A recommendation device, characterized in that, Applied to the edge side, the recommended device includes: The first acquisition module is used to acquire the first behavioral data of the target user; The first determining module is used to determine at least one first item to be recommended based on the first behavior data; The second determining module is configured to determine a first preference for a second item associated with the first item; the first preference is determined based on usage data of the second item by at least one other user different from the target user; and to determine a first rating score for the second item; the first rating score is determined based on rating information received by the second item. A first generation module is configured to: determine the target user's second preference for each of the first items associated with the second item; the second preference is determined based on the target user's usage data of the corresponding first item; determine a first recommendation degree including a first target sub-association rule for the second item; wherein the first target sub-association rule includes the second item and the first items associated with the second item, and the first recommendation degree characterizes the degree of association between the second item and the first items associated with the second item; determine a second evaluation score for the second item based on the first recommendation degree of each of the first target sub-association rules, the target user's second preference for each of the first items associated with the second item, and the first preference; determine a recommendation score for the second item based on the first evaluation score and the second evaluation score; and generate a first recommendation result based on the recommendation scores of the second items associated with each of the first items and the recommendation scores of each of the first items.

7. A recommended device, characterized in that, For cloud-based applications, the recommended device includes: The first receiving module is used to receive the third behavior data of the target user sent by the edge device; The third determining module is used to determine at least one third item to be recommended based on the third behavior data of the target user and the behavior data of at least one other user different from the target user. The fourth determining module is used to determine a third preference for the fourth item associated with the third item; the third preference is determined based on usage data of the fourth item by at least one other user different from the target user; and to determine a fifth evaluation score for the fourth item; the fifth evaluation score is determined based on evaluation information received by the fourth item. The second generation module is configured to: determine the target user's fourth preference for each of the third items associated with the fourth item; the fourth preference is determined based on the target user's usage data of the corresponding third items; determine a second recommendation degree including the fourth item as a second target sub-association rule; wherein the second target sub-association rule includes the fourth item and the third items associated with the fourth item, and the second recommendation degree characterizes the degree of association between the fourth item and the third items associated with the fourth item; determine a third evaluation score for the fourth item based on the second recommendation degree of each of the second target sub-association rules, the target user's fourth preference for each of the third items associated with the fourth item, and the third preference; determine a recommendation score for the fourth item based on the fifth evaluation score and the third evaluation score; and generate a third recommendation result based on the recommendation scores of the fourth items associated with each of the third items and the recommendation scores of each of the third items. The first sending module is used to send the third recommendation result to the edge device.

8. An electronic device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 4, or the method according to claim 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 4, or when the computer program is executed by a processor, it implements the method as described in claim 5.

Images