Incorporating author's activeness in online discussion in thread retrieval model

Online forum is one of user-generated contents available on the Internet that provides platform for knowledge sharing. However, not all messages posted can be considered of high quality and as it increases in its availability, finding quality information becomes more important and challenging. Thread retrieval model is very important in helping users to find relevance information pertaining to their topic search. As quality of post messages depends upon the author, this study aims to look at how ranking threads based on author’s activeness in a forum could improve thread retrieval task compared to non-quality based ranked list. Voting models were used to convert message level quality features into thread level features and learning to rank method to combine nine features of activeness dimension for thread scoring. Different combinations of nine features under the activeness dimension with different ranking strategies are studied and its performances also compared using normalized discounted cumulative gain (NDCG) as performance measure. 2555 models were generated and 23 models are identified as among the best model.