Eureka! Institutional Repository

In the present thesis we tackled the problem of sentiment analysis on hotel reviews found online. Sentiment Analysis is the process of detecting the positive or negative orientation of the writer, in this case of a hotel review, towards the subject of the text excerpt, in this case hotel. We utilized both probabilistic machine learning algorithms like Naïve Bayes and Maximum Entropy, and linear classifiers like Support Vector Machines. The classifiers were investigated on several feature extracting methods. One method was to use a general purpose sentimental lexicon and aggregate the sentiment orientation to the review level. The other method was to detect hidden aspects of the words used in the review and thus detect the hidden aspects discussed in the review. A third method was the Bag-ofWords model, where each word becomes a feature for the classifier. Finally we investigated combining the feature extraction methods and that proved the most successful method.