Background: Collaborative recommendation systems make predictions by information
filtering on the ratings of users on items. For data of a large volume and arriving dynamically, the
traditional approach of collaborative filtering (CF) based on user-item rating matrix directly is not
only expensive in terms of storage space for maintaining the matrix and time for training the prediction
model, but also in terms of accuracy in predications for the items.
Objective: The study aimed to overcome these deficiencies by associating content-based information
(user’s tags and item’s keywords) through categories.
Method: We propose a novel CF method based on the Tag-Category-Keyword three-level transitive
rating model that uses item’s category as a bridge for computing similarity between user’s tags and
item’s keywords and recommends the items with high similarity in content to users.
Results: Experimental results on real dataset demonstrate the above improvements of our approach
over the traditional CF methods.
Conclusion: By associating the content information of both items and users, our method improves
both efficiency and accuracy.