Dynamically Structured Holographic Memory (DSHM) is a cognitive model of associative memory that can be applied to the problem of recommendation. DSHM uses holographically reduced representations to encode the associations between objects that it learns about to generate recommendations. We compare the recommendations from this holographic recommender to a user-based collaborative filtering algorithm on several dataset, including MovieLens, and two bibliographic datasets from a scientific digital library. Off-line experiments show that the DSHM recommender predicts movie ratings as well as collaborative filtering and much better than collaborative filtering on very sparse bibliographic data sets. DSHM also has a unified underlying model that makes multi-dimensional recommendations and their explanations easier to develop. However, DSHM requires significant amounts of computational resources to generate recommendations and it may require a distributed implementation for it to be practical as a recommender for large data sets.

Additional Metadata
Keywords Dynamically Structured Holographic Memory, cognitive model, DSHM recommender, associative memory, learning associations
Publisher Institute of Cognitive Science
Series Cognitive Science Technical Report Series
Citation
Rutledge-Taylor, Matthew, Vellino, Andre, & West, R. (2010). Dynamically Structured Holographic Memory for a Recommendation. Technical Report 2010-01. Cognitive Science Technical Report Series. Institute of Cognitive Science.