Can I Hear You? Sentiment Analysis on Medical Forums

International Joint Conference on Natural Language Processing, pages 667–673, Nagoya, Japan, 14-18 October 2013. Can I hear you? Sentiment Analysis on Medical Forums Tanveer Ali University of Ottawa tali028@uottawa.ca David Schramm University of Ottawa and CHEO dschramm@ottawahospital.on.ca Marina Sokolova University of Ottawa and CHEO sokolova@uottawa.ca Diana Inkpen University of Ottawa Diana.Inkpen@uottawa.ca Abstract Text mining studies have started to investigae rela- tions between positive and negative opinions and pa- tients’ physical health. Several studies linked the per- sonal lexicon with health and the health-related be- havior of the individual. However, few text mining studies were performed to analyze opinions expressed in a large volume of user-written Web content. Our current study focused on performing sentiment analy- sis on several medical forums dedicated to Hearing Loss (HL). We categorized messages posted on the forums as positive, negative and neutral. Our study had two stages: first, we applied manual annotation of the posts with two annotators and have 82.01% over- all agreement with kappa 0.65 and then we applied Machine Learning techniques to classify the posts. 1 Introduction Natural language statements can be divided into two categories: factual and emotional. Factual statements can be expressed with a few topic keywords, while emotional statements express sentiments of the statement’s author and require a more complex analysis than the factual ones. Sentiment Analysis is often regarded as classifying and identifying the subjective infor- mation in the natural language text. In its appli- cation, Sentiment Analysis aims to detect the sentiments (e.g., opinions and emotions) of the speaker of the statement. Sentiments are charac- terized by polarity, intensity, strength and imme- diacy. In the current study, we focus on the polari- ty of sentiments that are expressed in messages posted on medical forums. Polarity can be bina- ry (e.g., positive vs. negative) or multi- categorical (e.g., positive, negative and un- known). Below we list examples found in online discussions about hearing aids. Positive1 This has the beneficial effect of making the quieter sounds audible but not blowing your head off with the louder sounds. Neutral/Unknown Now, you'll hear some people saying that compression is bad, and linearity is good, espe- cially for music. Negative Someone with 50 DB hearing aid gain with a total loss of 70 DB may not know that the place is producing 107 DB since it may not appear too loud to him since he only perceives 47 DB. In this work, we classified the subjective sen- tences into positive, negative and neutral. We have identified different syntactic features, i.e., patterns / rules (Yi and Nasukawa, 2003) which can indicate subjectivity and polarity of the sen- tences. The dataset of 3515 sentences from 26 threads were manually annotated by two annota- tors having overall agreement of 82.01% and kappa 0.65 which indicates substantial agreed data. Our experiments with different combinations of features using different classifiers have shown significant improvement in performance over the baseline. For example, with the Naïve Bayes classifier, the F1-score was10.5%better. The rest of the paper is organized as …

Content truncated to 500 words