A similarity dispersing service reordering method, device and electronic equipment
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI 2345 NETWORK TECH
- Filing Date
- 2023-05-08
- Publication Date
- 2026-05-19
AI Technical Summary
[0004]本申请提供一种相似度打散业务重排序方法、装置和电子设备，以解决现有技术中的排序效果不佳上述技术问题
[0027]本申请上述实施例的技术方案，可以基于推荐物料的标题向量信息，将算法推荐的各项指标的排序效率提升5％以上，提升效果显著。其中，上述指标如收入、点击率、arpu值等。之后，可以将物料进行更有针对性的个性化推荐。
Smart Images
Figure CN116610948B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of personalized recommendation technology, specifically to a method, apparatus, and electronic device for reordering business based on similarity scrambling. Background Technology
[0002] Vector feature information is a floating-point array consisting of 128 floating-point numbers, generated by processing reported user behavior or reported materials through a machine learning model. Common examples include material title vector features and user click sequence vector features.
[0003] The architecture of a recommendation system can be broadly divided into the following modules: recall, coarse ranking, fine ranking, and re-ranking. In the fine ranking stage, the goal is to obtain a globally optimal solution for the candidate ranking queue. However, in reality, the fine ranking model outputs a score for each user and each candidate item, and these candidate items also influence each other. Therefore, the re-ranking module mainly considers information related to the styles of items displayed on the same screen, using this information to re-rank the candidate item queue. Thus, the ranking model focuses on improving the effectiveness of fine ranking, while the subsequent re-ranking strategy handles the rest. Re-ranking algorithms or strategies are applied according to the actual business needs to ensure that the ranking results are not rigid, thereby achieving overall optimization. This approach of fine ranking first and then re-ranking allows the ranking model and specific business logic to be separated. Academia can focus on exploring better ranking models, while industry can design specific re-ranking strategies according to its own needs, decoupling the two to achieve overall efficiency improvements. Summary of the Invention
[0004] This application provides a method, apparatus, and electronic device for reordering similarity-based business processes to solve the aforementioned technical problem of poor sorting performance in the prior art.
[0005] According to one aspect of this application, one embodiment provides a similarity-based business reordering method, comprising:
[0006] In the recall pool mix1, a baseline material a1 is selected, and the other materials a2, b1, a3... after the baseline material a1 are selected as candidate materials, where the mix1 sequence is: [a1, a2, b1, a3...].
[0007] Calculate the vector similarity between the baseline material a1 and the adjacent candidate material a2. If the similarity between the baseline material a1 and the candidate material a2 is within the scrambling threshold range, the next candidate material b1 is selected sequentially to calculate the vector similarity until all candidate materials in the recall pool mix1 have been traversed. If the similarity between the baseline material a1 and the candidate material a2 is higher than the upper limit of the scrambling threshold range, the next material a3 is selected and compared with the baseline material a1 for similarity. Material a3 and the baseline material a1 come from the same content library. If the similarity between material a3 and the baseline material a1 is lower than the scrambling threshold range, the positions of material a3 and material a2 in the recall pool mix1 are swapped to form the mix2 sequence: [a1, a3, b1, a2...].
[0008] In the mix2 sequence, the next material a3 is selected as the reference material, and the similarity is calculated cyclically in the mix2 sequence; wherein, the cyclic similarity calculation includes calculating the vector similarity between the reference material a3 and the adjacent material b1 of the reference material a3;
[0009] The final sorting result is obtained when at least one content library in the recall pool mix1 is exhausted.
[0010] In one embodiment, after the positions of material a3 and material a2 are swapped, the cycle with material a1 as the reference material has been completed.
[0011] In one embodiment, the similarity is calculated as a cosine similarity calculation.
[0012] In one embodiment, the scattering threshold range is [75%, 95%].
[0013] In one embodiment, the similarity-based business reordering method further includes mixing multiple materials to form a recall pool mix1.
[0014] In one embodiment, the recall pool mix1 comes from at least two original recommended material recall pools; the content libraries of any two original recommended material recall pools are different.
[0015] In one embodiment, the original recommended material recall pool includes recall pool ar and recall pool br.
[0016] According to one aspect of this application, one embodiment provides a similarity-based business reordering apparatus, comprising:
[0017] The material selection module is used to select a baseline material a1 in the recall pool mix1. Other materials a2, b1, a3... after the baseline material a1 are selected are candidate materials, where the mix1 sequence is: [a1, a2, b1, a3...].
