Abstract: Sentiment element recognition is one of the key sub-tasks of sentiment analysis,and its goal is to identify the sentiment targets in the text.Sentiment target recognition is identified as the most fine-grained sentiment analysis task,and many researchers have conducted lots of research work on it.Since Chinese Microblog text is short and very flexible,which is often not standardized and contains a lot of noisy information in the text,it brings new challenges to the Chinese Microblog sentiment analysis research.At present,most of sentiment target recognition methods are based on rules or statistical learning methods using flat features,which can not distinguish between noisy information and sentiment targets very well,resulting in the low recognition performance.According to the characteristics of Chinese Microblog,a novel sentiment element recognition approach based on convolution tree kernel was proposed.Firstly,the approach analyses the part of speech(POS) and dependency relationship of the Microblog sentences,and takes the nouns in the sentences as candidate sentiment elements.Secondly,it adopts two different pruning strategies to obtain every candidate’s structured information.Finally,convolution tree kernel method is used to calculate the similarity of dependency tree,which is the foundation of sentiment elements recognition.The experiments of NLP & CC2012 and NLP & CC2013 Chinese Microblog sentiment target analysis tasks show the performance of this approach is improved significantly comparing to the baseline.

Key words: Sentiment target recognition,Chinese microblog,Convolution tree kernel,Pruning strategy