MusicLand: Exploratory Browsing in Music Space Heidi Lam December 15, 2004 CPSC 533CInformation Visualization Project Presentation

Agenda Motivation: Exploratory browsing? The ideal infovis solution: what should it be? MusicLand: my proposed solution –Visualization features –Demo –Evaluation –Future Work

Project Idea How can computer tools/interfaces better support exploratory browsing? What is exploratory browsing?

Two Scenarios at an Online Store 1.Find/Search: Looking for Ray Charles’ “The Spirit of Christmas” with using the store’s search tools, and input terms: “Ray Charles” and “spirit of Christmas” 2.Exploratory Browsing: Browsing at the “Classical” section  Came across a Jazzified version of Bach  Go to the “Jazz” section  Ray Charles’ Christmas album is being advertised

Two Scenarios at an Online Store The goals and assumptions of these scenarios are different: With find/search: –with clear and precise targets –want to find the target as quickly as possible With exploratory browsing: –need multiple queries to refine/define target –getting there is half of the fun/work?

Project Motivation Exploratory browsing is not well supported by current search tools… To better support exploratory browsing, the interface should … 1.Provide context: to allow users to interpret the query results based on their input terms  what am I looking at? 2.Guide navigation: going from the familiar to the unfamiliar  Given where I have been, where should I go next? 3.Assist refinement of target: based on retrieved results and query terms  what can I look for next?

Related Work Focus on query-based systems 4 approaches: 1.Spatial: retrieved results are clusters into groups based on query terms, and displayed spatially 2.List: retrieved results are displayed as a linear list 3.Temporal: retrieved results in the context of timelines 4.Integrated: multi-view with combinations of the above approaches

Related Work: Spatial 1.Show relationship between keywords— InfoCrystal (1993)

Related Work: Spatial 1.Show relationship between keywords—InfoCrystal (1993) 2.Show clusters only—Lighthouse (2000)

Related Work: Spatial 1.Show relationship between keywords—InfoCrystal (1993) 2.Show clusters only—Lighthouse (2000) 3.3D terrain— BEAD (1993)

Related Work: Spatial 1.Show relationship between keywords—InfoCrystal (1993) 2.Show clusters only— Lighthouse (2000) 3.3D terrain—BEAD (1993) 4.Arrange multiple query results spatially —Sparkler (2001)

MusicLand—Overall View

MusicLand—Query Display to put results in context of query terms to relate neigbouring regions by a query term Colour-coding the single term regions with primary colours, and the intersected areas with a mix of those colours Perceptual Layering to indicate the relative importance of each result region Arranges query results as a rectangularized Venn diagram

MusicLand—Semantic Zooming To increase display capacity –Full (artist, title, genre, style) –Partial (artist, title) –Minimal (artist) –Nil (number of results) Zoom

New query is an “extension” of old, linked by line, colour, and position MusicLand—New Query Select record of interest brings up a list of selectable attributes select new terms Old unused query regions fade and desize: Time metaphor

Dataset and Implementation 8556 mp3 files extracted from 714 albums by 315 different artists, labeled with English terms (by Eric Brochu) ALBFever to Tell ARTYeah Yeah Yeahs RELApr 29, 2003 GENRock STYIndie Rock, Garage Punk TONCathartic, Exuberant, Boisterous, Passionate, Brittle PAT/cs/beta/SCRATCH/music/mp3library/Yeah Yeah Yeahs/Fever to Tell Code written in Java with swt library

Demo

MusicLand—Evaluation Based on a 2-participant formative evaluation Strengths: Interprets query results in context Provides of possible query term choices Guides towards future queries with history Fades and desizes unused query regions  Looks like MusicLand has largely achieved its design goals :o)

MusicLand—Evaluation Weaknesses (Participants’ List): No album details Asymmetry of query term display No indication of new query terms in new query No Previews/Closing of query displays Weaknesses (My list): Scalability Perceptual layering Colour fading

MusicLand—Lessons Learned The importance of prototyping Colour is difficult Providing context and details has a price in scalability

MusicLand—Future Work Improve usability based on participants’ comments Launch and play the music albums Conduct usability study to verify the design principles behind MusicLand, and to study exploratory browsing

Conclusion MusicLand is a tool designed to support exploratory browsing … –Displays retrieved results in a rectangularized Venn diagram to cluster results and to provide query terms as context –Guides refinement of queries by providing a selectable list of attributes based on user’s current album of interest –Anchors and guides new query formation by providing a history Formative study with encouraging results

Questions