An accurate recommendation method, device, equipment, medium and product for e-commerce products

Abstract

The present invention discloses a method, device, equipment, medium and product for precise recommendation of e-commerce products, which relates to the technical field of big data analysis. The method first extracts sample data of various e-commerce products based on e-commerce product browsing record data and e-commerce product purchase record data, and trains purchase event prediction models of various e-commerce products based on the sample data. Then, based on the prediction models and e-commerce product recommendation schemes, the confidence levels of events in which various e-commerce products are purchased by target customers during the next startup and operation period are estimated. Finally, when it is determined that the sum of the confidence levels of various e-commerce products exceeds a preset threshold, it is confirmed that the recommendation accuracy of the recommendation scheme meets the preset goal, and during the next startup and operation period, the recommendation scheme is executed through the e-commerce platform application program. In this way, when applied to multi-category e-commerce product recommendation scenarios, it can effectively improve the user's purchase intention and achieve the purpose of further improving the accuracy of product recommendation.

Description

Technical Field

[0001] The present invention belongs to the technical field of big data analysis, and particularly relates to a method, device, equipment, medium and product for precise recommendation of e-commerce products. Background Art

[0002] With the booming development of e-commerce, users' demand for personalized recommendations is increasing day by day. Traditional e-commerce product recommendation solutions mainly rely on users' historical purchase records and evaluation data after purchase. For example, the existing patent "CN114429384A, Product Intelligent Recommendation Method and System Based on E-commerce Platform" provides a product intelligent recommendation solution based on an e-commerce platform, which specifically includes: querying the data of target products and their corresponding evaluation data in the historical shopping data of a target user; obtaining a group of similar users with similar shopping behaviors to the target user; obtaining a list of alternative products; selecting a preset number of alternative products from the list of alternative products; obtaining the sentiment scores corresponding to them for a preset key evaluation word category based on the evaluation data of the selected alternative products and the target products and forming a score input matrix with these sentiment scores; inputting the score input matrix into a recommendation model so that the recommendation model outputs recommended products. However, the foregoing solution obviously has problems such as insufficient continuity of product recommendation quality and insufficient comprehensiveness of portrait type generation due to the lack of precise adjustment of the product recommendation interval duration when users have less feedback on the products on the browsing page, resulting in a decrease in the accuracy of product recommendation.

[0003] In response to the above problems, the existing patent "CN117829965A, A Product Recommendation System and Method for Cross-border E-commerce Platform" provides another product recommendation solution for a cross-border e-commerce platform, which specifically includes: a data collection module; a data processing module for calculating and processing the first-level operation feature data of the browsing page to output the second-level operation feature data of the browsing page; a product update module, including a portrait generation unit for generating a user portrait according to the user feature data and the user's page browsing data, and a product recommendation unit connected to the portrait generation unit for scheduling the corresponding product information to be recommended in the user portrait to the browsing page; a control module for determining the corresponding update method of the product recommendation unit according to the proportion of product order quantity, or determining the portrait classification method of the portrait generation unit according to the proportion of product order quantity and the overlapping quantity of the searched and recommended product types. The foregoing solution has achieved an improvement in the accuracy of product recommendation on the cross-border e-commerce platform.

[0004] However, both of the above solutions only consider the impact of recommending a single e-commerce product on customer purchasing decisions, but do not consider the situation where recommending different types of e-commerce products will have a competitive or linkage impact on customer purchasing decisions. For example, recommending tablets will have a competitive impact on customers' decisions to purchase other electronic devices such as smartphones or laptops. For example, recommending smartphones will have a linkage impact on customers' decisions to purchase related supporting products such as Bluetooth headsets or mobile phone cases. Therefore, how to provide a new e-commerce product recommendation solution that is suitable for multiple e-commerce product recommendation scenarios and can effectively increase users' purchasing intention, so as to further improve the accuracy of product recommendations, is a topic that technical personnel in this field urgently need to study. Summary of the Invention

[0005] The purpose of the present invention is to provide a method, device, computer equipment, computer-readable storage medium and computer program product for accurately recommending e-commerce products, so as to solve the problem that the accuracy of product recommendations needs to be improved because existing e-commerce product recommendation schemes only consider the impact of recommending a single e-commerce product on customer purchasing decisions, but do not consider the situation where recommending different types of e-commerce products will have a competitive or linkage impact on customer purchasing decisions.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] First, a method for accurately recommending e-commerce products is provided, including:

[0008] Collect e-commerce product browsing history data and e-commerce product purchase history data generated by target customers during each historical launch and operation of the e-commerce platform application;

[0009] Obtaining browsing event indication values or browsing time proportion values of various types of e-commerce products during the various historical startups and operations based on the e-commerce product browsing record data, and obtaining purchasing event indication values of various types of e-commerce products during the various historical startups and operations based on the e-commerce product purchase record data, wherein the browsing event indication value is used to indicate whether an event of browsing the corresponding e-commerce product by the target customer occurs during the startup and operation of the e-commerce platform application, the browsing time proportion value refers to the ratio of the time the target customer browses the corresponding e-commerce product to the total time the target customer browses all e-commerce products during the startup and operation of the e-commerce platform application, and the purchasing event indication value is used to indicate whether an event of purchasing the corresponding e-commerce product by the target customer occurs during the startup and operation of the e-commerce platform application;

[0010] For the various types of e-commerce products, the corresponding sample data is extracted in the following manner: The browsing event indication value or the browsing time ratio during the consecutive m historical startup runs of all e-commerce products is used as the model input item, and the purchase event indication value during the last historical startup run among the consecutive m historical startup runs of the corresponding e-commerce product is used as the model output item. Then, the model input item and the model output item are used as a sample data, where m represents a positive integer greater than or equal to 2;

[0011] For the various types of e-commerce products, all the corresponding sample data is used to calibrate and verify the modeling of a machine learning model based on an artificial intelligence algorithm to obtain the corresponding purchase event prediction model;

[0012] For the various types of e-commerce products, the browsing event indication value or the browsing time ratio during the current most recent consecutive m - 1 historical startup runs of all e-commerce products and the e-commerce product recommendation plan during the next startup run of the e-commerce platform application are imported into the corresponding purchase event prediction model, and the confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup run is output. Among them, the e-commerce product recommendation plan includes the recommendation event indication value or the recommendation frequency ratio of the various types of e-commerce products during the next startup run. The recommendation event indication value is used to indicate whether an event of recommending the corresponding e-commerce product to the target customer for browsing is triggered during the next startup run, and the recommendation frequency ratio refers to the ratio of the frequency of recommending the corresponding e-commerce product to the target customer for browsing to the total frequency of recommending all e-commerce products to the target customer for browsing during the next startup run;

[0013] Judge whether the sum of the confidence levels of the various types of e-commerce products exceeds the first preset threshold;

[0014] If so, during the next startup run, execute the e-commerce product recommendation plan through the e-commerce platform application.

[0015] Based on the above invention content, a new solution for precise recommendation of e-commerce products in multi-category e-commerce product recommendation scenarios is provided. That is, first, based on e-commerce product browsing record data and e-commerce product purchase record data, sample data of various types of e-commerce products is extracted, and purchase event prediction models for various types of e-commerce products are trained based on the sample data. Then, based on the prediction models and e-commerce product recommendation schemes, the confidence levels of events where various e-commerce products are purchased by target customers during the current next startup operation are estimated. Finally, when it is determined that the sum of the confidence levels of various e-commerce products exceeds a preset threshold, it is confirmed that the recommendation accuracy of the recommendation scheme meets the preset goal, and during the current next startup operation, the recommendation scheme is executed through the e-commerce platform application program. In this way, both the impact of recommending a single e-commerce product on the customer's purchase decision and the situation where recommending different types of e-commerce products will have a competitive or linkage impact on the customer's purchase decision are considered, enabling effective improvement of the user's purchase willingness when applied to multi-category e-commerce product recommendation scenarios, achieving the purpose of further enhancing the accuracy of product recommendation, and facilitating practical application and promotion.

[0016] In a possible design, the method further includes:

[0017] For each type of e-commerce product, according to the purchase event indication value of the corresponding e-commerce product during each historical startup operation and the purchase event indication values of other types of e-commerce products during each historical startup operation, the linkage purchase impact coefficients between the corresponding e-commerce product and other types of e-commerce products are calculated respectively;

[0018] Judging whether the sum of the confidence levels of various types of e-commerce products exceeds a first preset threshold includes:

[0019] For each type of e-commerce product, according to the corresponding confidence level and the linkage purchase impact coefficient between the corresponding e-commerce product and other types of e-commerce products, the corresponding confidence level correction value is calculated according to the following formula:

[0020]

[0021] In the formula, N represents the total number of categories of e-commerce products, i represents a positive integer less than or equal to N, c i represents the confidence level of the i-th type of e-commerce product, represents the confidence level correction value of the i-th type of e-commerce product, i' represents a positive integer less than or equal to N - 1, c i,i′ represents the confidence level of the other i'-th type of e-commerce product relative to the i-th type of e-commerce product, η i,i′ represents the linkage purchase impact coefficient between the i-th type of e-commerce product and the other i'-th type of e-commerce product;

[0022] Judge whether the sum of the confidence level correction values of various types of e-commerce products exceeds the first preset threshold.

