Commodity intelligent recommendation optimization method, medium and equipment
By extracting user and product feature data from e-commerce transaction logs, using heavy weighted balance strategy to filter data and train artificial intelligence big models, the problem of inaccurate push of traditional product intelligent recommendation data is solved, and more efficient and accurate product recommendations are achieved.
Patent Information
- Application Number
- CN202411884328.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-04-04
AI Technical Summary
Traditional intelligent product recommendations have the problem of inaccurate data push, which is difficult to effectively meet user needs and improve the accuracy and rationality of recommendations.
By extracting user feature data and product feature data from e-commerce shopping transaction log data, the data is filtered using a balance strategy based on heavy weighting, and the filtered data is used to train artificial intelligence models to recommend products.
It improves the training efficiency and prediction accuracy of large models, reduces invalid exposure data, and enhances the accuracy and rationality of product recommendations.
Smart Images
Figure CN119762188A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent large model recommendation, and specifically relates to a commodity intelligent recommendation optimization method, medium and equipment. Background Art
[0002] Intelligent product recommendation is to infer the current and future needs of users through their historical behavior traces. The user's interactive behaviors such as searching, liking, commenting, and forwarding all provide the data basis for intelligent recommendation. For example, when a user searches for a product on a shopping platform, he or she will soon find that the platform also recommends similar products of similar types and prices to the user, and even extends to other related products.
[0003] As a means of increasing Internet users, intelligent product recommendation has been widely used in major companies. Its value lies in personalization and individuality. By comparing the recommendation system with the search system, we can find that in the recommendation system, users generally do not have a clear purpose for searching, or the purpose of obtaining information is not strong. Then, personalized recommendation is very important. The main functions of intelligent product recommendation include: on the one hand, it can meet user needs, improve user activity and the stickiness between the platform and users; on the other hand, for the platform, it not only improves the user experience, but also solves the long-tail problem to a certain extent. However, traditional intelligent product recommendation often has the problem of inaccurate data push. In the Web2.0 era, the application of large artificial intelligence models can bring new improvements to the performance of intelligent product recommendation. Summary of the invention
[0004] In view of the deficiencies in the prior art, the present invention provides a commodity intelligent recommendation optimization method, medium and device.
[0005] To achieve the above object, the present invention adopts the following technical solutions:
[0006] In a first aspect, the present invention provides a method for optimizing intelligent product recommendation, comprising the following steps:
[0007] Step 1: Extract user feature data and corresponding product feature data from e-commerce shopping transaction log data to form user feature data subsets and product feature data subsets respectively, and combine these two subsets into a big data set;
[0008] Step 2: Filter the large data set using a balancing strategy based on reweighting;
[0009] Step 3: Use the filtered big data set as training data to train the artificial intelligence model;
[0010] Step 4: Based on the trained artificial intelligence model, make product recommendations according to the characteristic data of the user to be recommended.
[0011] Optionally, in step 1, the user characteristic data in the user characteristic data subset is divided into characteristic data of frequent purchasing users and characteristic data of infrequent purchasing users, wherein the frequent purchasing users and infrequent purchasing users are users whose purchasing frequencies are higher and lower than a set frequency value, respectively.
[0012] Optionally, in step 2, the large data set is filtered using a weighted balance strategy, including filtering the balance between the user feature data subset and the product feature data subset, and also filtering the balance between the feature data of frequent purchasing users and the feature data of infrequent purchasing users.
[0013] Optionally, in step 2, the large data set is filtered using a weighted balancing strategy, as follows:
[0014] Step 2.1: For a large data set with C categories, calculate the weight ω of each category by the following formula: i :
[0015]
[0016] In the formula, category i has N i samples, category j has N j samples, i, j∈[1,...,C], θ is the adjustment factor to adjust the difference between weights of different categories;
[0017] Step 2.2: According to the calculated weights of each category, calculate the weighted loss L corresponding to each category by the following formula: weighted :
[0018] L weighted =-w i log(softmax(x i ));
[0019] In the formula, x i is the logits of the sample in category i, softmax(x i ) is the predicted probability of the sample in category i;
[0020] Step 2.3: Adjust the number of samples in each category so that the weighted loss of each category is within the set range.
