Chou, P.-H.P.-H.ChouTsai, R.T.-H.R.T.-H.TsaiHsu, J.Y.-J.J.Y.-J.HsuYUNG-JEN HSU2020-05-042020-05-04201714327643https://scholars.lib.ntu.edu.tw/handle/123456789/490532https://www.scopus.com/inward/record.uri?eid=2-s2.0-84979986867&doi=10.1007%2fs00500-016-2273-0&partnerID=40&md5=ed7edcc8b1d2ab2381ed5f7af8218e1dA sentiment dictionary is a valuable resource in sentiment analysis research. Previous work has propagated sentiment values from existing dictionaries via semantic networks to build wide-coverage dictionaries efficiently. Unfortunately, this blind propagation method tends to incorrectly estimate sentiment values the further along the chain it goes from the seed word because it does not consider word senses in context. In this work, we propose a context-aware propagation method on Chinese ConceptNet to help resolve this issue. In our approach, we represent contexts using LDA topic modeling by generating a topic for each context. We can then assign concepts different sentiment values for different topics when propagating sentiments on Chinese ConceptNet. Our experiments on both microblog posts and drama dialogue subtitles show that our context-aware approach improves the accuracy of sentiment polarity prediction. © 2016, Springer-Verlag Berlin Heidelberg.Commonsense knowledge; Context-aware; Sentiment analysis; Sentiment dictionary; Topic model; Value propagationSemantics; Commonsense knowledge; Context-Aware; Sentiment analysis; Sentiment dictionaries; Topic Modeling; Data miningjournal article10.1007/s00500-016-2273-0