Multi-service-line product rearrangement method, system and device, medium and program product

Abstract

The invention provides a multi-service-line product rearrangement method, system and device, a medium and a program product. The method comprises the steps of obtaining a correlation score of a user and each product; obtaining a product queue of each classification from high to low according to the correlation scores; obtaining a similarity score and a flow regulation score between the service line products in the sliding window; according to the correlation score, the similarity score and the flow regulation score, obtaining a rearrangement score of the head-of-queue product; obtaining a queue head product with the highest rearrangement score as a target product placed at the target position; adding the target product into the sorted product set and popping up, and moving the sliding window by one position according to a preset direction; and obtaining a product rearrangement table according to the sorted product set obtained after the preset number of times of loop execution. According to the invention, a set of reliable and universal rearrangement framework is designed, the correlation between users and products and the similarity between business line products are balanced, flow regulation and control are compatible, and the rationality and diversity of rearrangement results are ensured.

Description

Technical Field

[0001] The present disclosure relates to the technical field of product rearrangement, and particularly to a rearrangement method, system, device, medium and program product for multi-business line products. Background Art

[0002] In a search and recommendation system, the common system link is usually divided into a recall layer, a rough ranking layer, a filtering layer, a fine ranking layer, and a rearrangement layer. Among them, the fine ranking layer usually performs local optimization of products by using the single-point relationship between users and products. The rearrangement layer will make a secondary adjustment to the results of the fine ranking based on the global optimum and relevant business strategies, and usually takes into account the function of improving the user experience. Therefore, the rearrangement layer, as the module closest to the client at the end, usually plays an indispensable role. Some search and recommendation services involve multi-business line ranking, and there are significant differences between business lines. Simple rearrangement strategies are difficult to meet problems such as relevance, diversity, traffic skew, and cold start in small scenarios. For example, the search system in the travel scenario involves products from different business lines such as group tours, one-day tours, tickets, customized cases, and guide chartered tours. How to reasonably rank these products with huge differences in volume, unit price, play mode, etc., and maximize the total platform transaction volume on this basis has become a difficult problem. To solve various problems in multi-business line rearrangement, generally there are the following two solutions: 1) Adopt a single rearrangement or a combination of multiple rearrangement strategies; 2) Design a reliable and general integrated rearrangement algorithm framework to meet the requirements through a set of algorithms.

[0003] However, adopting a single rearrangement strategy, such as the commonly used linear weighting, interleaved sorting, and classification weight reduction in the industry, often can only solve one type of problem, and the benefit is minimal for the overall effect. The combination of multiple strategies can significantly improve this problem, but there may be conflicts between different strategies. Simple combination is likely to cause the phenomenon of attending to one thing and losing another, and maintaining multiple strategies in the long term also requires a large amount of human cost. Summary of the Invention

[0004] The technical problem to be solved by the present disclosure is to overcome the defect that the existing multi-business line rearrangement method cannot simultaneously meet relevance, diversity, and traffic regulation in an integrated manner, and provide a rearrangement method, system, device, medium and program product for multi-business line products.

[0005] The present disclosure solves the above technical problem through the following technical solutions:

[0006] In the first aspect of the present disclosure, a rearrangement method for multi-business line products is provided, and the rearrangement method includes:

[0007] S1. Obtain the relevance scores between the user and each product in the product set to be rearranged;

[0008] S2. Classify all products according to business lines, and obtain the product queue of each classification according to the sorting result from high to low of the correlation scores;

[0009] S3. Obtain the similarity scores between the products of each business line within the sliding window;

[0010] S4. Obtain the traffic regulation score;

[0011] S5. Obtain the head product from the product queue, and obtain the rearrangement scores of the head products in each business line according to the correlation scores, the similarity scores, and the traffic regulation scores;

[0012] S6. Obtain the head product with the highest rearrangement score from the rearrangement scores of the head products in each business line as the target product to be placed at the target position;

[0013] S7. Add the target product to the sorted product set, pop up the target product, and move the sliding window one position in the preset direction;

[0014] S8. Obtain the product rearrangement list according to the sorted product set obtained after executing S3 - S7 in a loop for a preset number of times.

[0015] Preferably, before step S2, the rearrangement method further includes:

[0016] Obtain the buried point data of the user;

[0017] Obtain the benchmark traffic of each business line according to the buried point data.

[0018] Preferably, step S4 includes:

[0019] Obtain the difference between the similarity score and the benchmark traffic of each business line, and use the difference as the traffic regulation score.

[0020] Preferably, the rearrangement method further includes:

[0021] Store the buried point data in the database;

[0022] And / or,

[0023] The rearrangement method further includes:

[0024] Synchronize the benchmark traffic of each business line to the remote dictionary service.