[0021] Optionally, in step 3, the filtered big data set is used as training data to train the artificial intelligence big model, specifically:
[0022] Step 3.1: Take the feature data of users with frequent purchase behavior as input and the corresponding product feature data as output to train the artificial intelligence model;
[0023] Step 3.2: Take the feature data of users with infrequent purchasing behavior as input and the corresponding product feature data as output to train the artificial intelligence model.
[0024] Optionally, in step 3, the artificial intelligence large model adopts deep neural networks DNNs, Transformer models or graph neural networks GNNs.
[0025] Optionally, in step 4, the product recommendation is performed based on the trained artificial intelligence model and the feature data of the user to be recommended, specifically:
[0026] Step 4.1: Input the feature data of the user to be recommended into the trained artificial intelligence model and output the feature data of the product;
[0027] Step 4.2: Select a list of products to be recommended based on the output product feature data;
[0028] Step 4.3: Recommend products to users based on the product list and record user feedback in real time;
[0029] Step 4.4: Based on user feedback, calculate the user preference data under actual transactions through the fairness strategy model;
[0030] Step 4.5: Match products according to user preference data and add them to the list of products to be recommended, and return to step 4.3 until the number of recommended products reaches the set threshold.
[0031] Optionally, in step 4.3, the fairness policy model is:
[0032] π t =A((x 1 , σ 1 , c 1 ),…,(x t , σ t , c t ));
[0033] In the formula, π t Indicates the score of the current product sorting strategy. The user preference in the actual transaction is determined according to the score. t represents the information of product t, σ t represents the ranking position of product t, c t Represents the user’s feedback on product t.
[0034] In a second aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the commodity intelligent recommendation optimization method as described in the first aspect.
[0035] In a third aspect, the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the commodity intelligent recommendation optimization method as described in the first aspect is implemented.
[0036] The beneficial effects of the present invention are as follows: the present invention adopts a balancing strategy based on weighting to filter data, filters out invalid data as much as possible under the premise of ensuring a certain amount of training data, and improves the training efficiency and prediction accuracy of the large model; and the present invention formulates an intelligent push strategy based on the artificial intelligence large model, in which user preferences are calculated based on a fair strategy model, and commodity recommendations that can be updated and adjusted in real time are made according to user preferences, avoiding the problems of biased user preference estimation and unfair exposure distribution, greatly reducing invalid exposure data, and improving the accuracy and rationality of commodity recommendations. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is an overall flow chart of a commodity intelligent recommendation optimization method of the present invention.
[0038] Figure 2 It is a flow chart of the present invention using a weighted balancing strategy to filter a large data set.
[0039] Figure 3 It is a flow chart of the present invention for recommending products based on the trained artificial intelligence big model. DETAILED DESCRIPTION
[0040] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.
[0041] In one embodiment, if Figure 1 As shown, for e-commerce networks, in order to improve the accuracy, rationality and network security of intelligent marketing, the present invention proposes a commodity intelligent recommendation optimization method, including the following steps:
[0042] Step 1: Extract user feature data and corresponding product feature data from the e-commerce shopping transaction log data to form user feature data subsets and product feature data subsets respectively, and combine these two subsets into a big data set.
[0043] In this embodiment, the user characteristic data in the user characteristic data subset is divided into characteristic data of frequent purchase behavior users and characteristic data of infrequent purchase behavior users. The division of frequent purchase behavior users and infrequent purchase behavior users can be determined by setting a frequency threshold.
[0044] Step 2: Use a balancing strategy based on reweighting to filter the large data set.
[0045] The reweighting strategy is a rebalancing strategy that increases the training weight of the tail categories by adjusting the weights of different categories in the loss function. The reweighting strategy assigns large loss weights to the tail categories and small loss weights to the head categories, so that the model can be adjusted to pay more attention to the tail categories during training. Generally, the category weights use the data distribution of the data set as a priori, and the inverse of the number of samples in each category is taken to calculate the corresponding weights. The reweighting strategy is a very simple and efficient rebalancing method. It only requires weight scaling on the corresponding items of each category in the loss function. It is generally combined with the conventional cross entropy loss function and is referred to as the reweighted loss function.
