Recall method and device for e-commerce recommendation system and readable medium
By performing full Swing computation in the e-commerce recommendation system and integrating it with historical similarity, the problems of high resource consumption and low efficiency in existing technologies are solved, and a highly efficient and stable recall effect is achieved.
Patent Information
- Application Number
- CN202411133958.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2026-03-03
AI Technical Summary
Existing Swing algorithms consume a lot of resources and are inefficient in e-commerce recommendation systems, affecting recall performance.
By performing a full Swing computation at task startup, only the Swing similarity of the previous day is calculated and merged with historical similarity to update the results for the next day. Hyperparameters are set to ensure the stability of the algorithm and data.
This greatly reduces resource consumption, improves recall efficiency, and ensures the non-destructive and stable nature of the recall.
Smart Images
Figure CN121599733A_ABST
Abstract
Description
Technical Field
[0001] This application generally relates to the field of recall technology for recommendation systems. More specifically, this application relates to a recall method, recall device, and non-transient machine-readable medium for e-commerce recommendation systems. Background Technology
[0002] The Swing algorithm is a recall algorithm. Swing is a relationship graph similar to a swing, where for each item, all users who clicked on it and all items clicked by those users form a user-item-user network structure. Swing represents the similarity relationships between items, which are propagated through user relationships.
[0003] The Swing algorithm is a user-item full-graph recall algorithm. Essentially, it's a memory-based paradigm within the i2i algorithm (a class of algorithms that recommend items based on similarity between items), belonging to the category of statistical, parameter-free algorithms. However, the full-graph Swing algorithm computation involves two data inflation processes. The first is traversing the user behavior sequence and constructing item pairs (e.g., ...).<i,j> Another time was for goods.<i,j> The product pairs that have been clicked by both users (e.g., u, v) are expanded. During these two data expansion processes, the two expansions consume memory and the two aggregations consume computation, making each full graph calculation both time-consuming and resource-intensive, thus affecting the recall effect.
[0004] In view of this, there is an urgent need to provide a recall solution for e-commerce recommendation systems in order to reduce resource consumption and improve recall efficiency. Summary of the Invention
[0005] In order to at least address one or more of the technical problems mentioned above, this application proposes a recall scheme for e-commerce recommendation systems in several aspects.
[0006] In a first aspect, this application provides a recall method for an e-commerce recommendation system, comprising: performing a full Swing calculation based on the full user behavior sequence in the e-commerce recommendation system to obtain a full Swing result; calculating a first Swing similarity of the user behavior sequence of the previous day based on the full Swing result; fusing the first Swing similarity and a second Swing similarity of the historical user behavior sequence to obtain a target Swing result; and updating the first Swing similarity of the previous day using the target Swing result, and using it as the historical Swing result for the next day, so as to realize the recall of the e-commerce recommendation system.
[0007] In a second aspect, this application provides a recall device for an e-commerce recommendation system, comprising: a processor; and a memory storing computer instructions for recall in the e-commerce recommendation system, wherein when the computer instructions are executed by the processor, the recall device implements the embodiments in the first aspect described above.
[0008] In a third aspect, this application provides a non-transient machine-readable medium having stored thereon computer program instructions for recall in an e-commerce recommendation system, which, when executed by one or more processors, cause the embodiments in the first aspect to be implemented.
[0009] The recall scheme for e-commerce recommendation systems provided above ensures that no product information is missed by performing a full Swing calculation based on the entire set of user behavior sequences, thus guaranteeing the lossless nature of the Swing algorithm. Compared to existing recall methods, this embodiment only needs to calculate the first Swing similarity of the user behavior sequences from the previous day, and then update the Swing result for the previous day by fusing the first Swing similarity with the second Swing similarity of the user behavior sequences. This eliminates the need to traverse all user behavior sequences and expand all product pairs each time, thereby significantly improving recall efficiency and reducing resource consumption. Furthermore, this embodiment also sets hyperparameters in the Swing calculation to ensure the stability of the algorithm and data during recall. Attached Figure Description
[0010] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein:
[0011] Figure 1 This is an exemplary schematic diagram illustrating a user-product interaction;
[0012] Figure 2 This is an exemplary flowchart illustrating a recall method for an e-commerce recommendation system according to an embodiment of this application;
[0013] Figure 3 This is an exemplary schematic diagram illustrating a recall method for an e-commerce recommendation system according to an embodiment of this application;
[0014] Figure 4 This is an overall exemplary flowchart illustrating a recall method for an e-commerce recommendation system according to an embodiment of this application;
[0015] Figure 5This is an exemplary structural block diagram of a recall device for an e-commerce recommendation system according to an embodiment of this application. Detailed Implementation
[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0017] It should be understood that the terms "comprising" and "including" used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0018] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.
