Information recommendation method, system, equipment and medium
The user-item interaction graph is denoised through the PTDNet denoising model, and the noise edges are eliminated, and the graph data is optimized. The problem of graph neural network being affected by noise interaction in the recommendation algorithm is solved, achieving a more efficient and personalized recommendation effect.
Patent Information
- Application Number
- CN202411879019.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-04-08
AI Technical Summary
In the prior art, graph neural networks are used in recommendation algorithms to capture deep and complex relationships in user-item interaction graphs, and the recommendation effect is affected by noise interaction, resulting in poor recommendation results.
The PTDNet denoising model is used to denoise the user-item real-time interactive graph, calculate the importance weight of each edge, filter and eliminate the noise edge, dynamically update the edge weight, repeat the denoising process, obtain the denoising graph data, and input the information to recommend the model for personalized recommendation.
Through denoising processing, the graph data is optimized, the recommendation accuracy is improved, the user's personalized recommendation experience is improved, and the recommended performance indicators such as Recall@K and NDCG are significantly improved.
Smart Images
Figure CN119782616A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of recommendation systems, and in particular relates to an information recommendation method, system, device and medium. Background Art
[0002] The information recommendation method is a method that predicts the items that users are interested in based on the interaction information between users and items, thereby making personalized recommendations.
[0003] In the prior art, graph neural networks (GNNs) are widely used in recommendation algorithms because they can capture deep and complex relationships in user-item interaction graphs. GNNs improve recommendation results by integrating multi-hop neighbor relationships into the representation of users or items. However, there are usually noise interactions in actual interaction graphs, which can reduce the accuracy of graph collaborative filtering recommendation algorithms and lead to poor recommendation results. Summary of the invention
[0004] In order to overcome the shortcomings of the prior art in that information recommendation has a poor effect, the present invention provides an information recommendation method, comprising the following steps:
[0005] Obtain the user-item real-time interaction graph, input the user-item real-time interaction graph into the PTDNet denoising model, calculate the importance weight of each edge in the user-item real-time interaction graph, filter and retain important edges according to the set threshold and the importance weight of each edge, and remove the noise edges; dynamically update the importance weight of the edge, and repeat the process of removing the noise edges to obtain the denoised graph data;
[0006] The denoised graph data is input into the information recommendation model to identify user preferences and generate personalized recommendation results for users.
[0007] The process of obtaining the information recommendation model includes the following steps:
[0008] Obtain the interaction dataset between users and items, and construct the original user-item interaction graph based on the data in the interaction dataset;
[0009] The original user-item interaction graph is input into the PTDNet denoising model, and the PTDNet denoising model is used to denoise the original user-item interaction graph to obtain a denoised sparse graph dataset;
[0010] The data in the denoised sparse graph dataset is input into the graph neural network model, and the graph neural network model is trained. Specifically, the sparse graph dataset is divided into a training set and a validation set, and the data in the training set is input into the graph neural network model. The embedding vectors of user nodes and items are calculated through the graph convolution operation of the graph neural network model; a multi-layer propagation mechanism is adopted to fuse the embedding vectors of users and items with the information of their neighbor nodes at each layer, gradually capture deep high-order interaction features, capture the potential relationship between users and items, use the validation set to adjust the model parameters, and dynamically adjust the network parameters by minimizing the loss function to obtain the information recommendation model.
[0011] Preferably, the user-item original interaction graph is constructed based on the data in the interaction data set, specifically: the interaction data set includes users, items and users' evaluation records of merchants, users and items are respectively used as nodes, users' evaluation records of items are used as edges, edges and nodes are connected to obtain the user-item original interaction graph.
[0012] Preferably, the method of inputting the data in the denoised sparse graph dataset into the graph neural network model, and before training the graph neural network model, also includes converting the data in the denoised sparse graph dataset into an adjacency matrix and a node feature matrix, wherein the adjacency matrix is the adjacency matrix of the sparse user-item graph in the denoised sparse graph dataset, which describes the edge connection relationship between the user nodes and the item nodes retained after denoising.
