Position-Aware Session Recommendation With SR-GNN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of session recommendation in SR-GNN is low due to treating multiple session sequences as the same session sequence for analysis.
Innovation Solution
Generate a position information sequence based on the arrangement of items in a session control sequence and combine it with the item embedding vector matrix to create a target embedding vector matrix, distinguishing different session sequences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If multiple session sequences corresponding to one directed graph are regarded as the same session sequence for analysis, then the computational complexity is reduced, but the accuracy of the recommendation result deteriorates
Solution Approach 1:
The patent segments the session sequence representation by introducing position information that divides the embedding representation into distinct positional components. Each item's embedding is combined with its position information to create unique position-aware embeddings, allowing the system to distinguish between different session sequences while maintaining manageable computational complexity through structured segmentation of the representation space.
Solution Approach 2:
The patent adds a positional dimension to the existing item embedding space. By generating position information sequences and combining them with item embeddings through concatenation or addition, the system transforms the original embedding representation into an enhanced representation that includes positional characteristics, thereby enabling differentiation of multiple session sequences without exponentially increasing computational complexity.
2Measurement precision
If position information sequence is generated and combined with item embedding vector matrix, then the accuracy of recommendation is improved, but the computational complexity increases
Solution Approach 1:
The computational process is segmented into distinct stages: generating position information sequences, obtaining embeddings for position information, combining with item embeddings, and feeding into SR-GNN. This segmentation allows for optimized computation at each stage, where position information can be precomputed and cached, reducing the overall computational burden while maintaining accuracy improvements.
Solution Approach 2:
The patent modifies the embedding parameters by incorporating position information, transforming the original item embeddings into position-aware embeddings. This parameter change enables the model to capture sequential patterns more effectively, improving recommendation accuracy while the computational overhead is managed through efficient parameter sharing and precomputation strategies.
Data Source
AI summary
A session recommendation method, a device and an electronic device are provided, related to the field of graph neural network technology. The session recommendation method includes: acquiring a session control sequence, and acquiring a first embedding vector matrix based on an embedding vector of each of items in the session control sequence; generating a position information sequence based on an arrangement sequence of the items in the session control sequence, and acquiring a second embedding vector matrix based on an embedding vector of each piece of position information in the position information sequence; determining a target embedding vector matrix based on the first embedding vector matrix and the second embedding vector matrix; and determining a recommended item, based on the target embedding vector matrix and through a Session-based Recommendation Graph Neural Network.