From keywords to social tags: Tagging for dialogues

This paper proposes an unsupervised method for generating informative tags for multi-party dialogue in an open domain. Our model first extracts keywords from text through a multi-weighting framework, which includes frequency weighting, sentence weighting, speaker weighting and position weighting. Then we get their bigrams through frequent pattern matching. In order to generate more flexible and socialized tags, we expand keywords and their bigrams by exploring tag associations mined from a famous bookmarking web del.icio.us. Finally we rank the three parts of tag candidates under a uniform metric. Unlike previous models used for tag recommendation task, our model needs neither seed tags for source texts nor a complete tag set. It can recommend new tags out of the historical tag set. We evaluate our methods on 10,265 Chinese dialogues. Experimental results show our method outperform previous models like TextRank, TFIDF rank and KNN.