[0019] As used in this specification and claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."
[0020] As described in the background technology above, Swing is a relationship graph similar to a swing. For example, suppose users u and v have both purchased the same item i; then, the three of them form a relationship graph similar to a swing. If users u and v have purchased both item i and item j, then the two items are considered to have some degree of similarity. That is, the similarity relationship between items is passed through user relationships. To measure the similarity between item i and item j, consider users u and v who have both purchased items i and j. The fewer items these two users have jointly purchased, the higher the similarity between items i and j. The similarity in Swing can be represented by the following formula:
[0021]
[0022] Among them, U i U represents the set of users who like product i. j Let I represent the set of users who like product j. u I represents the set of items that user u likes. v Let v represent the set of items liked by user v. The sum of the two sigma values in the formula represents selecting pairwise pairs of users from the set of users who like both items i and j. The 1 in the numerator represents a single user.<u,v> Yes. The intersection of Iu and Iv represents the similarity between users u and v, equal to the number of items clicked by both users. A higher value indicates greater overlap between the two users, thus its contribution should be reduced. α is a smoothing term to avoid a zero denominator; α can be a small positive number, such as 1. The following will combine... Figure 1 Describe in detail the similarity calculation method in Swing.
[0023] Figure 1 This is an exemplary schematic diagram illustrating a user-product interaction. (For example...) Figure 1 As shown in the diagram, assume A, B, and C represent three different users, and h, y, q, o, and x represent products. The diagram exemplifies that users A, B, and C all clicked on product h, and users A, B, and C also jointly clicked on product q. Furthermore, user C also clicked on product y, and user C also clicked on products o and x. As mentioned earlier, for each product, the user-item-user network structure formed by all users who clicked on it and all products clicked by these users is called a Swing. For example, for product h, users A and B both clicked on product q, therefore [A, q, B] forms a Swing structure. Similarly, [A, q, C] and [B, q, C] also each form a Swing structure.
[0024] In practical applications, when calculating Swing similarity, the first step is to construct product pairs, such as product pairs.<h,y> ,<h,q> ,<h,o> and<h,x> Next, based on the Swing structure, the number of items clicked by both users and the number of user pairs can be determined. The contribution of the joint clicks under each user pair can be summed to obtain the Swing similarity. That is, the Swing similarity is calculated based on the above formula (1). Among them, the smoothing coefficient α can be set to 1.
[0025] For example, to calculate the product pair<h,q> Taking the similarity between them as an example, since users only clicked q on [A,B], [B,C], and [A,C] respectively, the Swing structures of [A,B], [B,C], and [A,C] each contribute 1 / (1+1) to q.<h,q> The Swing similarity is: Similarly, it is also possible to obtain<h,y> ,<h,o> and<h,x> The similarity scores of each product are used to rank the products and achieve product recall.
[0026] As described above, the full-graph Swing algorithm requires traversing the user behavior sequence, assembling product pairs, and expanding product pairs that have been clicked by the user. This makes each full-graph calculation both time-consuming and resource-intensive, impacting recall performance.
[0027] Based on this, this application proposes a recall scheme for e-commerce recommendation systems. By calculating the full graph Swing only once when the task starts, and only calculating the Swing results of the previous day each time, resource consumption is greatly reduced and recall efficiency is improved.
[0028] The specific embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0029] Figure 2 This is an exemplary flowchart illustrating a recall method 200 for an e-commerce recommendation system according to an embodiment of this application. Figure 2 As shown, in step S201, a full Swing calculation is performed based on the full user behavior sequence in the e-commerce recommendation system to obtain the full Swing result. It can be understood that the full user behavior sequence in the aforementioned e-commerce recommendation system refers to all user behavior sequences in the database at the current moment, also known as full data. That is, a full Swing calculation is performed based on the full data. This is only calculated once when the task starts, as the initial version for recall. In one implementation scenario, a full Swing calculation can be performed based on the above formula (1) to obtain the full Swing result. In some embodiments, the aforementioned user behavior sequence is a list set formed when a user performs a target operation on a target product at different times, and the target operation includes at least a click operation. Furthermore, the aforementioned target operation may also include operations such as placing an order, adding to cart, or adding to favorites.
