120 likes | 391 Views
Riding the tide of sentiment change: sentiment analysis with evolving online reviews. Yang Liu, Xiaohui Yu, Aijun An, Xiangji Huang. General Idea. Perform sentiment analysis on IMDB movie review Using the sentiment to predict box office revenues. Sentiment Analysis: PLSA.
E N D
Riding the tide of sentiment change: sentiment analysis with evolving online reviews Yang Liu, Xiaohui Yu, Aijun An, Xiangji Huang
General Idea • Perform sentiment analysis on IMDB movie review • Using the sentiment to predict box office revenues
Sentiment Analysis: PLSA • PLSA can be used as an unsupervised classifier to determine the mixture of sentiment in a review • Incapacity of adapting itself as new data become available • Brutal way is to train the whole model again • This paper proposed two ways to deal with new data
Data structure • Only appraisals in reviews are recognized and frequency of appraisals are counted in each reviews • Appraisal lexicon • Polarity means whether there is negation like “not”, “no”, “non’, etc
Adaptive to new data • Light-weight incremental model • For data do not change dramatically throughout the time
Adaptive to new data • Quasi-Bayesian model • X is the available data till time n • Assumed probability distribution on each time interval determined by some parameters • Dirichlet distribution is used φ = {α, β} are the hyperparameters of the Dirichlet distribution
Application to sales prediction • Linear regression model
Evaluation • 28,353 reviews for 20 drama films released in the US from 1 May 2006 to 1 September 2006 • Box office revenue of the corresponding drama films from 1 May 2006 to 1 September 2006 • 4 time intervals for the training of PLSA • Mean absolute percentage error is used to measure the prediction accuracy
Evaluation • Time interval set to be days = 2,4,6,8
Evaluation • Time interval set to be day = 5, 10, 15, 20