[0018] The first processing module is used to calculate the vector similarity between the reference material a1 and the adjacent candidate material a2. If the similarity between the reference material a1 and the candidate material a2 is within the scrambling threshold range, the next candidate material b1 is selected sequentially to calculate the vector similarity until all candidate materials in the recall pool mix1 are traversed. If the similarity between the reference material a1 and the candidate material a2 is higher than the upper limit of the scrambling threshold range, the next material a3 is selected and compared with the reference material a1 for similarity. Material a3 and the reference material a1 come from the same content library. If the similarity between material a3 and the reference material a1 is lower than the scrambling threshold range, the positions of material a3 and material a2 in the recall pool mix1 are swapped to form the mix2 sequence: [a1, a3, b1, a2...].
[0019] The second processing module is used to select the next material a3 in the mix2 sequence as the reference material and calculate the similarity in the mix2 sequence in a loop.
[0020] The loop processing module is used to repeatedly calculate similarity, which includes calculating the vector similarity between the reference material a3 and the adjacent material b1 of the reference material a3; until at least one material in the content library of the recall pool mix1 is exhausted.
[0021] According to one aspect of this application, one embodiment provides an electronic device, including:
[0022] Memory;
[0023] and
[0024] processor;
[0025] The memory is used to store one or more computer instructions; the one or more computer instructions are executed by the processor to implement the similarity shuffling and reordering method as described above.
[0026] According to one aspect of this application, one embodiment provides a readable storage medium storing computer instructions; wherein, when the computer instructions are executed by a processor, they implement the similarity-dispersing service reordering method as described in any of the above claims.
[0027] The technical solution of the above embodiments of this application can improve the ranking efficiency of various indicators of the algorithm recommendation by more than 5% based on the title vector information of the recommended materials, with a significant improvement effect. These indicators include revenue, click-through rate, and ARPU value. Subsequently, materials can be recommended in a more targeted and personalized manner. Attached Figure Description
[0028] Figure 1This is a flowchart of a similarity-based business reordering method in one embodiment;
[0029] Figure 2 This is a flowchart of a quantitative prediction method for sandstone in tidal channels, as described in one embodiment.
[0030] Figure 3 This is a schematic diagram of the structure of a similarity-based business reordering device in one embodiment. Detailed Implementation
[0031] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0032] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some, not all, of the embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort should fall within the scope of protection of the present application.
[0033] It should be noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0034] It should be understood that when an element (such as a layer, film, region, or substrate) is described as being "on" another element, the element may be directly on the other element, or there may be an intermediate element present. Moreover, in this application, when an element is described as being "connected" to another element, the element may be "directly connected" to the other element, or "connected" to the other element via a third element.
[0035] Example 1
[0036] Please refer to Figure 1 One embodiment provides a similarity-based business reordering method, including the following steps:
[0037] S1. Select a reference material a1 in the recall pool mix1. Select other materials a2, b1, a3... after the reference material a1 as candidate materials. The mix1 sequence is: [a1, a2, b1, a3...].
[0038] S2. Calculate the vector similarity between the reference material a1 and the adjacent candidate material a2. If the similarity between the reference material a1 and the candidate material a2 is within the scrambling threshold range, the next candidate material b1 is selected sequentially to calculate the vector similarity until all candidate materials in the recall pool mix1 have been traversed. If the similarity between the reference material a1 and the candidate material a2 is higher than the upper limit of the scrambling threshold range, the next material a3 is selected and compared with the reference material a1 for similarity. Material a3 and the reference material a1 come from the same content library. If the similarity between material a3 and the reference material a1 is lower than the scrambling threshold range, the positions of material a3 and material a2 in the recall pool mix1 are swapped to form the mix2 sequence: [a1, a3, b1, a2...]. At this point, the loop with material a1 as the reference material has been completed. In the above, the scrambling threshold range can be set to [75%, 95%].
[0039] S3. Select the next material a3 in the mix2 sequence as the reference material, and calculate the similarity in the mix2 sequence in a loop; wherein, the loop calculation of similarity includes calculating the vector similarity between the reference material a3 and the adjacent material b1 of the reference material a3;
[0040] S4. Continue until at least one content library in the recall pool mix1 is exhausted, and obtain the final sorting result.
[0041] There are multiple ways to calculate similarity. In one embodiment, the similarity is calculated using cosine similarity.
[0042] In one embodiment, the similarity-based business reordering method further includes mixing multiple materials to form a recall pool mix1.
[0043] In one embodiment, the recall pool mix1 comes from at least two original recommended material recall pools; the content libraries of any two original recommended material recall pools are different.
[0044] In one embodiment, the original recommended material recall pool includes recall pool ar and recall pool br. Preferably, the quantity of material in recall pool ar and recall pool br is equal.
