Paper Reading #16: Creating Collections with Automatic Suggestions and ExampleBased Refinement (UIST 24)

Comments

http://shennessy11.blogspot.com/2011/03/paper-reading-15.html 

http://chi-garza.blogspot.com/2011/03/paper-reading-16-satellite-cursor.html

Reference Information

   Title:Creating Collections with Automatic Suggestions and ExampleBased Refinement

   Author:Adrian Secord, Holger Winnem¨oller, Wilmot Li, Mira Dontcheva

   Publisher: UIST' 10, October 3-6, 2010 New York

Summary

People love to create collections. One of the most created collections is that of music playlists from personal media libraries. Today, users do one of to things: they manually enter them one at a time or use an example-based recommendation system to automagically generate a collection. The first method is somewhat inconvenient but very precise. The second is much more convenient but usually limited in scope--the algorithm typically chooses an aspect of the song (date, genre, artist) and makes a playlist out of that.

The researchers in this paper want to go farther than that. They propose a semi-automatic interface that creates collections that combine automatic suggestions with manual refinement. They also add a suggestion widget, a mechanism for exploring alternatives for one or more collection items and a two-pane linked interface that helps users browse their libraries based on any selected collection item.

The keyword queries are made to be as flexible as possible. For  example, one could type "some rock, a lot of U2, no Alternative". Each phrase is a user criterion. More complex modifiers can be used such as "mostly," "some", "a lot" and "no". When a modifier is not specified, it is regarded as "some." More complex phases consist of "Michael Jackson before 1990." Should a phrase parse incorrectly, a message will prompt the user to fix the phrase.

Discussion
I found their work to be applicable to real life. I personally prefer to make my own playlists myself, that way I know exactly what is in them. There are several occasions when I use the "Genius" feature on iTunes and I feel like it works very well. I am surprised that the researchers didn't really mention iTunes' genius feature very much. I would've thought they would talk about it under related work. They don't mention until they ask people if they are aware of it and how well they think it works.