[0046] The big data set is filtered by using a weighted balance strategy, including filtering the balance between the user feature data subset and the product feature data subset, as well as filtering the balance between the feature data of frequent purchase users and the feature data of infrequent purchase users, and filtering out invalid data as much as possible under the premise of ensuring a certain amount of training data. The filtering method can directly use weight filtering or loss function filtering. This embodiment takes loss function filtering as an example, and the details are as follows:
[0047] Step 2.1: For a large data set with C categories, calculate the weight ω of each category by the following formula: i :
[0048]
[0049] In the formula, category i has N i samples, category j has N j samples, i, j∈[1,...,C], θ is the adjustment factor to adjust the difference between weights of different categories;
[0050] Step 2.2: According to the calculated weights of each category, calculate the weighted loss L corresponding to each category by the following formula: weighted :
[0051] L weighted =-w i log(softmax(x i ));
[0052] In the formula, xi is the logits of the sample in category i, softmax(x i ) is the predicted probability of the sample in category i;
[0053] Step 2.3: Adjust the number of samples in each category so that the weighted loss of each category is within the set range.
[0054] Step 3: Use the filtered big data set as training data to train the artificial intelligence model.
[0055] During the training process, the feature data of frequent purchase users and infrequent purchase users are used as input, and the corresponding product feature data are used as output to train the AI big model. Only in this way can the trained AI big model be applicable to different types of users. In order to finely divide the two types and improve the prediction effect of the model, two different prediction branches can be established in the AI big model for frequent purchase users and infrequent purchase users, and training and parameter adjustment can be performed separately. The AI big model can use deep neural networks DNNs, Transformer models or graph neural networks GNNs.
[0056] Step 4: Based on the trained artificial intelligence model, make product recommendations according to the characteristic data of the user to be recommended.
[0057] The fairness problem is a general term for the biased estimation of user preferences by the smart push module and the unfair distribution of exposure. The biased estimation of user preferences means that the estimated value cannot reflect the user's real preferences, but the user's real preferences are closely related to the actual transaction items. The resource model without actual transaction items will have inaccurate estimation.
[0058] Therefore, this embodiment further combines the fairness strategy model on the basis of the artificial intelligence model to calculate the user preference data of the actual transaction goods and filter out the invalid exposure data. The specific implementation steps are as follows:
[0059] Step 4.1: Input the feature data of the user to be recommended into the trained artificial intelligence model and output the feature data of the product;
[0060] Step 4.2: Select a list of products to be recommended based on the output product feature data;
[0061] Step 4.3: Recommend products to users based on the product list and record user feedback in real time;
[0062] The fairness strategy model is:
[0063] π t =A((x 1 , σ 1 , c1 ),…,(x t , σ t , c t ));
[0064] In the formula, π t Indicates the score of the current product sorting strategy. The user preference in the actual transaction is determined according to the score. t represents the information of product t, σ t represents the ranking position of product t, c t Represents the user’s feedback on product t.
[0065] Step 4.4: Based on user feedback, calculate the user preference data under actual transactions through the fairness strategy model;
[0066] Step 4.5: Match products according to user preference data and add them to the list of products to be recommended, and return to step 4.3 until the number of recommended products reaches the set threshold.
[0067] In addition, in the process of intelligent product recommendation, the infection mechanism of the SIS model can be used to predict network anomalies, thereby ensuring that the fluctuations in e-commerce shopping transaction data are not abnormal fluctuations caused by fake orders. The description of the SIS model is as follows:
[0068]
[0069] Assume that the density of individuals in the susceptible state and infected state in the network at any moment is s(t) and i(t) respectively. When the susceptible individuals and infected individuals are fully mixed, the dynamic behavior of the SIS model can be described as the following differential equations:
[0070]
[0071] In the formula, let the effective infection rate i(T) = 0. The equation has a threshold λ′. When λ<λ′, the steady-state solution i(T) = 0; and when λ>λ′, the steady-state solution i(T)>0, where T represents the time it takes to reach the steady state.
