Abstract: hhe paper mainly focused attention on how to mine opinion topic from online subjective text set,and identified the orientation of these opinion topics. It combined the heuristic rule and co-occurrence probability to identify area noun phrase, and adopted Latent Dirichlet Allocationn(LDA) to obtain local opinion topic. Then, it computed the sentence orientation with a method of multiple features fusion, and obtained the opinion topic orientation by adding up the number of each orientation of the sentences belonging to the specified topic. Experimental results show that the method can identify the topic and their orientation effectively.

Key words: LDA, Opinion topic, Orientation identification