[0030] Next, in step S202, the first Swing similarity of the user behavior sequence of the previous day is calculated based on the full Swing results. In one embodiment, the time index of the previous day can first be determined from the full user behavior sequence, and the user behavior sequence after the time index can be extracted from the full user behavior sequence. Then, based on the user behavior sequence after the time index and the historical user behavior sequence, product pairs are constructed to calculate the Swing similarity of the corresponding product pairs, so as to obtain the first Swing similarity of the user behavior sequence of the previous day.
[0031] It is important to understand that user behavior sequences are recorded continuously, thus requiring the determination of the time index of the previous day. After determining the aforementioned time index, the user behavior sequence following the time index corresponds to the user behavior sequence of the previous day. Furthermore, the user behavior sequence of the previous day is extracted, and product pairs are constructed based on the user behavior sequence of the previous day and the historical user behavior sequences. Then, the first Swing similarity of the user behavior sequence of the previous day is calculated according to the above formula (1).
[0032] Based on the first Swing similarity obtained above, in step S203, the first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result. In one embodiment, a weighted summation operation is performed on the first Swing similarity and the second Swing similarity of the historical user behavior sequence to obtain the target Swing result. In some embodiments, the weights in the aforementioned weighted summation operation can be set according to actual task requirements.
[0033] Finally, in step S204, the first Swing similarity of the previous day is updated using the target Swing result, and this is used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system. That is, the fused Swing result overwrites the content of the previous day, serving as a new recall version for that day, and is used as the historical Swing result for the next day to achieve recall in the recommendation system. Correspondingly, the second Swing similarity of the historical user behavior sequence in the context of this application is determined by the historical Swing result of the previous day.
[0034] As described above, this embodiment calculates the full Swing based on the complete user behavior sequence, thus preserving full graph information and ensuring lossless recall. Furthermore, this embodiment only needs to calculate the Swing result of the previous day each time, and then merge it with historical Swing results (e.g., weighted summation) to update the content of the previous day, which serves as the historical Swing result for the next day. This significantly improves recall efficiency and reduces resource consumption.
[0035] In some embodiments, this application also sets hyperparameters in calculating the full Swing or the first Swing similarity. These hyperparameters include at least the update time and / or a time window for the user behavior sequence. This is because Swing results based on user behavior sequences within a fixed time window are better; therefore, by setting hyperparameters such as update time and / or a time window for the user behavior sequence, the stability of the algorithm and data during recall can be ensured.
[0036] Figure 3 This is an exemplary schematic diagram illustrating a recall method for an e-commerce recommendation system according to an embodiment of this application. It should be understood that... Figure 3 The above Figure 2 A specific embodiment of the recall method 200, therefore the above regarding Figure 2 The description also applies to Figure 3 .
[0037] like Figure 3 As shown, in step S301, the full user behavior sequence is obtained, that is, all user behavior sequences in the database at the current moment. Next, in step S302, a full Swing calculation is performed to obtain the full Swing result in step S303. As mentioned above, the full Swing result can be obtained by performing a full Swing calculation based on the above formula (1). Through full calculation, the entire graph information can be preserved, ensuring lossless recall.
[0038] Further, in step S304, the first Swing similarity (denoted as sim_t-1) of the user behavior sequence of the previous day is calculated, and in step S305, the second Swing similarity (denoted as sim_t-2) of the historical user behavior sequence is obtained. Specifically, for the first Swing similarity, the user behavior sequence of the previous day is first extracted from the full set of user behavior sequences, and then product pairs are constructed based on the user behavior sequence of the previous day and the historical user behavior sequences.<i,j> Then, the first Swing similarity sim_t-1 of the user behavior sequence of the previous day is calculated according to the above formula (1). The second Swing similarity sim_t-2 can be determined by the historical Swing results of the previous day.