[0023] In a possible design, for each type of e-commerce product, according to the purchase event indication value of the corresponding e-commerce product during each historical startup operation and the purchase event indication values of other types of e-commerce products during each historical startup operation, the linkage purchase influence coefficients of the corresponding e-commerce product and other types of e-commerce products are calculated respectively, including:

[0024] For a certain type of e-commerce product, extract the purchase event indication values of the corresponding e-commerce product during the current most recent consecutive M historical startup operations to form a first sample S1, where M represents a positive integer greater than m;

[0025] For another certain type of e-commerce product relative to the said certain type of e-commerce product, extract the purchase event indication values of the corresponding e-commerce product during the current most recent consecutive M historical startup operations to form a second sample S2;

[0026] Perform a normal distribution KS test on the first sample S1 to calculate a first test statistic p-value p1, and also perform the normal distribution KS test on the second sample S2 to calculate a second test statistic p-value p2;

[0027] If both the first test statistic p-value p1 and the second test statistic p-value p2 are greater than a second preset threshold, then calculate the correlation coefficient r between the first sample S1 and the sample S2 according to the following formula q :

[0028]

[0029] In the formula, i represents a positive integer, x i represents the i-th sample value in the first sample S1, y i represents the i-th sample value in the second sample S2, represents the sample mean of the first sample S1, represents the sample mean of the second sample S2;

[0030] Take the correlation coefficient r q as the linkage purchase influence coefficient between the said certain type of e-commerce product and the other certain type of e-commerce product.

[0031] In a possible design, after determining whether the sum of the confidence levels of each type of e-commerce product exceeds a first preset threshold, if it is determined that the sum of the confidence levels of each type of e-commerce product does not exceed the first preset threshold, then the method further includes the following steps S701 - S704:

[0032] S701. Adjust the current e-commerce product recommendation plan to obtain a new e-commerce product recommendation plan, and then execute step S702;

[0033] S702. For each type of e-commerce product, import the browsing event indication value or browsing time ratio of all e-commerce products during the current most recent consecutive m - 1 historical startup runs and the new e-commerce product recommendation plan into the corresponding purchase event prediction model, and output the new confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup run, and then execute step S703;

[0034] S703. Determine whether the sum of the new confidence levels of each type of e-commerce product exceeds the first preset threshold. If so, execute step S704; otherwise, execute step S701;

[0035] S704. During the next startup run, execute the new e-commerce product recommendation plan through the e-commerce platform application.

[0036] In a possible design, when there are multiple e-commerce product recommendation plans, the method further includes:

[0037] For each type of e-commerce product and each e-commerce product recommendation plan among multiple e-commerce product recommendation plans, import the browsing event indication value or browsing time ratio of all e-commerce products during the current most recent consecutive m - 1 historical startup runs and the corresponding plan into the purchase event prediction model of the corresponding e-commerce product, and output the confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup run;

[0038] For each e-commerce product recommendation plan, accumulate and calculate the confidence levels of each type of e-commerce product and the corresponding confidence levels to obtain the corresponding total confidence level value;

[0039] Determine whether the maximum total confidence level value exceeds the first preset threshold;

[0040] If so, during the next startup run, execute a certain e-commerce product recommendation plan among multiple e-commerce product recommendation plans that corresponds to the maximum total confidence level value through the e-commerce platform application.

[0041] In a possible design, the method further includes:

[0042] Apply the browsing event indication values or browsing time ratio values of all the e-commerce products during the most recent consecutive m-1 historical startup runs, and the purchase event prediction models of the various types of e-commerce products, and optimize the e-commerce product recommendation plan based on an optimization algorithm to obtain the e-commerce product recommendation plan and the optimal search result for maximizing the sum of the confidence levels of the various types of e-commerce products;

[0043] During the next startup run, execute the e-commerce product recommendation plan obtained based on the optimal search result through the e-commerce platform application.

[0044] In a second aspect, there is provided an e-commerce product precise recommendation device, including a record data collection unit, a record data processing unit, a sample data extraction unit, a prediction model training unit, a prediction model application unit, a threshold comparison and judgment unit, and a recommendation plan execution unit that are sequentially communicatively connected;

[0045] The record data collection unit is used to collect the e-commerce product browsing record data and e-commerce product purchase record data generated by a target customer during each historical startup run of the e-commerce platform application;

[0046] The record data processing unit is used to process the e-commerce product browsing record data to obtain the browsing event indication values or browsing time ratio values of the various types of e-commerce products during each historical startup run, and process the e-commerce product purchase record data to obtain the purchase event indication values of the various types of e-commerce products during each historical startup run. Among them, the browsing event indication value is used to indicate whether the event that the corresponding e-commerce product is browsed by the target customer occurs during the startup run of the e-commerce platform application, the browsing time ratio value is the ratio of the duration of the target customer browsing the corresponding e-commerce product to the total duration of browsing all e-commerce products during the startup run of the e-commerce platform application, and the purchase event indication value is used to indicate whether the event that the corresponding e-commerce product is purchased by the target customer occurs during the startup run of the e-commerce platform application;

[0047] The sample data extraction unit is used to extract the corresponding sample data for the various types of e-commerce products in the following manner: use the browsing event indication values or browsing time ratio values of all e-commerce products during consecutive m historical startup runs as model input items, and use the purchase event indication value of the corresponding e-commerce product during the last historical startup run among the consecutive m historical startup runs as the model output item, and then use the model input item and the model output item as a sample data, where m represents a positive integer greater than or equal to 2;

[0048] The prediction model training unit is configured to calibrate and verify a machine learning model based on an artificial intelligence algorithm for each type of e-commerce product by applying all the corresponding sample data, so as to obtain a corresponding purchase event prediction model;

[0049] The prediction model application unit is configured to, for each type of e-commerce product, input the browsing event indication value or the browsing time ratio during the current most recent consecutive m-1 historical startup runs of all the e-commerce products and the e-commerce product recommendation scheme during the next startup run of the e-commerce platform application into the corresponding purchase event prediction model, and output the confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup run. Wherein, the e-commerce product recommendation scheme includes the recommendation event indication value or the recommendation frequency ratio of each type of e-commerce product during the next startup run. The recommendation event indication value is used to indicate whether an event that the corresponding e-commerce product is recommended to the target customer for browsing is triggered during the next startup run, and the recommendation frequency ratio refers to the ratio of the frequency that the corresponding e-commerce product is recommended to the target customer for browsing to the total frequency that all e-commerce products are recommended to the target customer for browsing during the next startup run;

[0050] The threshold comparison and judgment unit is configured to judge whether the sum of the confidence levels of each type of e-commerce product exceeds a first preset threshold;

[0051] The recommendation scheme execution unit is configured to, when it is determined that the sum of the confidence levels of each type of e-commerce product exceeds the first preset threshold, execute the e-commerce product recommendation scheme through the e-commerce platform application during the next startup run.

[0052] In a third aspect, the present invention provides a computer device, which includes a memory, a processor, and a transceiver that are communicatively connected in sequence. Wherein, the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the e-commerce product precise recommendation method as described in the first aspect or any possible design in the first aspect.

[0053] In a fourth aspect, the present invention provides a computer-readable storage medium, on which instructions are stored. When the instructions are run on a computer, the e-commerce product precise recommendation method as described in the first aspect or any possible design in the first aspect is executed.

[0054] In a fifth aspect, the present invention provides a computer program product, which includes a computer program or instructions. When the computer program or the instructions are executed by a computer, the e-commerce product precise recommendation method as described in the first aspect or any possible design in the first aspect is implemented.

