MIS 373 Lecture Notes - Lecture 13: Predictive Power, Sarcasm, Capital Asset Pricing Model

• School: UT-Austin
• Department: Management Information Systems
• Course: MIS 373
• Professor: A Barua

Document Summary

Numeric data easier to deal with than unstructured. Represent documents by some attributes (e. g. , presence or absence of terms ) Then use existing data mining methods on attributes. Binary representation: presence of absence of terms. Words that appear frequently in most documents are . Term frequency-inverse document frequency (tf-idf) = frequency of a term in a. Multiple products & attributes mentioned in the same post. Can social mentions predict box office hits? http://www. youtube. com/watch?v=uhmn2qsgbsu. Predictive power of social mentions for movies: Top 10 box office hits of 2012 (hollywood) Change in mentions volume from pre -> post release: +70% Average negative sentiment: 9% (of total mentions) Pre- to post-release: negatives stayed constant at ~ 9% Top 10 box office flops of 2012. Change in volume from pre -> post release: +17% Pre- to post-release: negatives increased from 23% to 27% Incorporating sentiments & mentions in the returns model.