Online Forum Thread Retrieval using Pseudo Cluster Selection and Voting Techniques

TL;DR

This paper introduces a combined model for online forum thread retrieval that integrates pseudo cluster selection and voting techniques, improving retrieval accuracy by focusing on message scoring and aggregation methods.

Contribution

It presents a novel combination of existing thread retrieval approaches, enhancing effectiveness through joint focus on input scoring and aggregation strategies.

Findings

Some combined models outperform baseline methods statistically.

The integrated approach improves thread retrieval accuracy.

Focus on input and aggregation enhances retrieval performance.

Abstract

Online forums facilitate knowledge seeking and sharing on the Web. However, the shared knowledge is not fully utilized due to information overload. Thread retrieval is one method to overcome information overload. In this paper, we propose a model that combines two existing approaches: the Pseudo Cluster Selection and the Voting Techniques. In both, a retrieval system first scores a list of messages and then ranks threads by aggregating their scored messages. They differ on what and how to aggregate. The pseudo cluster selection focuses on input, while voting techniques focus on the aggregation method. Our combined models focus on the input and the aggregation methods. The result shows that some combined models are statistically superior to baseline methods.