With the development of the Internet, short text has been everywhereon the Internet. Typical examples are social media sites like English basedTwitter and Chinese based Weibo. They are platforms that users share,publish and seek information. Users can access Twitter or Weibo via theweb, smartmobileterminalsorotherequipmentandwritetweetorcontentup to140characters. In mobile Internet era, more and more people areexpressing their views and ideas on the Internet everyday. Because socialmedia contains a large number of contents shared by users and the data isopen, so it attracts a lot of scholars in various felds. Sentiment analysis inshort text in these social media is a major task. In simple terms, sentimentanalysis is the analysis of the preferences and views expressed in the texttowards a specifc topic. It is used for decision support and public opinionanalysis. This paper attempts to do sentiment analysis on both English andChineseshorttextinsocialmediabasedonthetheircharacteristics. Ratherthan high-cost manual labeled data, this paper attempts to use a lot of non-artifcial noise labeled data to train the model. Lots of previous researchfocused on target independent sentiment analysis, this paper with the useof noise labeled data proposes to use dependency parsing to extract targetdependent feature to do target dependent sentiment analysis. |