Abstract: Online reviews are written by the user for the related products or services experience, online reviews have become an important data resource for obtaining user's needs. The quality of comments has seriously interfered with the accuracy and credibility of demand mining. Identifying useful online product review can contribute to exactly analyzing users' requirement. This paper proposes a method to calculate the similarity of product reviews by combining the similarity degree of semantic similarity of comment text and the similarity of product emotion, which can better reflect the similarity of reviews; then from the view of users' demand, uses the social network theory to construct a product reviews network that takes review as network nodes and the similarity of reviews as edges, thus obtain useful reviews by the outlier detection technique based on k-means clustering. The results show that this method can effectively identify the useless reviews in obtaining user requirement, which provides the prerequisite for improving the accuracy of user requirements analysis.