[0025] Preferably, the rearrangement method further includes:

[0026] Forcefully insert the product of a certain business line at the target position.

[0027] Preferably, the expression of the rearrangement score of the head product in each business line is as follows:

[0028]

[0029]

[0030] Wherein, represents the rearrangement score of the head product in each business line; Q represents the user; represents the product to be sorted; represents the relevance score between the user and each product in the set of products to be rearranged; represents the similarity score between products of each business line within the sliding window; represents the traffic regulation score; represents the business line to which product P belongs; represents the reference traffic of business line C; and both represent coefficients.

[0031] The second aspect of the present disclosure provides a rearrangement system for multi-business-line products, and the rearrangement system includes:

[0032] A first acquisition module, configured to acquire the relevance score between the user and each product in the set of products to be rearranged;

[0033] A second acquisition module, configured to classify all products according to business lines, and obtain the product queue of each classification according to the sorting result from high to low of the relevance scores;

[0034] A third acquisition module, configured to acquire the similarity score between products of each business line within the sliding window;

[0035] A fourth acquisition module, configured to acquire the traffic regulation score;

[0036] A fifth acquisition module, configured to acquire the head product from the product queue, and obtain the rearrangement score of the head product in each business line according to the relevance score, the similarity score, and the traffic regulation score;

[0037] A sixth acquisition module, configured to acquire the head product with the highest rearrangement score from the rearrangement scores of the head products in each business line as the target product to be placed at the target position;

[0038] A processing module, configured to add the target product to the set of sorted products, pop up the target product, and move the sliding window one position in a preset direction;

[0039] A seventh acquisition module, configured to obtain the product rearrangement list according to the set of sorted products obtained after the above loop is executed a preset number of times.

[0040] Preferably, the rearrangement system further includes:

[0041] An eighth acquisition module, configured to acquire the buried point data of the user;

[0042] A ninth acquisition module, configured to acquire the benchmark traffic of each business line according to the buried point data.

[0043] Preferably, the fourth acquisition module is configured to acquire the difference between the similarity score and the benchmark traffic of each business line, and use the difference as the traffic regulation score.

[0044] Preferably, the rearrangement system further includes:

[0045] A storage module, configured to store the buried point data in a database;

[0046] And / or

[0047] The rearrangement system further includes:

[0048] A synchronization module, configured to synchronize the benchmark traffic of each business line to a remote dictionary service.

[0049] Preferably, the rearrangement system further includes:

[0050] A forced insertion module, configured to forcibly insert a product of a certain business line at the target position.

[0051] Preferably, the expression of the rearrangement score of the head product in each business line is:

[0052]

[0053]

[0054] Wherein, represents the rearrangement score of the head product in each business line; Q represents the user; represents the product to be sorted; represents the relevance score between the user and each product in the set of products to be rearranged; represents the similarity score between the products of each business line within the sliding window; represents the traffic regulation score; represents the business line to which product P belongs; represents the benchmark traffic of business line C; And both represent coefficients.

[0055] A third aspect of the present disclosure provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the rearrangement method of the multi-business line products described in the first aspect is implemented.

[0056] A fourth aspect of the present disclosure provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the rearrangement method of the multi-business line products described in the first aspect is implemented.

[0057] A fifth aspect of the present disclosure provides a computer program product, including a computer program. When the computer program is executed by a processor, the rearrangement method of the multi-business line products described in the first aspect is implemented.

[0058] On the basis of conforming to the common knowledge in the art, the above preferred conditions can be combined arbitrarily to obtain various preferred examples of the present disclosure.

[0059] The positive and progressive effects of the present disclosure are as follows:

[0060] The present disclosure designs a set of reliable and general integrated rearrangement frameworks. The rearrangement framework is based on a multi-business line rearrangement algorithm related to the maximum margin, effectively weighing the correlation between users and products and the similarity between products of different business lines. It can be compatible with various rearrangement strategies such as traffic regulation and diversity scattering, ensuring the rationality and diversity of the rearrangement results, promoting the balance of multi-business line traffic, and improving the overall revenue and user experience of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flowchart of the rearrangement method of the multi-business line products provided in Embodiment 1 of the present disclosure.

[0062] Figure 2 It is a schematic diagram of the modules of the rearrangement system of the multi-business line products provided in Embodiment 2 of the present disclosure.

[0063] Figure 3 It is a schematic diagram of the structure of the electronic device for implementing the rearrangement method of the multi-business line products in Embodiment 3 of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] The present disclosure will be further described below by way of embodiments, but the present disclosure is not limited to the scope of the described embodiments.