[0039] After obtaining the first Swing similarity sim_t-1 and the second Swing similarity sim_t-2, in step S306, the first Swing similarity sim_t-1 and the second Swing similarity sim_t-2 are fused to obtain the target Swing result. In some embodiments, the aforementioned fusion can be achieved by weighted summation of the first Swing similarity sim_t-1 and the second Swing similarity sim_t-2. Specifically, the fusion can be based on weight_t-1*sim_t-1 + weight_t-2*sim_t-2 to obtain the target Swing result. Here, weight_t-1 and weight_t-2 represent weights, which can be set according to the actual task requirements.
[0040] Furthermore, in step S307, the first Swing similarity sim_t-1 of the previous day is updated using the target Swing result, and this is used as the historical Swing result for the next day, thus achieving recall in the recommendation system. Based on this, the embodiments of this application only need to calculate the Swing result of the previous day, without having to traverse all user behavior sequences and expand all product pairs every time, which greatly improves recall efficiency and reduces resource consumption.
[0041] Figure 4 This is an overall exemplary flowchart illustrating a recall method for an e-commerce recommendation system according to an embodiment of this application. Figure 4 As shown, in step S401, the full user behavior sequence is obtained, that is, all user behavior sequences in the database at the current moment. Next, in step S402, a full Swing calculation is performed to obtain the full Swing result in step S403.
[0042] Further, in step S404, the time index of the previous day is determined from the full user behavior sequence. Based on this time index, in step S405, the user behavior sequence after the time index is extracted, that is, the user behavior sequence of the previous day. In step S406, product pairs are constructed based on the user behavior sequence of the previous day and the historical user behavior sequences.<i,j> The Swing similarity is calculated using the formula (1) above, so that the first Swing similarity sim_t-1 of the user behavior sequence of the previous day is obtained at step S407. In addition, at step S408, the second Swing similarity sim_t-2 is obtained through the historical Swing results of the previous day.
[0043] Next, in step S409, the first Swing similarity sim_t-1 and the second Swing similarity sim_t-2 are fused according to weight_t-1*sim_t-1+weight_t-2*sim_t-2 to obtain the target Swing result. Based on the aforementioned target Swing result, in step S410, the first Swing similarity sim_t-1 of the previous day is updated using the target Swing result, and it is used as the historical Swing result for the next day to achieve recall of the recommendation system.
[0044] Figure 5 This is an exemplary structural block diagram of a recall device 500 for an e-commerce recommendation system according to an embodiment of this application. Figure 5 As shown, the recall device 500 of this application may include a processor 501 and a memory 502, wherein the processor 501 and the memory 502 communicate via a bus. The memory 502 stores program instructions for recall in an e-commerce recommendation system. When the program instructions are executed by the processor 501, they enable the implementation of the recall mechanism described above in conjunction with the appendix. Figures 2-4 The described method involves: performing a full Swing calculation based on the full user behavior sequence in the e-commerce recommendation system to obtain a full Swing result; calculating a first Swing similarity of the user behavior sequence of the previous day based on the full Swing result; fusing the first Swing similarity with a second Swing similarity of the historical user behavior sequence to obtain a target Swing result; and updating the first Swing similarity of the previous day using the target Swing result, which is then used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system.
[0045] Based on the foregoing description in conjunction with the accompanying drawings, those skilled in the art will understand that the embodiments of this application can also be implemented by software programs. Therefore, this application also provides a non-transient machine-readable medium. This non-transient machine-readable medium stores computer-readable instructions thereon for recall in an e-commerce recommendation system. When these computer-readable instructions are executed by one or more processors, they implement the embodiments of this application in conjunction with the accompanying drawings. Figures 2-4 The described recall method for e-commerce recommendation systems.
[0046] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0047] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.
[0048] The foregoing can be better understood in accordance with the following terms:
[0049] Clause A1, a recall method for an e-commerce recommendation system, comprising:
[0050] Full Swing computation is performed on the full user behavior sequence in the e-commerce recommendation system to obtain the full Swing result;
[0051] Calculate the first Swing similarity of the user behavior sequence of the previous day based on the full Swing results;
[0052] The first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result; and
[0053] The target Swing result is used to update the first Swing similarity of the previous day, and this is used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system.
[0054] Clause A2, the recall method according to Clause A1, wherein the user behavior sequence is a list set formed when the user performs a target operation on the target product at different times, and the target operation includes at least a click operation.