[0045] Example 2
[0046] Please refer to Figure 2One embodiment provides a similarity-based business reordering method, which is a similarity-based business reordering strategy based on material title vector information. The specific steps are as follows:
[0047] 1. The original recommended material recall pool has a recall pool (ar): [a1,a2,a3,a4,a5] and a recall pool (br): [b1,b2,b3,b4,b5].
[0048] 2. Suppose that some materials are selected from recall pool a and recall pool b and mixed to form a new recall pool mix1, which then enters the reordering process. Suppose that the sequence of the new material recall pool mix1 after mixing is: [a1,a2,b1,a3,b2,a4,b3,b4,a5,b5];
[0049] 3. In mix1, calculate the cosine similarity between the vectors of a1 and a2; where a1 is called the reference material and a2 is called the candidate material.
[0050] (1) If the similarity calculated between vectors a1 and a2 is within the scattering threshold range (e.g., [75%, 95%)), then the candidate material is set as the next material b1 in mix1; if the similarity between vectors a1 and b1 is still within the scattering threshold range, then a3 is used as the candidate material; and so on, until mix1 is completely traversed.
[0051] (2) If the vector similarity between a1 and a2 is higher than the upper limit of the scattering threshold interval, then select the next-ranked material a3 from the recall pool that belongs to the same content library as a1 and compare its similarity with a1; if the similarity between a1 and a3 is lower than the scattering threshold interval, then swap the positions of a3 and a2 in mix1, and the sequence of mix1 becomes mix2: [a1,a3,b1,a2,b2,a4,b3,b4,a5,b5]
[0052] At this point, the cycle with a1 as the reference material has been completed; take the next material a3 in mix1 as the reference material and continue the cycle in step 3, that is, compare a3 and b1 in mix2.
[0053] 4. Until the materials in a certain content library are exhausted (for example, after all the materials in recall pool A are used up), the last-resort materials will be used to cover the remaining materials. The last-resort materials are independent of the material pool in the process and are generally selected by a random algorithm. The last-resort materials are prepared separately to avoid the situation where the recommendation results are empty and to ensure that the recommendation results always have data.
[0054] Example 3
[0055] Please refer to Figure 3One embodiment provides a similarity-based business reordering apparatus, comprising:
[0056] 1. Material Selection Module 10
[0057] Material selection module 10 is used to select a baseline material a1 in the recall pool mix1. Other materials a2, b1, a3... after the baseline material a1 are selected are candidate materials, where the mix1 sequence is: [a1, a2, b1, a3...].
[0058] 2. First processing module 20
[0059] The first processing module 20 is used to calculate the vector similarity between the reference material a1 and the adjacent candidate material a2. If the similarity between the reference material a1 and the candidate material a2 is within the scrambling threshold range, the next candidate material b1 is selected in sequence to calculate the vector similarity until all candidate materials in the recall pool mix1 are traversed. If the similarity between the reference material a1 and the candidate material a2 is higher than the upper limit of the scrambling threshold range, the next material a3 is selected to compare its similarity with the reference material a1. Material a3 and the reference material a1 come from the same content library. If the similarity between material a3 and the reference material a1 is lower than the scrambling threshold range, the positions of material a3 and material a2 in the recall pool mix1 are swapped to form the mix2 sequence: [a1, a3, b1, a2...].
[0060] 3. Second processing module 30
[0061] The second processing module 30 is used to select the next sequential material a3 as the reference material in the mix2 sequence and calculate the similarity in the mix2 sequence in a loop.
[0062] 4. Loop Processing Module 40
[0063] The loop processing module 40 is used to loop through similarity calculation, which includes calculating the vector similarity between the reference material a3 and the adjacent material b1 of the reference material a3; until at least one material in the content library of the recall pool mix1 is exhausted.
[0064] In one embodiment, the similarity-based dispersal and reordering device further includes a mixing module 50; the mixing module 50 is used to mix multiple materials to form a recall pool mix1. Generally, the recall pool mix1 comes from at least two original recommended material recall pools; the content libraries of any two original recommended material recall pools are different. Preferably, the original recommended material recall pools include recall pool ar and recall pool br.
[0065] Example 4
[0066] Based on the same inventive concept, one embodiment of this application provides an electronic device, including: a memory and a processor; wherein, the memory is used to store one or more computer instructions; the one or more computer instructions are executed by the processor using the similarity shuffling and reordering method described in any of the above embodiments.
[0067] Example 5
[0068] Based on the same inventive concept, one embodiment of this application provides a readable storage medium storing computer instructions; wherein, when the computer instructions are executed by a processor, the similarity shuffling and reordering method described in any of the above embodiments is implemented.