[0065] In the embodiments of the present disclosure, prefix words such as "first" and "second" are only used to distinguish different described objects, and have no restrictive effect on the position, order, priority, quantity, content, etc. of the described objects. In the embodiments of the present disclosure, the use of prefix words such as ordinal numbers to distinguish described objects does not constitute a limitation on the described objects. For the statements of the described objects, refer to the descriptions in the claims or the context of the embodiments, and no redundant limitations should be formed due to the use of such prefix words. In addition, in the description of this embodiment, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0066] In the embodiments of the present disclosure, the processing of the collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information involved all complies with the provisions of relevant laws and regulations and does not violate public order and good customs.

[0067] Embodiment 1

[0068] Figure 1 is a flowchart of a rearrangement method for a multi-business-line product provided in Embodiment 1 of the present disclosure. As Figure 1 shown, the rearrangement method includes:

[0069] S1. Obtain the relevance scores of the user with each product in the product set to be rearranged;

[0070] In this embodiment, for the current request, a deep model or a linear weighting strategy is used to calculate the relevance scores of the user with each product in the product set to be rearranged;

[0071] S2. Classify all products according to the business line, and obtain the product queue of each classification according to the sorting result from high to low of the relevance scores;

[0072] In this embodiment, each business line includes a plurality of products.

[0073] S3. Obtain the similarity scores between the products of each business line within the sliding window;

[0074] In this embodiment, the steps of S3-S7 are looped N times, where N represents the size of the product set to be rearranged. Assume that the current loop is the i-th time; calculate the similarity scores between the products of each business line within the sliding window respectively;

[0075] S4. Obtain the traffic regulation scores;

[0076] S5. Obtain the head product from the product queue, and obtain the rearrangement scores of the head products of each business line according to the relevance scores, similarity scores, and traffic regulation scores;

[0077] In this embodiment, the head product is obtained from the product queue of each business line, and the rearrangement score of the head product in each business line is calculated by combining the relevance score, similarity score, and traffic regulation score;

[0078] S6. Obtain the head product with the highest rearrangement score among the head products of each business line as the target product to be placed at the target position;

[0079] In this embodiment, the target position refers to the position where the current loop is the i-th time, and the target product refers to the optimal product;

[0080] Specifically, select the head product with the highest rearrangement score from the rearrangement scores of the head products of each business line as the optimal product to be placed at the i-th position.

[0081] S7. Add the target product to the sorted product set, pop the target product, and move the sliding window one position in the preset direction;

[0082] In this embodiment, for example, add the target product to the sorted product set, pop the target product, and move the sliding window one position to the right;

[0083] S8. Obtain the product rearrangement list according to the sorted product set obtained after executing the loop of S3-S7 for a preset number of times.

[0084] In this embodiment, the sorted product set obtained after executing the loop of S3-S7 for N times is integrated into the final product rearrangement list, and the product rearrangement list is output.

[0085] This embodiment introduces a multi-business-line rearrangement method based on maximum margin correlation, designs a reliable and general integrated rearrangement framework. This rearrangement framework is based on the multi-business-line rearrangement algorithm of maximum margin correlation, effectively weighs the correlation between users and products and the similarity between products of different business lines, can be compatible with various rearrangement strategies such as traffic regulation and diversity scattering, ensures the rationality and diversity of the rearrangement results, and on this basis, proposes a sliding window traffic regulation algorithm framework, which promotes the balance of multi-business-line traffic and improves the overall platform revenue and user experience.

[0086] In an alternative embodiment, before step S2, the rearrangement method further includes:

[0087] Obtain the buried point data of the user;

[0088] In this embodiment, obtain the buried point data such as clicks and purchases of the user in the APP;

[0089] Obtain the benchmark traffic of each business line according to the buried point data.

[0090] In this embodiment, calculate the benchmark traffic of each business line according to the buried point data.

[0091] In an alternative embodiment, step S4 includes:

[0092] Obtain the difference between the similarity score and the baseline traffic of each business line, and use the difference as the traffic regulation score.

[0093] In an alternative embodiment, the rearrangement method further includes:

[0094] Store the buried point data in a database;

[0095] In this embodiment, the buried point data is stored in an offline Hive (data warehouse tool) table;

[0096] It should be noted that Hive is a data warehouse tool based on Hadoop, which is used for large-scale data storage, query, and analysis on Hadoop;

[0097] In an alternative embodiment, the rearrangement method further includes:

[0098] Synchronize the baseline traffic of each business line to a remote dictionary service.

[0099] In this embodiment, the baseline traffic of each business line is synchronized to Redis (remote dictionary service).