[0055] Clause A3, the recall method according to Clause A1, wherein the first Swing similarity is calculated by the following operations:
[0056] Determine the time index of the previous day from the full user behavior sequence, and extract the user behavior sequence after the time index from the full user behavior sequence;
[0057] Construct product pairs based on the user behavior sequence following the time index and the historical user behavior sequence; and
[0058] Calculate the Swing similarity of the corresponding product pairs to obtain the first Swing similarity of the user behavior sequence of the previous day.
[0059] Clause A4, the recall method described in Clause A3, wherein the second Swing similarity of the historical user behavior sequence is determined by the historical Swing results of the previous day.
[0060] Clause A5. The recall method described in Clause A4, wherein the first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result, includes:
[0061] The target Swing result is obtained by performing a weighted summation operation on the first Swing similarity and the second Swing similarity of the historical user behavior sequence.
[0062] Clause A6. The recall method described in Clause A1 further includes:
[0063] Set hyperparameters when calculating the full Swing or when calculating the similarity of the first Swing.
[0064] Clause A7, the recall method as described in Clause A6, wherein the hyperparameters include at least the update time and / or a time window for the user behavior sequence.
[0065] Clause A8. A recall device for an e-commerce recommendation system, comprising:
[0066] Processor; and
[0067] A memory stores computer instructions for recalling data in an e-commerce recommendation system. When these computer instructions are executed by a processor, the recall device performs the following operations:
[0068] Full Swing computation is performed on the full user behavior sequence in the e-commerce recommendation system to obtain the full Swing result;
[0069] Calculate the first Swing similarity of the user behavior sequence of the previous day based on the full Swing results;
[0070] The first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result; and
[0071] The target Swing result is used to update the first Swing similarity of the previous day, and this is used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system.
[0072] Clause A9, the recall device as described in Clause A8, wherein the user behavior sequence is a list set formed when the user performs a target operation on the target product at different times, and the target operation includes at least a click operation.
[0073] Clause A10, the recalled device as described in Clause A8, wherein the recalled device further calculates the first Swing similarity by:
[0074] Determine the time index of the previous day from the full user behavior sequence, and extract the user behavior sequence after the time index from the full user behavior sequence;
[0075] Construct product pairs based on the user behavior sequence following the time index and the historical user behavior sequence; and
[0076] Calculate the Swing similarity of the corresponding product pairs to obtain the first Swing similarity of the user behavior sequence of the previous day.
[0077] Clause A11, the recall device as described in Clause A10, wherein the second Swing similarity of the historical user behavior sequence is determined by the historical Swing results of the previous day.
[0078] Clause A12, pursuant to Clause A11, the recalled device further obtains the target Swing result by:
[0079] The target Swing result is obtained by performing a weighted summation operation on the first Swing similarity and the second Swing similarity of the historical user behavior sequence.
[0080] Clause A13. The recalled device as described in Clause A8, wherein the recalled device further performs the following operations:
[0081] Set hyperparameters when calculating the full Swing or when calculating the similarity of the first Swing.
[0082] Clause A14, the recalled device as described in Clause A13, wherein the hyperparameters include at least the update time and / or a time window for the user behavior sequence.
[0083] Clause A15. A non-transient machine-readable medium storing computer program instructions for recall in an e-commerce recommendation system, which, when executed by one or more processors, cause the following operations to be performed:
[0084] Full Swing computation is performed on the full user behavior sequence in the e-commerce recommendation system to obtain the full Swing result;
[0085] Calculate the first Swing similarity of the user behavior sequence of the previous day based on the full Swing results;
[0086] The first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result; and
[0087] The target Swing result is used to update the first Swing similarity of the previous day, and this is used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system.
[0088] Clause A16. A non-transient machine-readable medium as described in Clause A15, wherein the following operations are further implemented:
[0089] Determine the time index of the previous day from the full user behavior sequence, and extract the user behavior sequence after the time index from the full user behavior sequence;
[0090] Construct product pairs based on the user behavior sequence following the time index and the historical user behavior sequence; and
[0091] Calculate the Swing similarity of the corresponding product pairs to obtain the first Swing similarity of the user behavior sequence of the previous day.
[0092] Clause A17, a non-transient machine-readable medium as described in Clause A16, wherein the second Swing similarity of the historical user behavior sequence is determined by the historical Swing results of the previous day.
[0093] Clause A18, a non-transient machine-readable medium as described in Clause A17, further comprising the following operations:
[0094] The target Swing result is obtained by performing a weighted summation operation on the first Swing similarity and the second Swing similarity of the historical user behavior sequence.
