Predicting Prospective Peer Helpers to Provide Just-In-Time Help to Users in Question and Answer Forums

Question and answer forums are becoming more popular as increasing numbers of lifelong learners rely on such forums to receive help about their learning needs. Stack Overflow (SO) is an example of such a forum used by millions of programmers. The ability of users to receive timely answers to questions is crucial to the sustainability of such forums and for successful lifelong learning. In SO we have observed that the number of questions answered within 15 minutes have diminished with more questions taking a longer time to get answered or remaining unanswered in some cases. This suggests the need for an effective approach in predicting prospective helpers who can provide timely answers to the questions. In this paper, we seek to explore strategies to match helpers and help seekers. In particular we wish to use these strategies to predict which SO users will provide timely answers to questions asked in SO, and then compare these predictions to the users who actually answered the questions. In making these predictions we looked at 3 time frames of user data: 1 month, 3 months and 6 months. We used 5 basic strategies: frequency, knowledgeability, eagerness, willingness, recency; and we compared the success rates of each strategy in making predictions on 3 different success criteria: predicting the first answerer, predicting the answerer most liked by the asker of the question, and predicting the answerer rated most highly by other SO users. We then incorporated a timeliness measure, which takes into consideration how quickly the user provides answers to questions in the past, which helped us to achieve a higher success rate. The results of our study are an improvement over a similar previous study of SO and we hope will form the basis of methods for recommending peers in online forums who can provide just-intime help to lifelong learners as their knowledge needs evolve and change.