[0100] It should be noted that Redis is an open-source in-memory data structure storage system, which can be used as a database, cache, and message broker.

[0101] In an alternative embodiment, the rearrangement method further includes:

[0102] Forcefully insert a product of a certain business line at the target position.

[0103] In this embodiment, for example, for position i, forcefully insert a product of a certain business line.

[0104] It should be noted that only one of obtaining the similarity score between the products of each business line within the sliding window and forcefully inserting a product of a certain business line at the target position is executed.

[0105] In an alternative embodiment, the rearrangement score expression of the head product in each business line is as shown in formula (1):

[0106]

[0107] (1)

[0108] Wherein, represents the rearrangement score of the head product in each business line; Q represents the user; represents the product to be sorted; Represents the relevance score between the user and each product in the set of products to be rearranged; Represents the similarity score between products of each business line within the sliding window; Represents the traffic regulation score; Represents the business line to which product P belongs; Represents the baseline traffic of business line C; and both represent coefficients.

[0109] In the specific implementation process, in the sorting layer of the search and recommendation system, a large number of user features, candidate product features, context features, and cross features are usually mined, and a deep model is used to estimate the click-through rate, order completion rate, etc. of the current user for candidate products. This estimated value can be used as the relevance between the user and the product. In the mixed arrangement of products of multiple business lines, there is often a phenomenon that products of the same business line gather due to the influence of strong business lines, and some business lines with mediocre online performance have no chance of exposure. On the one hand, this will lead to an increasingly serious long-tail phenomenon in the entire recommendation and search system, thus affecting the normal operation of the entire system; on the other hand, when multiple products of the same type of business line are exposed simultaneously, the provided selectivity is too small, which is likely to cause visual fatigue to users and is not conducive to the user experience. Therefore, a greedy strategy can be adopted. In the set of products to be rearranged, find products that can maximize the relevance with the user and the diversity of the current list of products, and add them to the set of selected products, and so on, to obtain the final complete rearranged list, as shown in formula (2):

[0110]

[0111] (2)

[0112] Among them, represents the set of products to be rearranged, represents the set of selected products, represents a certain product, represents the relevance score between the user and each product in the set of products to be rearranged, usually the output probability value of the neurons in the last layer of the deep model, represents the similarity score between products of each business line within the sliding window. This similarity can usually be calculated by methods such as the category matching degree between products and the inner product of pre-trained embeddings in the e-commerce scenario. However, on the one hand, as the set of selected products becomes larger and larger, it is easy to cause to always be approximately equal to 1, and at this time, this method will degenerate into a sorting only affected by relevance and traffic regulation. On the other hand, when sorting by the above formula (2), it is necessary to calculate and traverse all the products in, and at the same time All products in need to be traversed To obtain the correlation of all products in the system, the time complexity is high. In order to solve the above problems at the same time, the present disclosure introduces the concept of sliding window, that is, The collection is changed to a fixed length Sliding Window , The size can be selected according to the actual situation. The larger the size, the closer it is to calculating the entire selected set. The smaller the size, the more it degenerates into sorting based on relevance only. According to practical experience, the number of products that can be exposed on about 3 screens can be selected as the window size. At the same time, the sliding window method also simulates the overall browsing process of users on the APP, which can better fit the user experience. In summary, The candidate products in only need to calculate the correlation with the product set in the sliding window, that is, to optimize formula (2) to get formula (1);

[0113] Furthermore, in order to further simplify the time complexity of the calculation, the present disclosure only considers the diversity of business lines as the similarity score, that is, it only needs to calculate the proportion of the business lines to which the candidate products in the sliding window belong in the sliding window. Compared with the original method, this calculation method of the business line dimension does not need to traverse all candidate products, but only needs to traverse the product with the highest correlation among the candidate products of each business line, because the re-ranking score of other products in the business line cannot exceed it, as shown in formula (3):

[0114] (3)

[0115] in, Indicates the number of products in a business line; Indicates the size of the sliding window.

[0116] Furthermore, in the actual re-ranking process, due to the influence of company strategy, user preferences, business experience, etc., there is often a need to enhance or suppress some business lines or categories. For example, in the tourism scenario, since group tour products account for a higher proportion of GMV than other product lines, it is often hoped that the ranking of group tour products will be improved in the mixed ranking process of multiple business lines, so that they have more exposure opportunities and are more likely to be noticed by users. Therefore, this disclosure additionally introduces the traffic control item , which is the candidate product for the current position The standard traffic of business line C to which it belongs The gap between them is shown in formula (4):

[0117] (4)

[0118] Among them, the business line benchmark traffic It can be set in combination with the business or calculated based on historical data. Since this item only involves the traffic proportion of the business line and is actually irrelevant to the product, it can be simplified to the difference between the current benchmark traffic and the similarity of the business line.