[0055] Beneficial effects of the above solution:

[0056] (1) The present invention creatively provides a new solution for precise recommendation of e-commerce products in multi-category e-commerce product recommendation scenarios. That is, first, based on e-commerce product browsing record data and e-commerce product purchase record data, sample data of various e-commerce products is extracted, and a purchase event prediction model for each type of e-commerce product is trained based on the sample data. Then, based on the prediction model and the e-commerce product recommendation scheme, the confidence level of the event that various e-commerce products will be purchased by the target customer during the next startup and operation period is estimated. Finally, when it is determined that the sum of the confidence levels of various e-commerce products exceeds a preset threshold, it is confirmed that the recommendation accuracy of the recommendation scheme meets the preset goal, and during the next startup and operation period, the recommendation scheme is executed through the e-commerce platform application program. In this way, both the impact of recommending a single e-commerce product on the customer's purchase decision and the situation where recommending different types of e-commerce products will have a competitive or linkage impact on the customer's purchase decision are considered, so that when applied to multi-category e-commerce product recommendation scenarios, it can effectively improve the user's purchase intention, achieve the purpose of further improving the accuracy of product recommendation, and facilitate practical application and promotion;

[0057] (2) By calculating the linkage purchase impact coefficient between e-commerce products based on the historical purchase situations of various e-commerce products and performing confidence correction, it is possible to improve the estimation accuracy of the user's purchase intention and further improve the accuracy of product recommendation while further considering the existence of linkage purchases between different types of e-commerce products;

[0058] (3) When the recommendation accuracy of the recommendation scheme does not meet the preset goal, the e-commerce product recommendation scheme can be continuously adjusted until the recommendation accuracy meets the preset goal, thereby ensuring the e-commerce product recommendation effect;

[0059] (4) When there are multiple e-commerce product recommendation schemes, the recommendation scheme can be preferentially selected to maximize the recommendation accuracy and meet the preset goal, thereby ensuring the best e-commerce product recommendation effect;

[0060] (5) It is also possible to optimize to obtain the optimal e-commerce product recommendation scheme to maximize the recommendation accuracy, thereby ensuring the best e-commerce product recommendation effect. Description of the Drawings

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0062] Figure 1 Schematic flowchart of the e-commerce product precise recommendation method provided by the embodiment of the present application.

[0063] Figure 2 Supplementary schematic flowchart of the e-commerce product precise recommendation method when the total confidence of various e-commerce products does not exceed the first preset threshold provided by the embodiment of the present application.

[0064] Figure 3 Supplementary schematic flowchart of the e-commerce product precise recommendation method when there are multiple e-commerce product recommendation schemes provided by the embodiment of the present application.

[0065] Figure 4 Schematic structural diagram of the e-commerce product precise recommendation device provided by the embodiment of the present application.

[0066] Figure 5 Schematic structural diagram of the computer device provided by the embodiment of the present application. Detailed implementation manners

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the present invention in combination with the accompanying drawings and the description of the embodiments or the prior art. Obviously, the following description of the structures of the accompanying drawings is only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other embodiments can be obtained based on these embodiments. It should be noted here that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation to the present invention.

[0068] It should be understood that although terms such as first and second etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, the first object can be called the second object, and similarly, the second object can be called the first object, without departing from the scope of the exemplary embodiments of the present invention.

[0069] It should be understood that for the term "and / or" that may appear in this article, it is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, B exists alone, or A and B exist simultaneously, etc. For another example, A, B, and / or C can represent any one of A, B, and C or any combination of them; for the term " / and" that may appear in this article, it is a description of another association object relationship, indicating that two relationships may exist. For example, A / and B can represent: A exists alone or A and B exist simultaneously, etc. In addition, for the character " / " that may appear in this article, generally, the objects before and after are in an "or" relationship.

[0070] Embodiment

[0071] As Figure 1 shown, the e-commerce product precise recommendation method provided in the first aspect of this embodiment can be, but is not limited to, executed by a computer device with certain computing resources, such as a cloud server, a personal computer (Personal Computer, PC, referring to a multi-purpose computer suitable for personal use in terms of size, price, and performance; desktop computers, laptops, small laptops, tablets, and ultrabooks, etc. all belong to personal computers), a smart phone, a personal digital assistant (Personal Digital Assistant, PDA), or a wearable device, etc. As Figure 1 shown, the e-commerce product precise recommendation method can be, but is not limited to, including the following steps S1 to S75.

[0072] S1. Collect the e-commerce product browsing record data and e-commerce product purchase record data generated during each historical startup and operation of the e-commerce platform application by the target customer.

[0073] In step S1, the e-commerce platform application refers to an APP (Application) program that can be started and run on a customer terminal (such as a smart phone or a tablet computer, etc.) to perform online purchases of e-commerce products. Specific examples include, but are not limited to, Taobao APP, JD.com APP, or Pinduoduo APP, etc. The e-commerce product browsing record data is used to reflect the e-commerce product browsing behavior of the target customer during each historical startup and operation of the e-commerce platform application, and the e-commerce product purchase record data is used to reflect the e-commerce product purchase behavior of the target customer during each historical startup and operation of the e-commerce platform application. They can both be routinely recorded and uploaded and collected by the e-commerce platform application through existing technologies.

[0074] S2. Process the e-commerce product browsing record data to obtain the browsing event indication value or browsing time occupancy ratio of various e-commerce products during each historical startup and operation, and process the e-commerce product purchase record data to obtain the purchase event indication value of various e-commerce products during each historical startup and operation. Among them, the browsing event indication value is used to indicate whether an event of the corresponding e-commerce product being browsed by the target customer occurs during the startup and operation of the e-commerce platform application, the browsing time occupancy ratio refers to the ratio of the duration of the target customer browsing the corresponding e-commerce product to the total duration of browsing all e-commerce products during the startup and operation of the e-commerce platform application, and the purchase event indication value is used to indicate whether an event of the corresponding e-commerce product being purchased by the target customer occurs during the startup and operation of the e-commerce platform application.

[0075] In the step S2, the specific processing method for the e-commerce product browsing record data and the e-commerce product purchase record data is the existing conventional data statistics method; for example, if there are e-commerce products of type A, type B, type C, type D, type E, and type F, and the e-commerce product browsing record data recorded during a certain historical startup and operation of the e-commerce platform application reflects that the target user browsed the e-commerce product of type B (the corresponding browsing duration is 3 minutes), the e-commerce product of type C (the corresponding browsing duration is 5 minutes), and the e-commerce product of type F (the corresponding browsing duration is 2 minutes), and the e-commerce product purchase record data recorded during the certain historical startup and operation reflects that the target user purchased the e-commerce product of type C (which is generally one of the browsed products), then it can be processed to obtain that the browsing event indication values of the e-commerce products of type A, type B, type C, type D, type E, and type F during the certain historical startup and operation are 0, 1, 1, 0, 0, and 1 in sequence (that is, for example but not limited to: using 0 to indicate that the event of the corresponding e-commerce product being browsed by the target customer did not occur during the certain historical startup and operation, and using 1 to indicate that the event of the corresponding e-commerce product being browsed by the target customer occurred during the certain historical startup and operation), or it can be processed to obtain that the browsing time proportion values of the e-commerce products of type A, type B, type C, type D, type E, and type F during the certain historical startup and operation are 0, 30%, 50%, 0, 0, and 20% in sequence, and it can also be processed to obtain that the purchase event indication values of the e-commerce products of type A, type B, type C, type D, type E, and type F during the certain historical startup and operation are 0, 0, 1, 0, 0, and 0 in sequence (that is, for example but not limited to: using 0 to indicate that the event of the corresponding e-commerce product being purchased by the target customer did not occur during the certain historical startup and operation, and using 1 to indicate that the event of the corresponding e-commerce product being purchased by the target customer occurred during the certain historical startup and operation). In addition, if each of the historical startup and operation periods is regarded as a time node, then for each type of e-commerce product, the browsing event indication values or browsing time proportion values arranged in sequence according to the time order and corresponding to each historical startup and operation period will constitute the corresponding time series data, and the purchase event indication values arranged in sequence according to the time order and corresponding to each historical startup and operation period will constitute another corresponding time series data.

[0076] S3. For the various types of e-commerce products, the corresponding sample data is extracted as follows: The browsing event indication value or the browsing time ratio during the continuous m historical startup runs of all e-commerce products is used as the model input item, and the purchase event indication value during the last historical startup run among the continuous m historical startup runs of the corresponding e-commerce product is used as the model output item. Then, the model input item and the model output item are used as a sample data, where m represents a positive integer greater than or equal to 2.

[0077] In step S3, based on the example in step S2, for the type B e-commerce products, the browsing event indication value or the browsing time ratio of the type A e-commerce products, the type B e-commerce products, the type C e-commerce products, the type D e-commerce products, the type E e-commerce products, and the type F e-commerce products during the continuous m historical startup runs can be used as the model input item (specifically a 6×m matrix), and the purchase event indication value during the last historical startup run among the continuous m historical startup runs of the type B e-commerce products is used as the model output item (specifically 0 or 1). In addition, the specific value of m can be preset according to actual requirements or optimization algorithms, for example, designed to be 5.