[0095] Clause A19. A non-transient machine-readable medium as described in Clause A16, wherein the following operations are further implemented:
[0096] Set hyperparameters when calculating the full Swing or when calculating the similarity of the first Swing.
[0097] Clause A20, a non-transient machine-readable medium as described in Clause A19, wherein the hyperparameters include at least the update time and / or a time window of a sequence of user actions.
[0098] It should be understood that when the terms "first," "second," "third," and "fourth," etc., are used in the claims, specification, and drawings of this application, they are used only to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" as used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.
[0099] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.
[0100] Although the embodiments of this application are described above, the content is merely an example adopted for the purpose of facilitating understanding of this application and is not intended to limit the scope and application scenarios of this application. Any person skilled in the art described in this application may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.
[0101] Furthermore, the collection and acquisition of various data in this application comply with relevant laws and regulations and are authorized by the data providers. Any organization or individual that needs to obtain external data shall obtain authorization in accordance with the law and ensure data security, and shall not illegally collect, use, process, or transmit unauthorized or unprotected data, nor shall it illegally buy, sell, provide, or disclose unauthorized or unprotected data.
Claims
1. A recall method for an e-commerce recommendation system, comprising: Full Swing computation is performed on the full user behavior sequence in the e-commerce recommendation system to obtain the full Swing result; Calculate the first Swing similarity of the user behavior sequence of the previous day based on the full Swing results; The first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result; as well as The target Swing result is used to update the first Swing similarity of the previous day, and this is used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system.
2. The recall method according to claim 1, wherein the user behavior sequence is a list set formed when the user performs a target operation on the target product at different times, and the target operation includes at least a click operation.
3. The recall method according to claim 1, wherein the first Swing similarity is calculated by the following operation: Determine the time index of the previous day from the full user behavior sequence, and extract the user behavior sequence after the time index from the full user behavior sequence; Product pairs are constructed based on the user behavior sequence following the time index and the historical user behavior sequence. as well as Calculate the Swing similarity of the corresponding product pairs to obtain the first Swing similarity of the user behavior sequence of the previous day.
4. The recall method according to claim 3, wherein the second Swing similarity of the historical user behavior sequence is determined by the historical Swing results of the previous day.
5. The recall method according to claim 4, wherein fusing the first Swing similarity and the second Swing similarity of the historical user behavior sequence to obtain the target Swing result includes: The target Swing result is obtained by performing a weighted summation operation on the first Swing similarity and the second Swing similarity of the historical user behavior sequence.
6. The recall method according to claim 1, further comprising: Set hyperparameters when calculating the full Swing or when calculating the similarity of the first Swing.
7. The recall method according to claim 6, wherein the hyperparameters include at least the update time and / or a time window for the user behavior sequence.
8. A recall device for an e-commerce recommendation system, comprising: processor; as well as A memory stores computer instructions for recalling data in an e-commerce recommendation system. When these computer instructions are executed by a processor, the recall device performs the following operations: Full Swing computation is performed on the full user behavior sequence in the e-commerce recommendation system to obtain the full Swing result; Calculate the first Swing similarity of the user behavior sequence of the previous day based on the full Swing results; The first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result; as well as The target Swing result is used to update the first Swing similarity of the previous day, and this is used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system.
9. The recall device according to claim 8, wherein the recall device further calculates the first Swing similarity by the following operation: Determine the time index of the previous day from the full user behavior sequence, and extract the user behavior sequence after the time index from the full user behavior sequence; Product pairs are constructed based on the user behavior sequence following the time index and the historical user behavior sequence. as well as Calculate the Swing similarity of the corresponding product pairs to obtain the first Swing similarity of the user behavior sequence of the previous day.
10. A non-transient machine-readable medium storing computer program instructions for recall in an e-commerce recommendation system, the computer program instructions, when executed by one or more processors, causing the following operations to be performed: Full Swing computation is performed on the full user behavior sequence in the e-commerce recommendation system to obtain the full Swing result; Calculate the first Swing similarity of the user behavior sequence of the previous day based on the full Swing results; The first Swing similarity and the second Swing similarity of the historical user behavior sequence are fused to obtain the target Swing result; as well as The target Swing result is used to update the first Swing similarity of the previous day, and this is used as the historical Swing result for the next day to achieve recall in the e-commerce recommendation system.