[0119] In this embodiment, the products at each position are selected preferentially in sequence. Therefore, it can simply and skillfully integrate the rearrangement strategy of forced insertion in fixed positions. It only needs to directly put the corresponding product when traversing this position, which increases the flexibility and adaptability of the algorithm. At the same time, in order to achieve a small and lightweight algorithm implementation for quick deployment and application, the benchmark traffic of the business line involved in this disclosure only needs to be calculated iteratively offline in a cycle, and the diversity only considers the business line dimension, which can reduce the complexity and resource consumption of the algorithm. And the entire process of traversing, selecting, and rearranging can be calculated in real time online to meet the needs of real-time applications.

[0120] In order to solve the problems of balancing the relevance between users and products, as well as the diversity between products of different business lines; reasonably regulating traffic in combination with the online performance of products of different business lines; the algorithm can conveniently integrate other rearrangement strategies to form an integrated framework; the overall algorithm implementation needs to be small and lightweight for quick implementation, etc. The above technical problems, this embodiment designs a set of reliable, general, flexible, lightweight, and quickly implementable integrated rearrangement framework for multiple business lines based on maximum margin correlation, which can be compatible with various rearrangement strategies such as traffic regulation, diversity scattering, and forced insertion. The actual effect depends on the business scenario. This rearrangement framework is based on the maximum margin correlation algorithm, weighing the relevance between users and products and the differences between products of different business lines, taking into account the user's interest points and browsing experience. At the same time, a traffic regulation algorithm based on a sliding window is incorporated into this rearrangement framework. By simulating the user's screen sliding behavior on the APP, the gap between the actual traffic of the business line within the window and the benchmark traffic is calculated, and the traffic distribution is regulated to return to the expectation. This rearrangement framework can balance the traffic within multiple business lines through the traffic regulation algorithm on the premise of meeting the relevance, personalization, and diversity of the rearrangement scenario, and effectively alleviate the poor sorting result phenomenon caused by the cold start problem in small scenarios, achieving the purpose of better serving merchants and pursuing the platform's own interests. In addition, its general integrated algorithm structure can be compatible with other common rearrangement strategies such as forced insertion.

[0121] Embodiment 2

[0122] Corresponding to the foregoing embodiment of a rearrangement method for multi-business line products, the present disclosure also provides an embodiment of a rearrangement system for multi-business line products.

[0123] Figure 2 It is a schematic diagram of the modules of a rearrangement system for multi-business line products provided in Embodiment 2 of the present disclosure, as Figure 2 shown, the rearrangement system includes:

[0124] The first acquisition module 21 is configured to acquire the relevance scores of the user with each product in the product set to be rearranged;

[0125] In this embodiment, for the current request, a deep model or a linear weighting strategy is adopted to calculate the relevance scores of the user with each product in the product set to be rearranged;

[0126] The second acquisition module 22 is configured to classify all products according to business lines, and acquire the product queues of each classification according to the sorting result from high to low of the relevance scores;

[0127] In this embodiment, each business line includes multiple products.

[0128] The third acquisition module 23 is configured to acquire the similarity scores between the products of each business line within the sliding window;

[0129] In this embodiment, the steps of S3 - S7 are repeatedly executed N times, where N represents the size of the product set to be rearranged. Assuming that the current loop is the i-th time; the similarity scores between the products of each business line within the sliding window are calculated respectively;

[0130] The fourth acquisition module 24 is configured to acquire the traffic regulation scores;

[0131] The fifth acquisition module 25 is configured to acquire the head product from the product queue, and acquire the rearrangement scores of the head products of each business line according to the relevance scores, similarity scores, and traffic regulation scores;

[0132] In this embodiment, the head product is acquired from the product queue of each business line, and the rearrangement scores of the head products of each business line are calculated in combination with the relevance scores, similarity scores, and traffic regulation scores;

[0133] The sixth acquisition module 26 is configured to acquire the head product with the highest rearrangement score among the rearrangement scores of the head products of each business line as the target product to be placed at the target position;

[0134] In this embodiment, the target position refers to the position where the current loop is the i-th time, and the target product refers to the optimal product;

[0135] Specifically, the head product with the highest rearrangement score is selected from the rearrangement scores of the head products of each business line as the optimal product to be placed at the i-th position.