[0078] S4. For the various types of e-commerce products, all the corresponding sample data is used to calibrate and verify the machine learning model based on the artificial intelligence algorithm to obtain the corresponding purchase event prediction model.

[0079] In the step S4, the artificial intelligence algorithm is a core artificial intelligence algorithm that specializes in studying how a computer simulates or implements human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve its own performance. It is the fundamental way to make a computer intelligent. Specifically, the machine learning model can but is not limited to adopting an LSTM model, a Bi-LSTM model, an Attention-LSTM model, etc. The aforementioned LSTM (Long Short-Term Memory) model, Bi-LSTM model, Attention-LSTM model, etc. are all existing machine learning models. Among them, the Attention-LSTM model is an improved model that adds an attention mechanism to the existing LSTM model (which is commonly used for time series data prediction). The added attention mechanism is used to allow the LSTM model to dynamically focus on partial information during training so as to be able to capture more important information and improve the performance of the model. Therefore, in this embodiment, the Attention-LSTM model is preferably adopted to train the purchase event prediction model. The specific calibration, verification, and modeling process of the machine learning model (where model calibration is to adjust parameters or structures to make the model fit the actual data; and model verification is to evaluate the accuracy and applicability of the model by comparing the model output with the actual data) is an existing technical means. Specifically, the parameters of the machine learning model can but are not limited to be obtained by using a grid search algorithm or a Bayesian search algorithm.

[0080] S5. For each type of e-commerce product, import the browsing event indication value or the browsing time occupancy ratio of all e-commerce products during the current most recent consecutive m - 1 historical startup runs and the e-commerce product recommendation plan of the e-commerce platform application during the next startup run into the corresponding purchase event prediction model, and output the confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup run. Among them, the e-commerce product recommendation plan includes but is not limited to the recommendation event indication value or the recommendation frequency occupancy ratio of each type of e-commerce product during the next startup run, etc. The recommendation event indication value is used to indicate whether an event of recommending the corresponding e-commerce product to the target customer for browsing is triggered during the next startup run. The recommendation frequency occupancy ratio refers to the ratio of the frequency of recommending the corresponding e-commerce product to the target customer for browsing during the next startup run to the total frequency of recommending all e-commerce products to the target customer for browsing.

[0081] In the step S5, the e-commerce product recommendation plan is used as an e-commerce product recommendation program, which can be randomly generated or directionally generated by the e-commerce platform service provider. Among them, the recommendation event indication value is used to be equivalent to the browsing event indication value (that is, because the e-commerce product will surely be browsed by the target customer after being recommended, and for example but not limited to: 0 is used to indicate that the corresponding e-commerce product will not be recommended to the target customer for browsing during the next startup and operation, and 1 is used to indicate that the corresponding e-commerce product will be recommended to the target customer for browsing during the next startup and operation). The recommended frequency ratio is used to be equivalent to the browsing time ratio (that is, because the recommended frequency must be positively correlated with the browsing duration). Based on the example in step S2, for the category B e-commerce products, the browsing event indication values or browsing time ratios of the category A, category B, category C, category D, category E, and category F e-commerce products during the current most recent consecutive m - 1 historical startup and operation periods, as well as the e-commerce product recommendation plan of the e-commerce platform application during the next startup and operation period (this plan is a 6×1 matrix, and thus the entire imported data is a 6×m matrix) are imported into the corresponding purchase event prediction model, and the confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup and operation period is output (that is, the confidence level of outputting the value 1). Specifically, when the purchase event prediction model is calibrated, verified, and modeled based on the browsing event indication value and the purchase event indication value, the data imported into the purchase event prediction model in step S5 is specifically the browsing event indication values of all e-commerce products during the current most recent consecutive m - 1 historical startup and operation periods and the recommendation event indication values of each category of e-commerce products during the next startup and operation period. When the purchase event prediction model is calibrated, verified, and modeled based on the browsing time ratio and the purchase event indication value, the data imported into the purchase event prediction model in step S5 is the browsing time ratios of all e-commerce products during the current most recent consecutive m - 1 historical startup and operation periods and the recommended frequency ratios of each category of e-commerce products during the next startup and operation period. In addition, the aforementioned step S5 and the subsequent step S6 are specifically executed between the end time of the current most recent historical startup and operation period and the start time of the next startup and operation period.

[0082] S6. Determine whether the sum of the confidence levels of each category of e-commerce products exceeds the first preset threshold.

[0083] In the step S6, the sum of the confidence levels of the various types of e-commerce products reflects the total purchase willingness of the target customer in the recommendation scenarios of multiple types of e-commerce products. The higher the value, the stronger the total purchase willingness and the higher the accuracy of product recommendation. The first preset threshold is used as the basis for judging whether the recommendation accuracy of the e-commerce product recommendation scheme meets the preset goal, and it can be set in advance by the platform service provider. In addition, the foregoing step S6 can also be replaced by: judging whether the average value of the confidence levels of the various types of e-commerce products exceeds another preset threshold (which can also be set in advance by the platform service provider).

[0084] S7. If so, during the next startup and operation, execute the e-commerce product recommendation scheme through the e-commerce platform application program.

[0085] In the step S7, based on the example in step S2, if the specific content of the e-commerce product recommendation scheme includes: the recommendation event indication values of the type A e-commerce products, the type B e-commerce products, the type C e-commerce products, the type D e-commerce products, the type E e-commerce products, and the type F e-commerce products during the next startup and operation are 1, 0, 0, 1, 0, and 0 respectively, then during the next startup and operation, executing the e-commerce product recommendation scheme through the e-commerce platform application program specifically includes: during the next startup and operation, only recommend the type A e-commerce products and the type D e-commerce products through the e-commerce platform application program. And if the specific content of the e-commerce product recommendation scheme includes: the recommendation frequency ratios of the type A e-commerce products, the type B e-commerce products, the type C e-commerce products, the type D e-commerce products, the type E e-commerce products, and the type F e-commerce products during the next startup and operation are 10%, 30%, 30%, 30%, 0, and 0 respectively, then during the next startup and operation, executing the e-commerce product recommendation scheme through the e-commerce platform application program specifically includes: among every 10 recommended e-commerce products, 1 e-commerce product is of the type A, 3 e-commerce products are of the type B, 3 e-commerce products are of the type C, and 3 e-commerce products are of the type D, or when recommending e-commerce products each time, recommend the type A e-commerce products with a probability of 10%, recommend the type B e-commerce products with a probability of 30%, recommend the type C e-commerce products with a probability of 30%, and recommend the type D e-commerce products with a probability of 30%.

[0086] Based on the e-commerce product precise recommendation method described in the foregoing steps S1 to S7, a new solution for precise recommendation of e-commerce products in multiple types of e-commerce product recommendation scenarios is provided. That is, first, based on the e-commerce product browsing record data and the e-commerce product purchase record data, sample data of various types of e-commerce products are extracted, and a purchase event prediction model for each type of e-commerce product is trained based on the sample data. Then, based on the prediction model and the e-commerce product recommendation scheme, the confidence level of the event that various types of e-commerce products are purchased by the target customer during the next startup and operation is estimated. Finally, when it is determined that the sum of the confidence levels of various types of e-commerce products exceeds a preset threshold, it is confirmed that the recommendation accuracy of the recommendation scheme meets the preset goal, and during the next startup and operation, the recommendation scheme is executed through the e-commerce platform application program. In this way, not only the impact of recommending a single e-commerce product on the customer's purchase decision is considered, but also the situation where recommending different types of e-commerce products will have a competitive or linkage impact on the customer's purchase decision is considered, so that when applied to multiple types of e-commerce product recommendation scenarios, the user's purchase intention can be effectively improved, the purpose of further improving the accuracy of product recommendation is achieved, and it is convenient for practical application and promotion.

[0087] Based on the technical solution of the foregoing first aspect, this embodiment also provides a possible design 1 for obtaining the linkage purchase influence coefficient between e-commerce products and performing confidence correction, that is, the method further includes, but is not limited to, the following step S8.

[0088] S8. For each type of e-commerce product, calculate the linkage purchase influence coefficient between the corresponding e-commerce product and other types of e-commerce products respectively according to the purchase event indication value of the corresponding e-commerce product during each historical startup and operation period and the purchase event indication value of other types of e-commerce products during each historical startup and operation period.

[0089] In step S8, based on the example of step S2, for the B-type e-commerce product, the A-type e-commerce product, the C-type e-commerce product, the D-type e-commerce product, the E-type e-commerce product, and the F-type e-commerce product are respectively the corresponding other types of e-commerce products. Specifically, for each type of e-commerce product, calculate the linkage purchase influence coefficient between the corresponding e-commerce product and other types of e-commerce products respectively according to the purchase event indication value of the corresponding e-commerce product during each historical startup and operation period and the purchase event indication value of other types of e-commerce products during each historical startup and operation period, including but not limited to the following steps S81 to S85.

