Research On Extraction Of Implicit Opinion Targets

The opinion targets are defined as implicit opinion targets if they are not shown in the comment text. In order to address the issue that current extract method of implicit o pinion targets is highly dependent on the tagged corpus and the limitation to be aware o f the relevant knowledge. The paper proposes a method to extract implicit opinion targets based on the feature terms and association relationship of the opinion targets to solve the problem that opinion targets are neglected in comments.In particular, a feature term refers to an adjective or a verb which is used to dominating or modifying opinion targets. Firstly, this method aims to create a new opinion target model through analysing grammatical structure characteristics of Chinese short text, and extract a pool of candidate opinion targets from corpus using the above-mentioned model. Through expanding the set of feature terms, filtering noise and clustering by the method of k-means to optimize candidate opinion targets, which solve the problem about information lacking, noisy and co-reference effectively. Thus we can gain the collection of opinion targets and feature terms in the field.Secondly,we also use the method of CHI to calculate the weight of the feature terms to get the indicative power vector. Finally, using it to extract the implicit opinion targets from the testing corpus. We have also validated the availability of this method by applying it to both the restaurant and hotel.