Improving sentiment analysis in an online cancer survivor community using dynamic sentiment lexicon

Online Health Communities is a major source for patients and their family members in the process of gathering information and seeking social support. The American Cancer Society Cancer Survivors Network has many users and presents a large number of users' interactions with regards to coping with cancer. Sentiment analysis is an important step in understanding participants' needs and concerns and the impact of users' responses on other members. We present an automated approach for sentiment analysis in an online cancer survivor community and compare it with a previous sentiment analysis approach. Both approaches are machine learning based and are tested on the same dataset. However, this work uses features derived from a dynamic sentiment lexicon, whereas the previous work uses a general sentiment lexicon to extract features. Tested on several classifiers, with only six features (versus thirteen), our results show 2.3% improvement on average, in terms of accuracy, and greater improvement in F-measure and AUC. An additional experiment was conducted that showed a positive impact of dimensionality reduction by extracting abstract features, instead of using term frequency (TF) vector space as attribute values.