[0090] S81. For a certain type of e-commerce product, extract the purchase event indication values of the corresponding e-commerce product during the current most recent consecutive M historical startup and operation periods to form a first sample S1, where M represents a positive integer greater than m.

[0091] In step S81, if m takes the value of 5, M can be exemplified as 50.

[0092] S82. For other certain types of e-commerce products relative to the certain type of e-commerce products, extract the purchase event indication values of the corresponding e-commerce products during the current most recent consecutive M historical startup operations to form the second sample S2.

[0093] S83. Perform a normal distribution KS test on the first sample S1 to calculate the first test statistic p-value p1, and also perform the normal distribution KS test on the second sample S2 to calculate the second test statistic p-value p2.

[0094] S84. If both the first test statistic p-value p1 and the second test statistic p-value p2 are greater than the second first preset threshold, calculate the correlation coefficient r between the first sample S1 and the sample S2 according to the following formula q :

[0095]

[0096] In the formula, i represents a positive integer, x i represents the i-th sample value in the first sample S1, and y i represents the i-th sample value in the second sample S2. represents the sample mean of the first sample S1. represents the sample mean of the second sample S2.

[0097] In step S84, the second preset threshold can be exemplified as 0.05.

[0098] S85. Take the correlation coefficient r q as the co-purchase influence coefficient between the certain type of e-commerce products and the other certain type of e-commerce products.

[0099] Judging whether the total confidence of all types of e-commerce products exceeds the first preset threshold includes, but is not limited to, the following steps S61 - S62.

[0100] S61. For all types of e-commerce products, calculate the corresponding confidence correction value according to the corresponding confidence and the co-purchase influence coefficient between the corresponding e-commerce products and other types of e-commerce products according to the following formula:

[0101]

[0102] In the formula, N represents the total number of categories of e-commerce products, i represents a positive integer less than or equal to N, c i represents the confidence of the i-th type of e-commerce products. denotes the confidence correction value of the i-th type of e-commerce product, where i' denotes a positive integer less than or equal to N - 1, and c i,i′ denotes the confidence of other i'-th type of e-commerce products relative to the i-th type of e-commerce product, η i,i′ denotes the co-purchase influence coefficient between the i-th type of e-commerce product and the other i'-th type of e-commerce products.

[0103] S62. Determine whether the sum of the confidence correction values of all types of e-commerce products exceeds a first preset threshold.

[0104] Based on the foregoing, it is possible to design a method. By calculating the co-purchase influence coefficient between e-commerce products based on the historical purchase situation of various types of e-commerce products and performing confidence correction, it is possible to improve the estimation accuracy of the user's purchase intention and further improve the accuracy of product recommendation when further considering the co-purchase between different types of e-commerce products.

[0105] Based on the technical solution of the foregoing first aspect or possible design one, this embodiment further provides a possible design two for how to adjust the recommendation scheme and re-evaluate it when the recommendation accuracy of the recommendation scheme does not meet the preset goal, that is, after determining whether the sum of the confidence levels of all types of e-commerce products exceeds a first preset threshold, if it is determined that the sum of the confidence levels of all types of e-commerce products does not exceed the first preset threshold, then as Figure 2 shown, the method further includes but is not limited to the following steps S701 to S704.

[0106] S701. Adjust the current e-commerce product recommendation scheme to obtain a new e-commerce product recommendation scheme, and then execute step S702.

[0107] In step S701, the specific adjustment method of the e-commerce product recommendation scheme is an existing conventional method, such as randomly modifying the corresponding recommendation event indication value or recommendation frequency ratio of a certain type of e-commerce product.

[0108] S702. For all types of e-commerce products, import the browsing event indication value or browsing time ratio of all e-commerce products during the current most recent consecutive m - 1 historical startup runs and the new e-commerce product recommendation scheme into the corresponding purchase event prediction model, and output the new confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup run, and then execute step S703.

[0109] In step S702, the specific technical details can be referred to step S5 above and will not be elaborated here.

[0110] S703. Determine whether the sum of the new confidence levels of the various types of e-commerce products exceeds the first preset threshold. If so, execute step S704; otherwise, execute step S701.

[0111] S704. During the current next startup and operation, execute the new e-commerce product recommendation plan through the e-commerce platform application.

[0112] Based on the foregoing possible design two, when the recommendation accuracy of the recommendation plan does not meet the preset target, the e-commerce product recommendation plan can be continuously adjusted until the recommendation accuracy meets the preset target, thereby ensuring the e-commerce product recommendation effect.

[0113] Based on the technical solution of the foregoing first aspect or possible design one, this embodiment further provides a possible design three on how to preferentially select a recommendation plan when there are multiple e-commerce product recommendation plans, that is, when there are multiple e-commerce product recommendation plans, as Figure 3 shown, the method further includes but is not limited to the following steps S91 to S94.

[0114] S91. For each type of e-commerce product and each e-commerce product recommendation plan among multiple e-commerce product recommendation plans, input the browsing event indication value or browsing time ratio of all e-commerce products during the current most recent consecutive m - 1 historical startup and operation periods and the corresponding plan into the purchase event prediction model of the corresponding e-commerce product, and output the event that the corresponding e-commerce product is purchased by the target customer during the current next startup and operation period and the corresponding confidence level.

[0115] In step S91, for specific technical details, refer to the foregoing step S5, which will not be elaborated here.

[0116] S92. For each e-commerce product recommendation plan, accumulate and calculate the confidence levels of each type of e-commerce product and obtain the corresponding total confidence level value.

[0117] S93. Determine whether the maximum total confidence level value exceeds the first preset threshold.

[0118] S94. If so, during the current next startup and operation, execute, through the e-commerce platform application, an e-commerce product recommendation plan among multiple e-commerce product recommendation plans that corresponds to the maximum total confidence level value.

[0119] Based on the foregoing possible design three, when there are multiple e-commerce product recommendation plans, the recommendation plan can be preferentially selected to maximize the recommendation accuracy and meet the preset target, thereby ensuring the best e-commerce product recommendation effect.

[0120] Based on the technical solutions of the foregoing first aspect or possible design one, this embodiment further provides a possible design four for obtaining an optimal e-commerce product recommendation solution, that is, the method further includes but is not limited to the following steps S10 to S11.

[0121] S10. Apply the browsing event indication values or browsing time ratios of all the e-commerce products during the current most recent consecutive m-1 historical startup runs and the purchase event prediction models of various types of e-commerce products, and optimize the e-commerce product recommendation solution based on an optimization algorithm to obtain the e-commerce product recommendation solution and the optimal search result for maximizing the sum of the confidence levels of various types of e-commerce products.

[0122] In step S10, specifically, the optimization algorithm can be but is not limited to using a particle swarm optimization algorithm, a Newton optimization algorithm, a genetic optimization algorithm, a grey wolf optimization algorithm, a whale optimization algorithm, or a tuna swarm optimization algorithm, etc. Considering that different optimization algorithms have different advantages and disadvantages, in order to comprehensively utilize the performance of the foregoing optimization algorithms to achieve the purpose of making the best use of advantages and avoiding disadvantages, preferably, apply the browsing event indication values or browsing time ratios of all the e-commerce products during the current most recent consecutive m-1 historical startup runs and the purchase event prediction models of various types of e-commerce products, and optimize the e-commerce product recommendation solution based on an optimization algorithm to obtain the e-commerce product recommendation solution and the optimal search result for maximizing the sum of the confidence levels of various types of e-commerce products, including but not limited to the following steps S101 to S116.

[0123] S101. Initialize the optimization algorithm parameters including the number of search individuals I in the search population, the maximum number of iterations T, the first learning factor α1, the second learning factor α2, the third learning factor α3, the fourth learning factor α4, the fifth learning factor α5, and the sixth learning factor α6, and randomly generate the initial positions of each search individual in the search population, and then execute step S102, where the initial positions of each search individual are randomly generated based on the following formula:

[0124]

[0125] In the formula, i′ represents a positive integer less than or equal to I, d′ represents a positive integer less than or equal to D′, D′ represents the number of dimensions of the parameter vector to be optimized, and the parameter vector to be optimized includes the e-commerce product recommendation solution. represents the component of the initial position of the i′-th search individual in the search population on the d′-th dimension of the parameter vector to be optimized, u c,d′ represents the upper limit of the parameter search space on the d′-th dimension, l c,d′represents the lower limit of the parameter search space in the d′th dimension, and rand(0,1) represents a pure decimal random generator function.

