As the Internet becomes more widespread, the comments according to various products are also growing, these comments which contain abundant information are extremely analysis value for the manufacturers and consumers. Based on the correct analysis of opinion object, the results of opinion mining can be accurate and reliable. Based on the summarization of predecessors' achievements, according to the customer comments on the network, this paper proposes a new evaluation object extraction method. This method is based on the ICTParser syntactic analysis and IR dependence analysis, adopts the likelihood testing methods to screen irrelative candidate opinion objects. The experimental results preliminarily verified the validity of this method.