Mining online forums for valuable contributions

Web 2.0 resources such as online forum messages are of growing value in supporting education and research, and new information systems are aimed at synthesizing these resources into environments facilitating collaboration, learning and other goals. To support the automatic identification of valuable Web 2.0 resources, this paper presents an approach for estimating the information value of messages posted to an online forum. The system integrates key insights that include (1) how to reliably obtain large quantities of both positive and negative examples of such resources for use in training a machine learning algorithm, (2) the importance of considering authorship when analyzing the value of an online message, (3) the benefit of considering the subsequent message when analyzing a given message for its importance, and (4) how to develop and leverage a custom sentiment analysis model for use in automatically identifying high-value messages. In an evaluation using almost 90000 messages from an online forum, a baseline model for identifying valuable messages made over 50% more errors than the proposed model.