Task to be done
1. For each system user, we concatenate all titles of the papers he/she has every read, then use the joined title as a input to my tag suggestion engine. Finally I will select top-N represented tags for each user according to our tag suggestion results.
2. For each paper, I can use tag suggestion engine to get the most popular tag set, then the tag set can represent a paper’s attribute.
3. Similarity calculation
a. user-item tag set overlap ratio
b. weighted sum approach
a. according to different recommendation size, calculate the precision and recall.