[0136] The processing module 27 is configured to add the target product to the sorted product set, pop up the target product, and move the sliding window one position in the preset direction;

[0137] In this embodiment, for example, the target product is added to the sorted product set, the target product is popped up, and the sliding window is moved one position to the right;

[0138] The seventh acquisition module 28 is configured to obtain a product rearrangement list according to the sorted product set obtained after the above loop is executed a preset number of times.

[0139] This embodiment introduces a multi-business line rearrangement method based on maximum margin correlation, designs a reliable and general integrated rearrangement framework. The rearrangement framework is based on the multi-business line rearrangement algorithm of maximum margin correlation, effectively balances the correlation between users and products and the similarity between products of different business lines, can be compatible with various rearrangement strategies such as traffic regulation and diversity scattering, ensures the rationality and diversity of the rearrangement results, and on this basis, proposes a sliding window traffic regulation algorithm framework, which promotes the balance of multi-business line traffic and improves the overall revenue and user experience of the platform.

[0140] In an optional embodiment, the rearrangement system further includes:

[0141] The eighth acquisition module is configured to acquire the buried point data of the user;

[0142] In this embodiment, the buried point data such as clicks and purchases of the user within the APP is acquired;

[0143] The ninth acquisition module is configured to obtain the benchmark traffic of each business line according to the buried point data.

[0144] In this embodiment, the benchmark traffic of each business line is calculated according to the buried point data.

[0145] In an optional embodiment, the fourth acquisition module is configured to obtain the difference between the similarity score and the benchmark traffic of each business line, and use the difference as the traffic regulation score.

[0146] In an optional embodiment, the rearrangement system further includes:

[0147] The storage module is configured to store the buried point data in a database;

[0148] In this embodiment, the buried point data is stored in an offline Hive (data warehouse tool) table;

[0149] It should be noted that Hive is a data warehouse tool based on Hadoop, which is used for large-scale data storage, query and analysis on Hadoop;

[0150] In an optional embodiment, the rearrangement system further includes:

[0151] The synchronization module is configured to synchronize the benchmark traffic of each business line to a remote dictionary service.

[0152] In this embodiment, the benchmark traffic of each business line is synchronized to Redis.

[0153] It should be noted that Redis is an open-source in-memory data structure storage system that can be used as a database, cache, and message broker.

[0154] In an optional embodiment, the rearrangement system further includes:

[0155] A forced insertion module for forcibly inserting a product of a certain business line at the target position.

[0156] In this embodiment, for example, at the i position, a product of a certain business line is forcibly inserted.

[0157] It should be noted that only one of obtaining the similarity scores between the products of each business line within the sliding window and forcibly inserting a product of a certain business line at the target position is executed.

[0158] In an optional embodiment, the expression of the rearrangement score of the head product in each business line is as shown in formula (1) of Embodiment 1.

[0159] In the specific implementation process, in the sorting layer of the search and recommendation system, a large number of user features, candidate product features, context features, and cross features are usually mined, and a deep model is used to estimate the click-through rate, order completion rate, etc. of the current user for the candidate products. This estimated value can be used as the correlation between the user and the product. In the mixed arrangement of products of multiple business lines, the phenomenon of products of the same business line clustering often occurs due to the influence of strong business lines, and some business lines with mediocre online performance have no chance of exposure. On the one hand, this will lead to an increasingly serious long-tail phenomenon in the entire recommendation and search system, thus affecting the normal operation of the entire system; on the other hand, when multiple products of the same type of business line are exposed at the same time, the selectivity provided is too small, which is likely to cause visual fatigue to users and is not conducive to the user experience. Therefore, a greedy strategy can be adopted to find the product that can maximize the correlation with the user and the diversity of the current list of products in the set of products to be rearranged, add it to the set of selected products, and so on, to obtain the final complete rearranged list, that is, as shown in formula (2) of Embodiment 1.

[0160] Among them, represents the set of products to be rearranged, represents the set of selected products, represents a certain product, represents the correlation score between the user and each product in the set of products to be rearranged, usually the output probability value of the neurons in the last layer of the deep model, represents the similarity score between the products of each business line within the sliding window. This similarity can usually be calculated by means of category matching degree between products, vector inner product of pre-trained embeddings, etc. However, on the one hand, as the set of selected products becomes larger and larger, it is easy to cause The phenomenon that is always approximately equal to 1, in which case the method will degenerate into a sorting method that is only affected by correlation and traffic control. On the other hand, the sorting method based on the above formula (2) requires the calculation of the traversal All products in All products in need to be traversed To obtain the correlation of all products in the system, the time complexity is high. In order to solve the above problems at the same time, the present disclosure introduces the concept of sliding window, that is, The collection is changed to a fixed length Sliding Window , The size can be selected according to the actual situation. The larger the size, the closer it is to calculating the entire selected set. The smaller the size, the more it degenerates into sorting based on relevance only. According to practical experience, the number of products that can be exposed on about 3 screens can be selected as the window size. At the same time, the sliding window method also simulates the overall browsing process of users on the APP, which can better fit the user experience. In summary, The candidate products in only need to calculate the correlation with the product set in the sliding window, that is, to optimize formula (2) to get formula (1);

