Journal of Digital Information Management Jen-Yuan Yeh, Shihn-Yuarn Chen 14 1 2016 http://dline.info/fpaper/jdim/v14i1/v14i1_8.pdf Social tags have been considered to indirectly reflect authorized opinions of taggers. This paper proposes an unsupervised method which derives implicit sentiment information from social tags to decide, in one document, which sentences are opinionated, as well as to annotate them with proper polarity labels. First, for a social tag, its opinion degree is measured by aggregating the opinion degree of related sentiment words, in proportion to the co-occurrence relations between sentiment words and the tag. Second, the opinion degree of a sentence is determined by a combination function of the opinion degree of the tags, in proportion to the similarity between the sentence and each tag. Finally, sentences are sorted in order of their opinion degree, followed by a partition of the ranked list to distinguish sentences into positively opinionated, negatively opinionated, neutral, and non-opinionated ones. The proposed method is examined using the Chinese dataset of the NTCIR Opinion Analysis Task Test Collection and found to perform well. Experimental results testify that social tags are positively conducive to opinion analysis.