Retrieving Similar Discussion Forum Threads: A Structure based Approach

Online forums are becoming a popular way of finding useful<br/>information on the web. Search over forums for existing discussion<br/>threads so far is limited to keyword-based search due<br/>to the minimal effort required on part of the users. However,<br/>it is often not possible to capture all the relevant context in a<br/>complex query using a small number of keywords. Examplebased<br/>search that retrieves similar discussion threads given<br/>one exemplary thread is an alternate approach that can help<br/>the user provide richer context and vastly improve forum<br/>search results. In this paper, we address the problem of<br/>finding similar threads to a given thread. Towards this, we<br/>propose a novel methodology to estimate similarity between<br/>discussion threads. Our method exploits the thread structure<br/>to decompose threads in to set of weighted overlapping<br/>components. It then estimates pairwise thread similarities<br/>by quantifying how well the information in the threads are<br/>mutually contained within each other using lexical similarities<br/>between their underlying components. We compare our<br/>proposed methods on real datasets against state-of-the-art<br/>thread retrieval mechanisms wherein we illustrate that our<br/>techniques outperform others by large margins on popular<br/>retrieval evaluation measures such as NDCG, MAP, Precision@k<br/>and MRR. In particular, consistent improvements of<br/>up to 10% are observed on all evaluation measures