Date and time: 19th April 2010, 1:00 PM
Location: 219 SEO
Committee: Prof. Barbara Di Eugenio, Prof. Peter Nelson, Prof. Bing Liu,
Dr. Susan McRoy (University Of Wisconsin, Madison), Dr. Swee Mok (Motorola
Research Lab, Schaumburg, IL)

Abstract:

With the drastic increase in the personal music collections of individuals and the increased availability of data on the internet, recommendation systems are becoming increasingly popular. Many users decide whether or not to try a product by looking at reviews. However, there are far too many reviews for each product and it is a tedious task to read all of them. The main focus of this research is to produce a coherent and grammatical summary of reviews of individual songs. As reviews of songs are rarely available, this information has to be extracted from album reviews, which encompass several songs from the same album.

A music recommendation system has been developed. A formative study was carried out to determine that users are interested to read reviews in the recommendation system. An automatic summarization framework was proposed and developed, that combines extraction of information about songs from album reviews and generation techniques to produce summaries of reviews of individual songs. Fine-grained song feature summaries were incorporated into a music recommendation system, SongRecommend. Evaluation of SongRecommend with 39 users showed that users were able to make quicker decisions when presented with the summary as compared to the full album review; additionally, their decisions appeared to be more informed as their choices of recommendations to follow were more varied than in the control condition.