Abstract—Sentiment analysis is a more difficult task than topic classification, spam etc. The challenges include how to describe the topic of a given review and how to find the topic-specific sentiment features. This paper applies sentiment analysis to psychological questionnaires. A role-based sentiment analyzer is proposed to detect sentiment features in thousands of questionnaires coming from sandplay therapy and to determine the polarity of them automatically, which help guiding and evaluating the therapeutic process. We decompose the method into 3 steps: detecting topic terms in each review according metadata words, labeling semantic roles on all topic sentences and identifying sentiment features related to topic terms through the labeled roles. A key of our method is that it can find topic-specific sentiment words depending on semantic orientation of roles. Addressing the shortage of manually tagged questionnaires, one semi-supervised learning model -- Spectral Graph Transducer (SGT) is applied to sentiment analysis based on a new feature representation on reviews, i.e. Positive-Related Vector Percentage. The experiments on sandplay questionnaires showed our sentiment analyzer performed well especially on long reviews, and when using SGT it improved rapidly in a small amount of tagged training data.

Index Terms—Sentiment analysis, sandplay questionnaires, topic-specific sentiment features, topic sentences, spectral graph transducer, semantic role labelling.

Chen Yaodong is with Department of Information and Engineering, Changsha Normal University, Changsha, China. Peng Diefei is with Department of Research & Discipline Development, Changsha Normal University, Changsha, China. Liu Qin is with Department of Vocational & Adult Education Research, Hunan Provincial Research Institute of Education, Changsha, China.

Cite:Chen Yaodong, Peng Diefei, Liu Qin, "Sentiment Analysis for Psychological Questionnaires with Semi-supervised Learning," International Journal of Computer and Electrical Engineering vol. 10, no. 3, pp. 221-232, 2018.