1. Review Mining for Music Digital Libraries: Phase II J. Stephen Downie, Xiao Hu The International Music Information Retrieval Systems Evaluation Lab (IMIRSEL) University of Illinois at Urbana-Champaign THE ANDREW W. MELLON FOUNDATION

2. Background & Motivation Classify Reviews Identify User Descriptions Connect to Objects Customer Reviews Epinions.com Positive Negative Description 1 D1D2D3D1D2D3 Phase IIPhase I Future

3. Review mining: phase I

4. Phase II Mining frequent descriptive patterns in positive and negative reviews ReviewsPositiveNegative Total Reviews400 Total Sentences63118 30053 Total Words1027713 447603 Avg. (STD ) sentences per review157.80 (75.49)75.13 (41.62) Avg. (STD) words per sentence16.28 (14.43)14.89 (12.24) sets of words used by users to express feelings/opinions

5. Frequent Descriptive Pattern Mining (FDPM) z Finds patterns consisting of items that frequently occur together in individual transactions  Items = candidate descriptive words (terms) = adjectives, adverbs and verbs, no nouns zTransactions = review sentences Items Transactions

6. Data/Text-to-Knowledge (D2K/T2K) Toolkits POS taggingFrequent pattern mining

7. Findings Digging deeper and deeper to find out what makes good things good and bad things bad….

8. Single term patterns Positive ReviewsNegative Reviews not – 3417 sentences good – 1621 sentences: 1/4 of all sentences not – 1915 sentences good – 1025 sentences: 1/3 of all sentences Good = Bad?!

9. good in a negative context Negation: “Nothing is good.” “It just doesn't sound good.” Song titles: “Good Charlotte, you make me so mad.” “Feels So Good is dated and reprehensibly bad.” Rhetoric: “And this is a good ruiner: …” “What a waste of my good two dollars…” Faint praise: “…the only good thing… is the packaging.” Expressions: “You all have heard … the good old cliché.”

10. Double term patterns Positive ReviewsNegative Reviews not good not realli realli good not listen not great not good not bad not realli not sound realli good Good  Bad?!

11. Triple term patterns Positive ReviewsNegative Reviews sing open melod sing smooth melod sing fill melod sing smooth open not realli good sing lead melod sound realli good sing plai melod accompani sing melod sing soft melod not realli good not realli listen bad not good bad not sound pretti tight spit bad not don’t realli not don’t realli bad not pretti bad not not sing sound

12. Comparison to an earlier study zCunningham et al. "The Pain, The Pain": Modeling music information behavior and the songs we hate. In Proc. of ISMIR ’05 What is the worst song ever?

13. Comparison to an earlier study This StudyCunningham et al ‘05 badreally annoyingbad hateworst reallyannoying inaneboring horrible stupidawful worsthate awfulstupid crap boreinane

14. Conclusions zTriple-term patterns necessary: Need to dig deeper to capture users’ emotional orientation/feelings toward music objects zFindings consistent with earlier work zCustomer reviews are an excellent resource for studying the underlying intentions and contributing features of user-generated metadata

15. Future work zNon-music cases Criticism mining on book and movie reviews zOther facets of music reviews Recommended usage metadata zOther feature studies Stylistics in customer reviews Naïve Bayesian feature ranking Noun pattern mining in different genres

16. Questions? Thank you! THE ANDREW W. MELLON FOUNDATION

17. References Han, J., Pei, J., and Yin, Y. Mining frequent patterns without candidate generation. In Proceedings of the ACM SIGMOD 2000. 1-12. Hu, X., Downie J.S., West K., and Ehmann A. Mining Music Reviews: Promising Preliminary Results. In Proceedings of the 6th International Symposium on Music Information Retrieval. 2005, 536-539. Welge, M., et al. Data to Knowledge (D2K) An Automated Learning Group Report. NCSA, University of Illinois at Urbana-Champaign, 2003. (http://alg.ncsa.uiuc.edu)http://alg.ncsa.uiuc.edu