Many online communities allow their members to provide information helpfulness judgments that can be used to guide other users to useful contents quickly. However, it is a serious challenge to solicit enough user participation in providing feedbacks in online communities. Existing studies on assessing the helpfulness of user-generated contents are mainly based on heuristics and lack of a unifying theoretical framework. In this article we propose a text classification framework for finding helpful user-generated contents in online knowledge-sharing communities. The objective of our framework is to help a knowledge seeker find helpful information that can be potentially adopted. The framework is built on the Knowledge Adoption Model that considers both content-based argument quality and information source credibility. We identify 6 argument quality dimensions and 3 source credibility dimensions based on information quality and psychological theories. Using data extracted from a popular online community, our empirical evaluations show that all the dimensions improve the performance over a traditional text classification technique that considers word-based lexical features only.