[0072] In another embodiment, the present invention proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the commodity intelligent recommendation optimization method of the aforementioned embodiment.
[0073] In another embodiment, the present invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the commodity intelligent recommendation optimization method of the aforementioned embodiment is implemented.
[0074] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.
[0075] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0076] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A commodity intelligent recommendation optimization method, characterized in that: The steps include: Step 1: Extract user feature data and corresponding product feature data from e-commerce shopping transaction log data to form user feature data subsets and product feature data subsets respectively, and combine these two subsets into a big data set; Step 2: Filter the large data set using a balancing strategy based on reweighting; Step 3: Use the filtered big data set as training data to train the artificial intelligence model; Step 4: Based on the trained artificial intelligence model, make product recommendations according to the characteristic data of the user to be recommended.
2. A commodity intelligent recommendation optimization method according to claim 1, characterized in that: In step 1, the user characteristic data in the user characteristic data subset is divided into characteristic data of frequent purchasing behavior users and characteristic data of infrequent purchasing behavior users, wherein the frequent purchasing behavior users and infrequent purchasing behavior users are users whose purchasing frequencies are higher and lower than a set frequency value, respectively.
3. A commodity intelligent recommendation optimization method according to claim 2, characterized in that: In step 2, the large data set is filtered using a weighted balance strategy, including filtering the balance between the user feature data subset and the product feature data subset, as well as filtering the balance between the feature data of frequent purchasing users and the feature data of infrequent purchasing users.
4. The method for optimizing intelligent product recommendation according to claim 1, characterized in that: In step 2, the large data set is filtered using a weighted balancing strategy, as follows: Step 2.1: For a large data set with C categories, calculate the weight ω of each category by the following formula: i : In the formula, category i has N i samples, category j has N j samples, i, j∈[1,...,C], θ is the adjustment factor to adjust the difference between weights of different categories; Step 2.2: According to the calculated weights of each category, calculate the weighted loss L corresponding to each category by the following formula: weighted : L weighted =-w i log(softmax(x i )); In the formula, x i is the logits of the sample in category i, softmax(x i ) is the predicted probability of the sample in category i; Step 2.3: Adjust the number of samples in each category so that the weighted loss of each category is within the set range.
5. The method for optimizing intelligent commodity recommendation according to claim 2, characterized in that: In step 3, the filtered big data set is used as training data to train the artificial intelligence big model, specifically: Step 3.1: Take the feature data of users with frequent purchase behavior as input and the corresponding product feature data as output to train the artificial intelligence model; Step 3.2: Take the feature data of users with infrequent purchasing behavior as input and the corresponding product feature data as output to train the artificial intelligence model.
6. A commodity intelligent recommendation optimization method according to claim 2, characterized in that: In step 3, the artificial intelligence large model adopts deep neural networks DNNs, Transformer models or graph neural networks GNNs.
7. A commodity intelligent recommendation optimization method according to claim 1, characterized in that: In step 4, the product recommendation is performed based on the trained artificial intelligence model and the characteristic data of the user to be recommended, specifically: Step 4.1: Input the feature data of the user to be recommended into the trained artificial intelligence model and output the feature data of the product; Step 4.2: Select a list of products to be recommended based on the output product feature data; Step 4.3: Recommend products to users based on the product list and record user feedback in real time; Step 4.4: Based on user feedback, calculate the user preference data under actual transactions through the fairness strategy model; Step 4.5: Match products according to user preference data and add them to the list of products to be recommended, and return to step 4.3 until the number of recommended products reaches the set threshold.
8. The method for optimizing intelligent commodity recommendation according to claim 1, characterized in that: In step 4.3, the fairness strategy model is: p t =A((x1,σ1,c1),…,(x t ,s t ,c t )); In the formula, π t Indicates the score of the current product sorting strategy. The user preference in the actual transaction is determined according to the score. t Represents the information of the product, σ t represents the ranking position of product t, c t Represents the user’s feedback on product t.
9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to execute the commodity intelligent recommendation optimization method as described in any one of claims 1 to 8.
10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the commodity intelligent recommendation optimization method as described in any one of claims 1 to 8 is implemented.