This paper presents a general detection framework, and develops a variety of content and structure features to find high quality threads. The feature selection algorithm, which is a combination of genetic algorithm, Tabu search and a machine learning algorithm, is designed to attain a better assessment of key features. In this paper, an experiment is done that focuses on the Tencent Message Boards. The experimental results, obtained from a large scale evaluation of over thousands of real web forum threads and user ratings, demonstrate the feasibility of modeling and detecting high quality threads. The proposed feature extraction methods, feature selection algorithms, and detection framework can be useful for a variety of domains such as Blogs and social network platforms.