[0126] S102. For each search individual, the corresponding initial position is used as the e-commerce product recommendation scheme, and then the browsing event indication values or browsing time proportion values of all the e-commerce products during the current most recent m-1 consecutive historical startup operations and the e-commerce product recommendation scheme are respectively imported into the purchase event prediction model of each type of e-commerce product, and the first confidence level of the event that the each type of e-commerce product is purchased by the target customer during the current next startup operation is outputted. Finally, the cumulative sum of each of the first confidence levels corresponding one-to-one to the each type of e-commerce product is used as the corresponding fitness level, and then step S103 is executed.

[0127] In step S102, the specific technical details can be found in the aforementioned step S5 and will not be repeated here. In addition, considering that some values in the e-commerce product recommendation scheme (such as the recommended event indicator value) need to be integers, when the initial position is used as the e-commerce product recommendation scheme, it is necessary to round the corresponding values, such as rounding the recommended event indicator value to the nearest integer.

[0128] S103. Take the initial position of a search individual with the maximum fitness as the initial global optimal position x best , and also initialize and set the current iteration number t′=0, and then execute step S104.

[0129] S104. Calculate the current average position of the search population based on the current positions of the individual searchers, and determine a corresponding first new position for each individual searcher, then proceed to step S105, wherein the first new position of each individual searcher is determined according to the following formula:

[0130] x new1,i′,d′ =x best,d′ +α1×rand(0,1)×(x mean,d′ -x i′,d′ )

[0131] Where x new1,i′,d′ represents the component of the first new position of the i′th search individual in the d′th dimension, x best,d′ Denotes the global optimal position x best The component in the d′th dimension, x mean,d′ represents the component of the current average position of the search population in the d′th dimension, x i′,d′ Represents the component of the current position of the i′th search individual in the d′th dimension.

[0132] S105. For each of the search individuals, use the corresponding first new position as the e-commerce product recommendation solution. Then, import the browsing event indication values or browsing time ratio values of all the e-commerce products during the current most recent consecutive m-1 historical startup operations and the e-commerce product recommendation solution into the purchase event prediction models of the respective types of e-commerce products, and output the second confidence levels of the events that the respective types of e-commerce products are purchased by the target customer during the next startup operation. Finally, use the sum of the accumulations of the respective second confidence levels corresponding to the respective types of e-commerce products as the corresponding new fitness, and then execute step S106.

[0133] In step S105, for the specific technical details, reference can be made to the aforementioned step S5, which will not be elaborated here. In addition, considering that some values in the e-commerce product recommendation solution (such as the recommendation event indication value) need to be integers, when using the first new position as the e-commerce product recommendation solution, it is necessary to perform rounding processing on the corresponding values. For example, perform rounding to the nearest integer on the recommendation event indication value.

[0134] S106. Determine whether the fitness corresponding to the current position of the i'-th search individual is less than the current fitness of the i'-th search individual. If so, update the current position of the i'-th search individual to the first new position of the i'-th search individual, and then execute step S107. Otherwise, directly execute step S107.

[0135] S107. Determine whether the fitness corresponding to the global optimal position x best is less than the minimum value among the current fitnesses of all the search individuals. If so, update the global optimal position x best to the current position of any search individual with this minimum value, and then execute step S108. Otherwise, directly execute step S108.

[0136] S108. Update and calculate the current average position of the search population based on the current positions of all the search individuals, and determine the corresponding second new positions for all the search individuals, and then execute step S109. Among them, the second new positions of all the search individuals are determined according to the following formula:

[0137]

[0138] where x new2,i′,d′represents the component of the second new position of the \(i'\)-th search individual in the \(d'\)-th dimension, \(\beta(i')\) represents the first intermediate variable corresponding to the \(i'\)-th search individual, \(\delta(i')\) represents the second intermediate variable corresponding to the \(i'\)-th search individual, \(i''\) represents a positive integer less than or equal to \(I\), \(mod()\) represents the modulo function, \(x\) i″,d′ represents the component of the current position of the \(i''\)-th search individual in the search population in the \(d'\)-th dimension, \(\beta_r(i')\) represents the third intermediate variable corresponding to the \(i'\)-th search individual, \(\delta_r(i')\) represents the fourth intermediate variable corresponding to the \(i'\)-th search individual, \(max()\) represents the maximum value function, \(\theta(i')\) represents the fifth intermediate variable corresponding to the \(i'\)-th search individual, \(r(i')\) represents the sixth intermediate variable corresponding to the \(i'\)-th search individual, and \(\pi\) represents 180 degrees.

[0139] S109. For each of the search individuals, use the corresponding second new position as the e-commerce product recommendation plan. Then, import the browsing event indication value or the browsing time ratio of all the e-commerce products during the current most recent consecutive \(m - 1\) historical startup runs and this e-commerce product recommendation plan into the purchase event prediction models of the various types of e-commerce products, and output the third confidence levels of the events that the various types of e-commerce products are purchased by the target customer during the next startup run. Finally, use the sum of the accumulations of the respective third confidence levels corresponding to the various types of e-commerce products as the corresponding new fitness, and then execute step S110.

[0140] In step S109, for the specific technical details, reference can be made to the aforementioned step S5, which will not be elaborated here. In addition, considering that some values in the e-commerce product recommendation plan (such as the recommendation event indication value) need to be integers, when using the second new position as the e-commerce product recommendation plan, it is necessary to round the corresponding values. For example, round the recommendation event indication value by the method of rounding.

[0141] S110. Determine whether the fitness corresponding to the current position of the \(i'\)-th search individual is less than the current fitness of the \(i'\)-th search individual. If so, update the current position of the \(i'\)-th search individual to the second new position of the \(i'\)-th search individual, and then execute step S111. Otherwise, directly execute step S111.

[0142] S111. Determine whether the fitness corresponding to the global optimal position \(x\) best is less than the minimum value among the current fitnesses of all the search individuals. If so, update the global optimal position \(x\) best to the current position of any search individual with this minimum value, and then execute step S112. Otherwise, directly execute step S112.

[0143] S112. Update and calculate the current average position of the search population based on the current positions of the respective search individuals, and for each of the search individuals, determine a corresponding third new position, and then execute step S113, wherein the third new position of each of the search individuals is determined according to the following formula:

[0144]

[0145] In the formula, x new3,i′,d′ represents the component of the third new position of the i'-th search individual in the d'-th dimension, represents the seventh intermediate variable corresponding to the i'-th search individual, represents the eighth intermediate variable corresponding to the i'-th search individual, represents the ninth intermediate variable corresponding to the i'-th search individual, represents the tenth intermediate variable corresponding to the i'-th search individual, represents the eleventh intermediate variable corresponding to the i'-th search individual.

[0146] S113. For each of the search individuals, use the corresponding third new position as the e-commerce product recommendation plan, and then import the browsing event indication values or browsing time ratios of all the e-commerce products during the current most recent consecutive m - 1 historical startup runs and the e-commerce product recommendation plan into the purchase event prediction models of the respective types of e-commerce products, and output the fourth confidence levels of the events that the respective types of e-commerce products are purchased by the target customer during the next startup run, and finally use the sum of the accumulations of the respective fourth confidence levels corresponding to the respective types of e-commerce products as the corresponding new fitness, and then execute step S114.

[0147] In step S113, for specific technical details, reference can be made to the aforementioned step S5, which will not be elaborated here. In addition, considering that some values in the e-commerce product recommendation plan (such as the recommendation event indication value) need to be integers, so when using the third new position as the e-commerce product recommendation plan, it is necessary to round the corresponding values, for example, round the recommendation event indication value by rounding.

[0148] S114. Determine whether the fitness corresponding to the current position of the i'-th search individual is less than the current fitness of the i'-th search individual. If so, update the current position of the i'-th search individual to the third new position of the i'-th search individual, and then execute step S115; otherwise, directly execute step S115.

[0149] S115. Determine whether the fitness corresponding to the global optimal position xbest whether the corresponding fitness is less than the minimum value among the current fitness values of the respective search individuals. If so, update the global optimal position x best to the current position of any search individual having this minimum value, and then execute step S116. Otherwise, directly execute step S116.

[0150] S116. Increment the current iteration number t′ by 1, and determine whether the current iteration number t′ has reached the maximum iteration number T. If so, use the global optimal position x best as the e-commerce product recommendation solution and the optimal search result for maximizing the sum of the confidence levels of various e-commerce products, otherwise return to execute step S104.

[0151] Based on the foregoing steps S101 - S116, since in each iteration process, each search individual performs three fitness calculations by means of a spiral and the group center respectively, it can be regarded as a fusion of the particle swarm optimization algorithm and the whale optimization algorithm. Therefore, under all the same conditions, its optimization performance will be better than other optimization algorithms, which is conducive to quickly and accurately obtaining the e-commerce product recommendation solution and the optimal search result for maximizing the sum of the confidence levels of various e-commerce products.