[0161] Furthermore, in order to further simplify the time complexity of the calculation, the present disclosure only considers the diversity of business lines as the similarity score, that is, it only needs to calculate the proportion of the business lines to which the candidate products in the sliding window belong in the sliding window. Compared with the original method, this calculation method of the business line dimension does not need to traverse all candidate products, but only needs to traverse the product with the highest correlation among the candidate products of each business line, because the re-ranking score of other products in the business line cannot exceed it, as shown in formula (3) in Example 1.

[0162] Furthermore, in the actual re-ranking process, due to the influence of company strategy, user preferences, business experience, etc., there is often a need to enhance or suppress some business lines or categories. For example, in the tourism scenario, since group tour products account for a higher proportion of GMV than other product lines, it is often hoped that the ranking of group tour products will be improved in the mixed ranking process of multiple business lines, so that they have more exposure opportunities and are more likely to be noticed by users. Therefore, this disclosure additionally introduces the traffic control item , which is the candidate product for the current position The standard traffic of business line C to which it belongs The difference between them is shown in formula (4) in Example 1.

[0163] In this embodiment, the products at each position are selected preferentially in sequence. Therefore, it is possible to simply and ingeniously integrate the rearrangement strategy of strong insertion with fixed positions. It only needs to directly place the corresponding product when traversing this position, which increases the flexibility and adaptability of the algorithm. At the same time, in order to achieve a small and lightweight algorithm implementation for quick deployment and application, the benchmark traffic of the business line involved in this disclosure only needs to be calculated iteratively in an offline cycle, and the diversity only considers the business line dimension, which can reduce the complexity and resource consumption of the algorithm. And the entire process of traversing and preferentially rearranging can be calculated in real time online to meet the needs of real-time applications.

[0164] In order to solve the above technical problems such as the need to balance the relevance between users and products, and the diversity between products of different business lines; to carry out reasonable traffic regulation by combining the online performance of products of different business lines; the algorithm can easily integrate other rearrangement strategies to form an integrated framework; the overall algorithm implementation needs to be small and lightweight for quick implementation, etc., this embodiment designs a reliable, general, flexible, lightweight, and quickly implementable integrated rearrangement framework for multiple business lines based on maximum margin correlation. It can be compatible with multiple rearrangement strategies such as traffic regulation, diversity scattering, and strong insertion. The actual effect depends on the business scenario. This rearrangement framework is based on the maximum margin correlation algorithm, weighing the relevance between users and products and the differences between products of different business lines, and taking into account the user's interest points and browsing experience. At the same time, a traffic regulation algorithm based on a sliding window is incorporated into this rearrangement framework. By simulating the user's screen-swiping behavior on the APP, the gap between the actual traffic of the business line within the window and the benchmark traffic is calculated, and the traffic distribution is regulated to return to the expectation. This rearrangement framework can balance the traffic within multiple business lines through the traffic regulation algorithm on the premise of meeting the relevance, personalization, and diversity of the rearrangement scenario, and effectively alleviate the poor sorting result phenomenon caused by the cold start problem in small scenarios, achieving the purpose of better serving merchants and pursuing the platform's own interests. In addition, its general integrated algorithm structure can be compatible with other common rearrangement strategies such as strong insertion.

[0165] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The system embodiments described above are only illustrative. The units described as separate components may or may not be physically separated. The components as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this disclosure.

[0166] Embodiment 3

[0167] Figure 3FIG. 0 is a schematic structural diagram of an electronic device shown in Embodiment 3 of the present disclosure. The electronic device includes a memory, a processor, and a computer program stored in the memory and configured to run on the processor. When the processor executes the computer program, the rearrangement method of the multi-service line product described in any of the above embodiments is implemented. Figure 3 The illustrated electronic device 90 is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present disclosure.

[0168] As Figure 3 shown, the electronic device 90 may be presented in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 90 may include, but are not limited to: at least one of the above-mentioned processors 91, at least one of the above-mentioned memories 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0169] The bus 93 includes a data bus, an address bus, and a control bus.

[0170] The memory 92 may include a volatile memory, such as a random access memory (RAM) 921 and / or a cache memory 922, and may further include a read-only memory (ROM) 923.

