Generating Fine-Grained Reviews of Songs from Album Reviews

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1376–1385, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics Generating Fine-Grained Reviews of Songs From Album Reviews Swati Tata and Barbara Di Eugenio Computer Science Department University of Illinois, Chicago, IL, USA {stata2 | bdieugen}@uic.edu Abstract Music Recommendation Systems often recommend individual songs, as opposed to entire albums. The challenge is to gen- erate reviews for each song, since only full album reviews are available on-line. We developed a summarizer that combines in- formation extraction and generation tech- niques to produce summaries of reviews of individual songs. We present an intrinsic evaluation of the extraction components, and of the informativeness of the sum- maries; and a user study of the impact of the song review summaries on users’ de- cision making processes. Users were able to make quicker and more informed deci- sions when presented with the summary as compared to the full album review. 1 Introduction In recent years, the personal music collection of many individuals has significantly grown due to the availability of portable devices like MP3 play- ers and of internet services. Music listeners are now looking for techniques to help them man- age their music collections and explore songs they may not even know they have (Clema, 2006). Currently, most of those electronic devices follow a Universal Plug and Play (UPNP) protocol (UPN, 2008), and can be used in a simple network, on which the songs listened to can be monitored. Our interest is in developing a Music Recommendation System (Music RS) for such a network. Commercial web-sites such as Amazon (www. amazon.com) and Barnes and Nobles (www. bnn.com) have deployed Product Recommen- dation Systems (Product RS) to help customers choose from large catalogues of products. Most Product RSs include reviews from customers who bought or tried the product. As the number of reviews available for each individual product in- creases, RSs may overwhelm the user if they make all those reviews available. Additionally, in some reviews only few sentences actually describe the recommended product, hence, the interest in opin- ion mining and in summarizing those reviews. A Music RS could be developed along the lines of Product RSs. However, Music RSs recom- mend individual tracks, not full albums, e.g. see www.itunes.com. Summarizing reviews be- comes more complex: available data consists of album reviews, not individual song reviews (www. amazon.com, www.epinions.com). Com- ments about a given song are fragmented all over an album review. Though some web-sites like www.last.fm allow users to comment on indi- vidual songs, the comments are too short (a few words such as “awesome song”) to be counted as a full review. In this paper, after presenting related work and contrasting it to our goals in Section 2, we discuss our prototype Music RS in Section 3. We devote Section 4 to our summarizer, that extracts com- ments on individual tracks from album reviews and produces a summary of those comments for each individual track recommended to …

Content truncated to 500 words