[0152] S11. During the next startup and operation, execute the e-commerce product recommendation solution obtained based on the optimal search result through the e-commerce platform application program.

[0153] In step S11, considering that some values in the e-commerce product recommendation solution (such as the recommendation event indication value) need to be integers, therefore, in the process of obtaining the e-commerce product recommendation solution based on the optimal search result, it is necessary to round the corresponding values, for example, perform rounding on the recommendation event indication value.

[0154] Based on the foregoing possible design four, it is also possible to optimize and obtain the optimal e-commerce product recommendation solution to maximize the recommendation accuracy, thereby ensuring the best e-commerce product recommendation effect. In addition, the foregoing steps S10 - S11 can be specifically executed when it is determined that the sum of the confidence levels / maximum confidence level value of various e-commerce products does not exceed the first preset threshold, so as to optimize and obtain the optimal e-commerce product recommendation solution.

[0155] As Figure 4 shown, in the second aspect of this embodiment, a virtual device for implementing the e-commerce product precise recommendation method described in the first aspect or any possible design in the first aspect is provided, including a recorded data collection unit, a recorded data processing unit, a sample data extraction unit, a prediction model training unit, a prediction model application unit, a threshold comparison and judgment unit, and a recommendation scheme execution unit that are sequentially communicatively connected;

[0156] The recorded data collection unit is used to collect the e-commerce product browsing record data and e-commerce product purchase record data generated by the target customer during each historical startup and operation of the e-commerce platform application;

[0157] The recorded data processing unit is used to process the e-commerce product browsing record data to obtain the browsing event indication value or browsing time ratio of various e-commerce products during each historical startup and operation, and process the e-commerce product purchase record data to obtain the purchase event indication value of various e-commerce products during each historical startup and operation. Among them, the browsing event indication value is used to indicate whether the event that the corresponding e-commerce product is browsed by the target customer occurs during the startup and operation of the e-commerce platform application. The browsing time ratio refers to the ratio of the duration of the target customer browsing the corresponding e-commerce product to the total duration of browsing all e-commerce products during the startup and operation of the e-commerce platform application. The purchase event indication value is used to indicate whether the event that the corresponding e-commerce product is purchased by the target customer occurs during the startup and operation of the e-commerce platform application;

[0158] The sample data extraction unit is used to extract the corresponding sample data for each type of e-commerce product in the following manner: taking the browsing event indication value or browsing time ratio of all e-commerce products during consecutive m historical startup and operation periods as the model input item, and taking the purchase event indication value of the corresponding e-commerce product during the last historical startup and operation period among the consecutive m historical startup and operation periods as the model output item, and then taking the model input item and the model output item as a sample data, where m represents a positive integer greater than or equal to 2;

[0159] The prediction model training unit is used to calibrate, verify and model the machine learning model based on the artificial intelligence algorithm for each type of e-commerce product by applying all the corresponding sample data to obtain the corresponding purchase event prediction model;

[0160] The prediction model application unit is configured to, for each type of e-commerce product, import the browsing event indication value or browsing time ratio of all e-commerce products during the current most recent consecutive m-1 historical startup operations and the e-commerce product recommendation plan of the e-commerce platform application during the next startup operation into the corresponding purchase event prediction model, and output the confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup operation. Wherein, the e-commerce product recommendation plan includes the recommendation event indication value or recommendation frequency ratio of each type of e-commerce product during the next startup operation. The recommendation event indication value is used to indicate whether an event of recommending the corresponding e-commerce product to the target customer for browsing is triggered during the next startup operation. The recommendation frequency ratio refers to the ratio of the frequency of recommending the corresponding e-commerce product to the target customer for browsing to the total frequency of recommending all e-commerce products to the target customer for browsing during the next startup operation;

[0161] The threshold comparison and judgment unit is configured to judge whether the sum of the confidence levels of each type of e-commerce product exceeds a first preset threshold;

[0162] The recommendation plan execution unit is configured to, when it is determined that the sum of the confidence levels of each type of e-commerce product exceeds the first preset threshold, execute the e-commerce product recommendation plan through the e-commerce platform application during the next startup operation.

[0163] For the working process, working details and technical effects of the foregoing device provided in the second aspect of this embodiment, reference can be made to the e-commerce product precise recommendation method described in the first aspect or any possible design in the first aspect, which will not be elaborated herein.

[0164] Such as Figure 5As shown, in the third aspect of this embodiment, a computer device for executing the accurate recommendation method for e-commerce products described in the first aspect or any possible design in the first aspect is provided, including a memory, a processor, and a transceiver that are communicatively connected in sequence. Among them, the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the accurate recommendation method for e-commerce products described in the first aspect or any possible design in the first aspect. Specifically, for example, the memory may but is not limited to including a random access memory (RAM), a read-only memory (ROM), a flash memory, a first input first output (FIFO), and / or a first input last output (FILO), etc.; the processor may but is not limited to adopt a microprocessor of the STM32F105 series. In addition, the computer device may but is not limited to include a power module, a display screen, and other necessary components.

[0165] For the working process, working details, and technical effects of the aforementioned computer device provided in the third aspect of this embodiment, reference may be made to the accurate recommendation method for e-commerce products described in the first aspect or any possible design in the first aspect, which will not be elaborated here.

[0166] In the fourth aspect of this embodiment, a computer-readable storage medium storing instructions including the accurate recommendation method for e-commerce products described in the first aspect or any possible design in the first aspect is provided, that is, instructions are stored on the computer-readable storage medium, and when the instructions are run on a computer, the accurate recommendation method for e-commerce products described in the first aspect or any possible design in the first aspect is executed. Among them, the computer-readable storage medium refers to a carrier for storing data, and may but is not limited to include computer-readable storage media such as floppy disks, optical discs, hard disks, flash memories, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a dedicated computer, a computer network, or other programmable devices.

[0167] For the working process, working details, and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment, reference may be made to the accurate recommendation method for e-commerce products described in the first aspect or any possible design in the first aspect, which will not be elaborated here.

[0168] The fifth aspect of this embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, implement the precise recommendation method for e-commerce products as described in the first aspect or any possible design in the first aspect. Among them, the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0169] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for precise recommendation of e-commerce products, characterized in that, Including: Collecting e-commerce product browsing record data and e-commerce product purchase record data generated during each historical startup and operation of the e-commerce platform application by the target customers; Processing the e-commerce product browsing record data to obtain browsing event indication values or browsing time ratio values of various e-commerce products during each historical startup and operation, and processing the e-commerce product purchase record data to obtain purchase event indication values of various e-commerce products during each historical startup and operation, wherein the browsing event indication value is used to indicate whether an event that the corresponding e-commerce product is browsed by the target customer occurs during the startup and operation of the e-commerce platform application, the browsing time ratio value is the ratio of the duration of the target customer browsing the corresponding e-commerce product to the total duration of browsing all e-commerce products during the startup and operation of the e-commerce platform application, and the purchase event indication value is used to indicate whether an event that the corresponding e-commerce product is purchased by the target customer occurs during the startup and operation of the e-commerce platform application; For the various e-commerce products, extracting corresponding sample data in the following manner: taking the browsing event indication values or browsing time ratio values of all e-commerce products during consecutive m historical startup and operation periods as model input items, and taking the purchase event indication value of the corresponding e-commerce product during the last historical startup and operation period among the consecutive m historical startup and operation periods as the model output item, and then taking the model input item and the model output item as a sample data, where m represents a positive integer greater than or equal to 2; For the various e-commerce products, calibrating, verifying and modeling a machine learning model based on an artificial intelligence algorithm by using all the corresponding sample data to obtain a corresponding purchase event prediction model; For the various e-commerce products, inputting the browsing event indication values or browsing time ratio values of all e-commerce products during the current most recent consecutive m-1 historical startup and operation periods and the e-commerce product recommendation scheme during the next startup and operation of the e-commerce platform application into the corresponding purchase event prediction model, and outputting the confidence level of an event that the corresponding e-commerce product is purchased by the target customer during the next startup and operation, wherein the e-commerce product recommendation scheme includes recommendation event indication values or recommendation frequency ratio values of various e-commerce products during the next startup and operation, the recommendation event indication value is used to indicate whether an event that the corresponding e-commerce product is recommended to the target customer for browsing is triggered during the next startup and operation, and the recommendation frequency ratio value is the ratio of the frequency of the corresponding e-commerce product being recommended to the target customer for browsing to the total frequency of all e-commerce products being recommended to the target customer for browsing during the next startup and operation; Judging whether the sum of the confidence levels of the various e-commerce products exceeds a first preset threshold; If so, during the next startup and operation, executing the e-commerce product recommendation scheme through the e-commerce platform application; 2. The e-commerce product precise recommendation method according to claim 1, wherein The method further includes: For each type of e-commerce product, based on the purchase event indication values of the corresponding e-commerce product during each historical startup operation and the purchase event indication values of other types of e-commerce products during each historical startup operation, calculate the co-purchase influence coefficients of the corresponding e-commerce product and other types of e-commerce products respectively; Determine whether the sum of the confidence levels of each type of e-commerce product exceeds a first preset threshold, including: For each type of e-commerce product, calculate the corresponding confidence level correction value according to the corresponding confidence level and the co-purchase influence coefficient of the corresponding e-commerce product and other types of e-commerce products using the following formula: Wherein, N represents the total number of categories of e-commerce products, i represents a positive integer less than or equal to N, and c i represents the confidence of the i-th category of e-commerce products, represents the confidence correction value of the i-th category of e-commerce products, i' represents a positive integer less than or equal to N - 1, and c i,i′ represents the confidence of other i'-th category of e-commerce products relative to the i-th category of e-commerce products, and η i,i′ represents the co-purchase influence coefficient between the i-th category of e-commerce products and the other i'-th category of e-commerce products; Determine whether the sum of the confidence level correction values of each type of e-commerce product exceeds the first preset threshold.

