Second session on Recommender Systems

The second session on Recommender Systems starts with the first paper on Automatic Construction of Travel Itineraries using Social Breadcrumbs by Munmun De Choudhury, Moran Feldman, Sihem Amer-Yahia, Nadav Golbandi, Ronny Lempel, and Cong Yu. They create these travel itineraries by taking photo streams from the Flickr dataset that map the photo with the city, then extract candidate POIs using the Yahoo! Maps API, then segment the photo streams to created timed paths to form the travel itineraries. They create the POI graph from this. They compare their automated travel itineraries with real itineraries using Amazon Mechanical Turk. The extensive survey-based user studies on Amazon Mechanical Turk gave promising results against bus tour companies' itineraries.

The second paper is on Speak the Same Language with Your Friends: Augmenting Tag Recommenders with Social Relations by Kaipeng Liu and Binxing Fang. Their proposed graph is to add edges between users and resources in addition to users linked to tags and tags linked to resources. They use random walk-based similarity measures. They compare personalized-CF with User-CF and showed that the proposed Personalized-CF algorithm with MFA as similarity measure performs best.

The third paper is Connecting Users and Items with Weighted Tags for Personalized Item Recommendations by Huizhi Liang, Yue Xu, Yuefeng Li, and Richi Nayak. The problem is that we have a tag quality problem, there is semantic ambiguity, tags could be personal and there are synonyms of tags that mean the same thing. Their proposed apprach is to use the multiple relationships of tagging. Usually only 2 dimensional relationships are used (user-item, user-tag, and item-tag), but in actuality there are three dimensional relationships of user-tag-item relationship: (userXtag)-item mapping, item-(userXtag) mapping. But what about user-(itemXtag) mapping?