[0013] Preferably, the graph neural network model shown is a LightGCN model.
[0014] Preferably, the interaction dataset is the Yelp2018 dataset.
[0015] The present invention also provides an information recommendation system, comprising:
[0016] The denoising module is used to obtain the user-item real-time interaction graph, input the user-item real-time interaction graph into the PTDNet denoising model, calculate the importance weight of each edge in the user-item real-time interaction graph, filter and retain important edges according to the set threshold and the importance weight of each edge, and remove the noise edges; dynamically update the importance weight of the edge, and repeat the process of removing the noise edges to obtain the denoised graph data;
[0017] The recommendation module is used to input the denoised graph data into the information recommendation model, identify the user's preferences, and generate personalized recommendation results for the user.
[0018] The present invention also provides a computer device, comprising a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the information recommendation method.
[0019] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the information recommendation method.
[0020] The information recommendation method, system, device and medium provided by the present invention have the following beneficial effects:
[0021] The present invention can calculate the importance weight of each edge in the user-item real-time interaction graph by using the PTDNet denoising model, screen and retain important edges according to the set threshold and the importance weight of each edge, and remove noise edges; by dynamically updating the importance weight of the edge and repeating the process of removing the noise edge, the process can denoise the graph data, optimize the interaction graph, and obtain denoised graph data; by inputting the denoised graph data into the information recommendation model, the information recommendation model can improve the information recommendation accuracy and enhance the user's personalized recommendation experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0023] Figure 1 is a flow chart of an information recommendation method according to an embodiment of the present invention;
[0024] Figure 2 A user-item interaction diagram provided by an embodiment of the present invention;
[0025] Figure 3 A schematic diagram of the training process of the LightGCN network model provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0026] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.
[0027] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the technical solutions of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.
[0028] In addition, the terms "first", "second", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance. In the description of the present invention, it should be noted that, unless otherwise clearly specified or limited, the terms "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In the description of the present invention, unless otherwise specified, "plurality" means two or more, which will not be described in detail here.
[0029] Example
[0030] The present invention provides an information recommendation method, specifically as follows Figure 1 As shown, the following steps are included:
[0031] Step 1: Obtain the interaction dataset between users and items, and construct the user-item original interaction graph based on the data in the interaction dataset.
[0032] The present invention uses a public user-item interaction dataset, such as the Yelp2018 dataset, which contains approximately 1,561,406 user-merchant interaction records, involving approximately 45,919 users and 45,538 merchants. The interaction dataset mainly includes user, merchant, and user-merchant interaction information. Each user has interaction records with at least 10 merchants, and each merchant is also evaluated by at least 10 users.
[0033] Before constructing the user-item original interaction graph, the data in the interaction dataset is first preprocessed to remove incomplete, invalid or duplicate interaction records to improve data quality. Next, the user-item interaction graph is constructed, with users and items as nodes, user evaluation records of items as edges, and edges and nodes connected to obtain the user-item original interaction graph, as shown in Figure 2 shown.
[0034] Step 2: Input the original user-item interaction graph into the PTDNet denoising model, and use the PTDNet denoising model to denoise the original user-item interaction graph to obtain the denoised sparse graph dataset.
[0035] After constructing the user-item interaction graph, the PTDNet denoising model is used to denoise the original user-item interaction graph to eliminate noise and redundant information and improve the quality of the graph. Specifically, the PTDNet denoising model is used to denoise, identify and remove noise and redundant information in the user-item graph, evaluate the importance of edges in the user-item graph, selectively retain useful edges, and remove those that have a negative impact on capturing user preferences to obtain a sparse graph dataset. To prevent overfitting, nuclear norm regularization is applied to the generated sparse graph to keep its low-rank property and improve the generalization ability of the model.
[0036] Step 3: Input the data in the denoised sparse graph dataset into the graph neural network model, train the graph neural network model, and obtain the information recommendation model; input the user-item real-time interaction graph into the PTDNet denoising model, calculate the importance weight of each edge in the user-item real-time interaction graph, and select and retain important edges according to the set threshold and the importance weight of each edge, and remove the noise edges; dynamically update the importance weight of the edge, and repeat the process of removing the noise edges to obtain the denoised graph data.
