Analyzing the Sentence Level Sentiment of Bengali People Through Facebook Comments

Abstract

Of late, with immense interest in research field a bunch of studies are being conducted on sentiment analysis mostly on English. There remain some studies in Bangla sentiment analysis focusing on computational analysis without implementing any guideline. So, this study aims firstly to propose a comprehensive linguistic guideline for sentence level sentiment analysis. Secondly, annotating Facebook comments by defining the Subjective, Objective and polarities—Strongly Positive, Weakly Positive, Strongly Negative, Weakly Negative, Neutral of Bangladeshi countrymen following proposed guideline and lastly a comparative analysis of five class polarities based on three different timelines. Data were collected through Graph API from public pages and profiles and after pre-processing a total of 13,852 sentences were selected for further analysis. Total of 50 university students wherein 40 as annotators and 10 as validators allocating in 10 groups whereases each group comprised of 4 annotators and 1 validator served to annotate sentiment. These 13,852 sentences were annotated by Group-3 of 10 groups. Kappa value of >0.80 was set for inter-annotator agreement. Result shows, the highest percentage of 27.76% Strongly Negative sentences which represents the Negativity of our society. Further studies should be done finding whys and wherefores rectifying the situation by implementing appropriate solution.