[0069] One or more of the aforementioned computer instructions can form a program.
[0070] The aforementioned program can run on a processor or be stored in memory (or computer-readable medium). Computer-readable medium includes both permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable medium does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0071] These computer programs may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes can be implemented using different modules, and different steps can be implemented using different modules.
[0072] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for reordering business processes based on similarity scrambling, characterized in that, include: Materials from at least two original recommended material recall pools are mixed to form a recall pool mix1, wherein any two original recommended material recall pools correspond to different content libraries, and the recall pool mix1 sequence is: [a1,a2,b1,a3……], where a1, a2, and a3 come from the same content library, and b1 comes from another content library; In the recall pool mix1, a baseline material a1 is selected, and the other materials a2, b1, a3... after the baseline material a1 are selected as candidate materials, where the mix1 sequence is: [a1, a2, b1, a3...]. Calculate the vector similarity between the baseline material a1 and the adjacent candidate material a2. If the similarity between the baseline material a1 and the candidate material a2 is within the scrambling threshold range, the next candidate material b1 is selected sequentially to calculate the vector similarity until all candidate materials in the recall pool mix1 have been traversed. If the similarity between the baseline material a1 and the candidate material a2 is higher than the upper limit of the scrambling threshold range, the next material a3 is selected and compared with the baseline material a1 for similarity. Material a3 and the baseline material a1 come from the same content library. If the similarity between material a3 and the baseline material a1 is lower than the scrambling threshold range, the positions of material a3 and material a2 in the recall pool mix1 are swapped to form the mix2 sequence: [a1, a3, b1, a2...]. In the mix2 sequence, the next material a3 is selected as the reference material, and the similarity is calculated cyclically in the mix2 sequence; wherein, the cyclic similarity calculation includes calculating the vector similarity between the reference material a3 and the adjacent material b1 of the reference material a3; The final sorting result is obtained when at least one content library in the recall pool mix1 is exhausted.
2. The similarity-based business reordering method according to claim 1, characterized in that, After the positions of materials a3 and a2 are swapped, the cycle with material a1 as the reference material is completed.
3. The similarity-based business reordering method according to claim 1, characterized in that, The similarity is calculated using cosine similarity.
4. The similarity-based business reordering method according to claim 1, characterized in that, The scattering threshold range is [75%, 95%).
5. A similarity-based business reordering method according to any one of claims 1-4, characterized in that, It also includes mixing multiple materials to form a recall pool mix1.
6. The similarity-based business reordering method according to claim 1, characterized in that, The original recommended material recall pool includes recall pool ar and recall pool br.
7. A similarity-based business reordering device, characterized in that, include: The mixing module is used to mix materials from at least two original recommended material recall pools to form a recall pool mix1, wherein any two original recommended material recall pools correspond to different content libraries, and the recall pool mix1 sequence is: [a1,a2,b1,a3……], where a1, a2, and a3 come from the same content library, and b1 comes from another content library; The material selection module is used to select a baseline material a1 in the recall pool mix1. Other materials a2, b1, a3... after the baseline material a1 are selected are candidate materials, where the mix1 sequence is: [a1, a2, b1, a3...]. The first processing module is used to calculate the vector similarity between the reference material a1 and the adjacent candidate material a2. If the similarity between the reference material a1 and the candidate material a2 is within the scrambling threshold range, the next candidate material b1 is selected sequentially to calculate the vector similarity until all candidate materials in the recall pool mix1 are traversed. If the similarity between the reference material a1 and the candidate material a2 is higher than the upper limit of the scrambling threshold range, the next material a3 is selected and compared with the reference material a1 for similarity. Material a3 and the reference material a1 come from the same content library. If the similarity between material a3 and the reference material a1 is lower than the scrambling threshold range, the positions of material a3 and material a2 in the recall pool mix1 are swapped to form the mix2 sequence: [a1, a3, b1, a2...]. The second processing module is used to select the next material a3 in the mix2 sequence as the reference material and calculate the similarity in the mix2 sequence in a loop. The loop processing module is used to repeatedly calculate similarity, which includes calculating the vector similarity between the reference material a3 and the adjacent material b1 of the reference material a3; until at least one material in the content library of the recall pool mix1 is exhausted.
8. An electronic device, characterized in that, include: Memory; and processor; The memory is used to store one or more computer instructions; the one or more computer instructions are executed by the processor to implement the similarity shuffling and reordering method according to any one of claims 1 to 6.
9. A readable storage medium, characterized in that, The readable storage medium stores computer instructions; wherein, when the computer instructions are executed by a processor, they implement the similarity shuffling and reordering method according to any one of claims 1 to 6.