[0037] After denoising is completed, the data in the denoised sparse graph dataset is first converted into an adjacency matrix and a node feature matrix, where the adjacency matrix is the adjacency matrix of the sparse user-item graph in the denoised sparse graph dataset, which describes the edge connection relationship between the user nodes and the item nodes retained after denoising; then the data in the obtained sparse graph dataset is divided into training set, validation set and test set in a ratio of 8:1:1. This division method ensures the generalization ability of the graph neural network model on different data sets. The training set and validation set are input into the lightweight graph neural network model, and the training set is used for model training (such as Figure 3As shown), specifically, the embedding vectors of user and item nodes are calculated through efficient graph convolution operations; a multi-layer propagation mechanism is used to fuse the embedding vectors of users and items with the information of their neighbor nodes at each layer to gradually capture deep high-order interaction features; during the training process, the validation set is used to adjust the model parameters, the optimal parameter configuration is found, and the model is evaluated using indicators such as Recall@K and NDCG, and finally a high-precision, low-computational burden recommendation system is obtained. At the same time, during the training process, the network parameters are dynamically adjusted by minimizing the loss function to ensure that the model can accurately capture user preferences and item characteristics. The test set is used to finally evaluate the model performance and finally obtain the information recommendation model. The present invention selects a suitable lightweight graph neural network model, the LightGCN recommendation model, which can effectively learn the relationship between users and items and reduce computational complexity through a simple graph propagation mechanism.
[0038] The real-time user-item interaction graph is input into the information recommendation model. The model calculates the items that are most likely to be of interest to the user based on the user's historical behavior and the denoised graph structure, and sorts them to generate personalized recommendation results for the user.
[0039] In addition, in order to verify the performance of the model, the recommendation results of the information recommendation model of the present invention are compared and analyzed with the traditional recommendation method, and the improvement in the Recall@K and NDCG recommendation performance indicators is evaluated. The recommendation results of each user are recorded and compared with their actual behavior. The results are shown in Table 1.
[0040] Table 1 Comparison of recommended effects
[0041] Recommended Method Recall@20 NDCG@20 LightGCN 0.0639％ 0.0525％ PTDNet+LightGCN 0.0677％ 0.0552％
[0042] As can be seen from Table 1, compared with the LightGCN method alone, the PTDNet+LightGCN method (the method of the present invention) combined with PTDNet shows a certain improvement in the recommendation effect. In the Recal l@20 index, the result of PTDNet+LightGCN is 0.0677%, which is 5.95% higher than that of LightGCN; in the NDCG@20 index, the result of PTDNet+LightGCN is 0.0552%, which is 5.14% higher than that of LightGCN.
[0043] This shows that by introducing the graph denoising mechanism of PTDNet and removing the noise edges in the user-item interaction graph, the performance of the recommendation model can be effectively improved. The PTDNet+LightGCN method further strengthens the effective interaction signals between users and items, allowing the model to more accurately capture user preferences and generate better recommendation results.
[0044] According to the user-item interaction dataset, the present invention uses the PTDNet denoising model to denoise the data to obtain an optimized user-item interaction graph; the denoised graph data is divided into a training set, a validation set and a test set, and the training set and the validation set are input into a lightweight graph neural network for training and verification, the test set is input into the trained model for recommendation, and the recommendation accuracy and recall rate indicators for evaluating the model performance are obtained.
[0045] Among them, after using the denoised graph obtained by the PTDNet model to input the lightweight graph neural network for recommendation, the recommendation accuracy and recall rate indicators obtained are significantly higher than the recommendation results generated by the traditional method, showing the effectiveness of the present invention in improving the accuracy and efficiency of recommendations. This result shows that by adopting a method based on graph topology denoising, the performance of the graph collaborative filtering recommendation system can be enhanced, and the user's personalized recommendation experience can be improved.