[0171] The memory 92 may further include a program tool 925 (or utility) having a set (at least one) of program modules 924. Such program modules 924 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0172] The processor 91 executes various functional applications and data processing by running the computer program stored in the memory 92, such as the rearrangement method of the multi-service line product provided in any of the above embodiments.

[0173] The electronic device 90 may also communicate with one or more external devices 94 (such as a keyboard, a pointing device, etc.). Such communication may be performed through an input / output (I / O) interface 95. And, the electronic device 90 may further communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 96. As Figure 3 shown, the network adapter 96 communicates with other modules of the electronic device 90 through the bus 93. It should be understood that although not shown in the figure, other hardware and / or software modules may be used in combination with the electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0174] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0175] Embodiment 4

[0176] Embodiment 4 of the present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the rearrangement method of the multi-service line product provided in any of the above embodiments.

[0177] Among them, the more specific forms that the readable storage medium can adopt may include but are not limited to: portable disks, hard disks, random access memories, read-only memories, erasable programmable read-only memories, optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0178] Embodiment 5

[0179] Embodiment 5 of the present disclosure also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the rearrangement method of the multi-service line product described in any of the above.

[0180] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on the user device, partially on the user device, executed as an independent software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0181] Although the specific embodiments of the present disclosure are described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Without departing from the principles and essence of the present disclosure, those skilled in the art can make various changes or modifications to these embodiments, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A method for rearranging products of multiple business lines, characterized in that: The rearrangement method comprises: S1. Obtain the relevance score between the user and each product in the set of products to be rearranged; S2. Classify all products according to business lines, and obtain a product queue for each category in descending order according to the relevance scores; S3, obtaining the similarity scores between the products of each business line in the sliding window; S4. Obtaining a traffic control score; S5. Obtain the head product from the product queue, and obtain the reordering score of the head product in each business line according to the relevance score, the similarity score, and the traffic regulation score; S6. Obtaining the top product with the highest reordering score from the reordering scores of the top products in each business line as the target product to be placed at the target position; S7, adding the target product to the sorted product set, popping up the target product, and moving the sliding window by one position in a preset direction; S8. Obtain a product rearrangement list based on the sorted product set obtained after executing S3-S7 repeatedly for a preset number of times.

2. The method for rearranging products of multiple business lines according to claim 1, characterized in that: Before step S2, the rearrangement method further comprises: Get the user's buried data; Obtain the baseline traffic for each business line based on the tracking data.

3. The method for rearranging products of multiple business lines according to claim 2, characterized in that: Step S4 includes: Obtain the difference between the similarity score and the baseline traffic of each business line, and use the difference as the traffic regulation score.

4. The method for rearranging products of multiple business lines according to claim 2 or 3, characterized in that: The rearrangement method further comprises: The buried point data is stored in a database; and / or, The rearrangement method further comprises: The baseline traffic of each business line is synchronized to the remote dictionary service.

5. The method for rearranging products of multiple business lines according to claim 1, characterized in that: The rearrangement method further comprises: Forcefully insert a product of a certain business line at the target location.

6. The method for rearranging products of multiple business lines according to claim 1, characterized in that: The expression for the reordering score of the head product in each business line is: in, represents the re-ranking score of the top products in each business line; Q represents the user; Indicates the products to be sorted; Represents the relevance score between the user and each product in the product set to be rearranged; Represents the similarity score between products of each business line within the sliding window; represents the flow control score; Indicates the business line to which product P belongs; Indicates the baseline traffic of business line C; and All represent coefficients.

7. A multi-business line product reordering system, characterized in that: The rearrangement system comprises: A first acquisition module is used to obtain a relevance score between the user and each product in the set of products to be rearranged; A second acquisition module is used to classify all products according to business lines, and obtain a product queue for each category in a descending order according to the relevance scores; The third acquisition module is used to obtain the similarity scores between the products of each business line in the sliding window; A fourth acquisition module is used to obtain a traffic control score; A fifth acquisition module, used to acquire the head product from the product queue, and acquire the reordering score of the head product in each business line according to the relevance score, the similarity score and the traffic regulation score; A sixth acquisition module, configured to acquire the head product with the highest reordering score from the reordering scores of the head products in each business line as the target product to be placed at the target position; A processing module, used for adding the target product to the sorted product set, popping out the target product, and moving the sliding window by one position in a preset direction; The seventh acquisition module is used to obtain a product rearrangement list according to the sorted product set obtained after the above cycle is executed a preset number of times.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, the method for rearranging multi-business line products according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for rearranging multi-business line products according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for rearranging multi-business line products according to any one of claims 1 to 6 is implemented.

Images