Social Attentional Collaborative Filtering for Recommendation Accuracy

The patent segments social relationships into different types (friend relationships and group relationships) and processes each type separately through dedicated attention mechanisms. The attention module calculates attention scores for different relationship types independently, then combines them to generate comprehensive recommendation scores. This segmentation allows the system to handle complex multi-type social relationships while maintaining manageable computational structure.

The patent introduces an attention mechanism as an intermediary component that processes and weights information from different social relationship types. The attention module acts as a mediator between raw social interaction data and final recommendation scores, dynamically determining the importance of different relationship types and group members. This intermediary structure enables accurate integration of diverse social signals without requiring complex manual feature engineering.

The patent implements partial action by selectively processing only the most relevant social relationships and group members based on attention scores. Instead of uniformly processing all social interactions, the system identifies and focuses computational resources on high-attention relationships that contribute most to recommendation accuracy. This selective processing reduces overall computational burden while maintaining recommendation quality.

The patent dynamically changes the weighting parameters of different social relationship types based on user behavior and context. The attention mechanism adjusts the importance weights of friend relationships versus group relationships, and the importance of individual group members, according to observed user interactions. This parameter adaptation allows the system to optimize computational resource allocation by emphasizing more influential relationships while reducing processing of less relevant ones.