[0046] The present invention also provides an information recommendation system, comprising:
[0047] The denoising module is used to obtain the user-item real-time interaction graph, input the user-item real-time interaction graph into the PTDNet denoising model, calculate the importance weight of each edge in the user-item real-time interaction graph, filter and retain important edges according to the set threshold and the importance weight of each edge, and remove the noise edges; dynamically update the importance weight of the edge, and repeat the process of removing the noise edges to obtain the denoised graph data;
[0048] The recommendation module is used to input the denoised graph data into the information recommendation model, identify the user's preferences, and generate personalized recommendation results for the user.
[0049] The present invention also provides a computer device, including a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the information recommendation method.
[0050] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is suitable for being loaded by a processor to execute the information recommendation method.
[0051] The embodiments described above are only preferred specific implementation modes of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of the technical solutions that can be obviously obtained by any technician familiar with the field within the technical scope disclosed in the present invention belong to the protection scope of the present invention.
Claims
1. An information recommendation method, characterized in that: The steps include: Obtain the user-item real-time interaction graph, input the user-item real-time interaction graph into the PTDNet denoising model, calculate the importance weight of each edge in the user-item real-time interaction graph, filter and retain important edges according to the set threshold and the importance weight of each edge, and remove the noise edges; dynamically update the importance weight of the edge, and repeat the process of removing the noise edges to obtain the denoised graph data; The denoised graph data is input into the information recommendation model to identify user preferences and generate personalized recommendation results for users.
2. The information recommendation method according to claim 1, characterized in that: The process of obtaining the information recommendation model includes the following steps: Obtain the interaction dataset between users and items, and construct the original user-item interaction graph based on the data in the interaction dataset; The original user-item interaction graph is input into the PTDNet denoising model, and the PTDNet denoising model is used to denoise the original user-item interaction graph to obtain a denoised sparse graph dataset; The data in the denoised sparse graph dataset is input into the graph neural network model, and the graph neural network model is trained. Specifically, the sparse graph dataset is divided into a training set and a validation set, and the data in the training set is input into the graph neural network model. The embedding vectors of user nodes and items are calculated through the graph convolution operation of the graph neural network model; a multi-layer propagation mechanism is adopted to fuse the embedding vectors of users and items with the information of their neighbor nodes at each layer, gradually capture deep high-order interaction features, capture the potential relationship between users and items, use the validation set to adjust the model parameters, and dynamically adjust the network parameters by minimizing the loss function to obtain the information recommendation model.
3. The information recommendation method according to claim 2, characterized in that: The method of constructing a user-item original interaction graph based on the data in the interaction data set is as follows: the interaction data set includes users, items, and user evaluation records of merchants, and the users and items are respectively used as nodes, and the user evaluation records of items are used as edges. The edges and nodes are connected to obtain the user-item original interaction graph.
4. The information recommendation method according to claim 2, characterized in that: The method of inputting the data in the denoised sparse graph dataset into the graph neural network model and, before training the graph neural network model, also includes converting the data in the denoised sparse graph dataset into an adjacency matrix and a node feature matrix, wherein the adjacency matrix is the adjacency matrix of the sparse user-item graph in the denoised sparse graph dataset, and describes the edge connection relationship between the user nodes and the item nodes retained after denoising.
5. The information recommendation method according to claim 2, characterized in that: The neural network model shown in the figure is the LightGCN model.
6. The information recommendation method according to claim 2, characterized in that: The interaction dataset is the Yelp2018 dataset.
7. An information recommendation system, characterized in that: include: The denoising module is used to obtain the user-item real-time interaction graph, input the user-item real-time interaction graph into the PTDNet denoising model, calculate the importance weight of each edge in the user-item real-time interaction graph, and select and retain important edges and remove noise edges according to the set threshold and the importance weight of each edge; Dynamically update the importance weights of edges and repeat the process of removing noise edges to obtain denoised graph data; The recommendation module is used to input the denoised graph data into the information recommendation model, identify the user's preferences, and generate personalized recommendation results for the user.
8. A computer device, characterized in that: It comprises a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the information recommendation method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the information recommendation method according to any one of claims 1 to 6.