3. The e-commerce product precise recommendation method according to claim 2, wherein For each type of e-commerce product, based on the purchase event indication values of the corresponding e-commerce product during each historical startup operation and the purchase event indication values of other types of e-commerce products during each historical startup operation, calculate the co-purchase influence coefficients of the corresponding e-commerce product and other types of e-commerce products respectively, including: For a certain type of e-commerce product, extract the purchase event indication values of the corresponding e-commerce product during the current most recent consecutive M historical startup operations to form a first sample S1, where M represents a positive integer greater than m; For another certain type of e-commerce product relative to the certain type of e-commerce product, extract the purchase event indication values of the corresponding e-commerce product during the current most recent consecutive M historical startup operations to form a second sample S2; Perform a normal distribution KS test on the first sample S1 to calculate a first test statistic p value p1, and also perform the normal distribution KS test on the second sample S2 to calculate a second test statistic p value p2; If both the p-value p1 of the first test statistic and the p-value p2 of the second test statistic are greater than a second preset threshold, the correlation coefficient r between the first sample S1 and the sample S2 is calculated according to the following formula q : where i represents a positive integer, x i represents the i-th sample value in the first sample S1, y i represents the i-th sample value in the second sample S2, represents the sample mean of the first sample S1, represents the sample mean of the second sample S2; Take the correlation coefficient r q as the co-purchase influence coefficient of a certain type of e-commerce product and another certain type of e-commerce product.

4. The accurate recommendation method for e-commerce products according to claim 1, wherein After determining whether the sum of the confidence levels of each type of e-commerce product exceeds the first preset threshold, if it is determined that the sum of the confidence levels of each type of e-commerce product does not exceed the first preset threshold, the method further includes the following steps S701 to S704: S701. Adjust the current e-commerce product recommendation plan to obtain a new e-commerce product recommendation plan, and then execute step S702; S702. For each type of e-commerce product, import the browsing event indication values or browsing time ratio values of all e-commerce products during the current most recent consecutive m - 1 historical startup operations and the new e-commerce product recommendation plan into the corresponding purchase event prediction model, and output the new confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup operation, and then execute step S703; S703. Determine whether the sum of the new confidence levels of each type of e-commerce product exceeds the first preset threshold. If so, execute step S704; otherwise, execute step S701; S704. During the next startup operation, execute the new e-commerce product recommendation plan through the e-commerce platform application.

5. The e-commerce product precise recommendation method according to claim 1, characterized in that When there are multiple e-commerce product recommendation plans, the method further includes: For each of the various e-commerce products and each e-commerce product recommendation plan among multiple e-commerce product recommendation plans, input the browsing event indication value or browsing time ratio of all the e-commerce products during the current most recent consecutive m-1 historical startup runs and the corresponding plan into the purchase event prediction model of the corresponding e-commerce product, and output the event that the corresponding e-commerce product is purchased by the target customer during the next startup run and the corresponding confidence level. For each e-commerce product recommendation plan, accumulate and calculate the various e-commerce products and the corresponding confidence levels to obtain the corresponding total confidence level value. Determine whether the maximum total confidence level value exceeds the first preset threshold. If so, during the next startup run, execute, through the e-commerce platform application, an e-commerce product recommendation plan among multiple e-commerce product recommendation plans that corresponds to the maximum total confidence level value.

6. The precise recommendation method for e-commerce products according to claim 1, wherein The method further includes: Apply the browsing event indication value or browsing time ratio of all the e-commerce products during the current most recent consecutive m-1 historical startup runs and the purchase event prediction model of the various e-commerce products, and optimize the e-commerce product recommendation plan based on an optimization algorithm to obtain the e-commerce product recommendation plan and the optimal search result for maximizing the total confidence level of the various e-commerce products. During the next startup run, execute, through the e-commerce platform application, the e-commerce product recommendation plan obtained based on the optimal search result.

7. An e-commerce product precise recommendation device, characterized in that, It includes a record data collection unit, a record data processing unit, a sample data extraction unit, a prediction model training unit, a prediction model application unit, a threshold comparison and judgment unit, and a recommendation plan execution unit that are sequentially communicatively connected. The record data collection unit is used to collect the e-commerce product browsing record data and e-commerce product purchase record data generated by the target customer during each historical startup run of the e-commerce platform application. The record data processing unit is used to process the e-commerce product browsing record data to obtain the browsing event indication value or browsing time ratio of various e-commerce products during each historical startup run, and process the e-commerce product purchase record data to obtain the purchase event indication value of the various e-commerce products during each historical startup run. Among them, the browsing event indication value is used to indicate whether the event that the corresponding e-commerce product is browsed by the target customer occurs during the startup run of the e-commerce platform application, the browsing time ratio is the ratio of the duration of the target customer browsing the corresponding e-commerce product to the total duration of browsing all e-commerce products during the startup run of the e-commerce platform application, and the purchase event indication value is used to indicate whether the event that the corresponding e-commerce product is purchased by the target customer occurs during the startup run of the e-commerce platform application. The sample data extraction unit is configured to extract corresponding sample data for each type of e-commerce product in the following manner: taking the browsing event indication value or the browsing time ratio during consecutive m historical startup operations of all e-commerce products as model input items, and taking the purchase event indication value during the last historical startup operation among the consecutive m historical startup operations of the corresponding e-commerce product as the model output item, and then taking the model input items and the model output item as a sample data, where m represents a positive integer greater than or equal to 2; The prediction model training unit is configured to calibrate and verify the modeling of a machine learning model based on an artificial intelligence algorithm for each type of e-commerce product by applying all the corresponding sample data, and obtain a corresponding purchase event prediction model; The prediction model application unit is configured to, for each type of e-commerce product, import the browsing event indication value or the browsing time ratio during the current most recent consecutive m - 1 historical startup operations of all e-commerce products and the e-commerce product recommendation scheme during the next startup operation of the e-commerce platform application into the corresponding purchase event prediction model, and output the confidence level of the event that the corresponding e-commerce product is purchased by the target customer during the next startup operation, where the e-commerce product recommendation scheme includes the recommendation event indication value or the recommendation frequency ratio of each type of e-commerce product during the next startup operation, the recommendation event indication value is used to indicate whether an event of recommending the corresponding e-commerce product to the target customer for browsing is triggered during the next startup operation, and the recommendation frequency ratio refers to the ratio of the frequency of recommending the corresponding e-commerce product to the target customer for browsing to the total frequency of recommending all e-commerce products to the target customer for browsing during the next startup operation; The threshold comparison and judgment unit is configured to judge whether the sum of the confidence levels of each type of e-commerce product exceeds a first preset threshold; The recommendation scheme execution unit is configured to, when it is determined that the sum of the confidence levels of each type of e-commerce product exceeds the first preset threshold, execute the e-commerce product recommendation scheme through the e-commerce platform application during the next startup operation; 8. A computer device, characterized in that, It includes a memory, a processor, and a transceiver that are communicatively connected in sequence, where the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the e-commerce product precise recommendation method according to any one of claims 1 to 6; 9. A computer-readable storage medium, characterized in that , An instruction is stored on the computer-readable storage medium, and when the instruction runs on a computer, it executes the e-commerce product precise recommendation method according to any one of claims 1 to 6; 10. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or the instruction, when executed by a computer, implements the e-commerce product precise recommendation method according to any one of claims 1 to 6.

Images