Conversation-Based Recommender Using Attention Weights on Directed Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current recommendation systems using graph neural networks struggle to accurately learn useful information from directed graphs representing conversations, due to the inefficiency of averaging feature vectors from adjacent nodes, leading to inaccurate recommendation results.
Innovation Solution
A conversation-based recommending method that involves obtaining a directed graph corresponding to a current conversation, determining attention weights for directed edges based on node feature vectors and relationship data, calculating new feature vectors for nodes using these weights, and then determining a feature vector for the current conversation to recommend items accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If graph neural networks use average pooling to aggregate feature vectors from adjacent nodes, then the computation is simple, but the recommendation accuracy deteriorates due to inability to learn useful information efficiently
Solution Approach 1:
The patent applies local quality by assigning different attention weights to different edges connected to each node, allowing the model to selectively emphasize important relationships while suppressing less relevant ones. This is achieved through the attention mechanism that computes weight coefficients for each edge based on node features and edge attributes, enabling differentiated information aggregation rather than uniform averaging.
Solution Approach 2:
The patent changes the aggregation parameter from fixed uniform weights (average pooling) to dynamic attention weights that are learned during training. The attention weights are computed as functions of node features and edge attributes, allowing the model to adaptively adjust the importance of different neighbors for each node, thereby improving information aggregation quality.
2Productivity
If the model processes all edges in the directed graph uniformly, then the processing is straightforward, but the learning efficiency of useful information deteriorates
Solution Approach 1:
The patent introduces dynamics by making the edge weights adaptive rather than static. The attention mechanism dynamically computes weight coefficients for each edge based on the current node's features and the edge's attributes, allowing the model to flexibly adjust the importance of different relationships during the forward propagation process, thereby improving learning efficiency.
Solution Approach 2:
The patent applies partial action by selectively attending to only the most relevant neighbors for each node rather than uniformly processing all neighbors. The attention mechanism identifies and emphasizes important edges while down-weighting less relevant ones, enabling the model to focus computational resources on the most informative relationships in the graph.
Data Source
AI summary
The disclosure discloses a conversation-based recommending method. A directed graph corresponding to a current conversation is obtained. The current conversation includes clicked items, the directed graph includes nodes and directed edges between the nodes, each node corresponds to a clicked item, and each directed edge indicates relationship data between the nodes. For each node of the directed graph, an attention weight is determined for each directed edge corresponding to the node based on a feature vector of the node and the relationship data for each node of the directed graph. A new feature vector of the node is determined based on the relationship data and the attention weight of each directed edge. A feature vector of the current conversation is determined based on the new feature vector of each node. An item is recommended based on the feature vector of the current conversation.