Expanding Comments Using Previously Responded Posts for Short Text Conversation in Open Social Media

Abstract

We report a study on improving the retrieval of comments made to previous posts that can potentially be relevant to new posts in the open social media domain. This retrieval task, known as an important component of short text conversation, is challenging due to various reasons, among which is the fact that both posts and comments are usually very short, thus not providing enough context for typical information retrieval systems to be very effective. We designed a technique of expanding each comment by adding words to it that are selected from the post(s) that the comment previously responded to. Our experiment using a Sina WeiBo collection shows that by properly considering the number of posts that a comment has previously responded to, the number of responding comments of each of these posts, as well as the number of words to be selected for expansion, the proposed technique can result in slightly improved retrieval effectiveness. Furthermore, the technique can retrieve relevant comments that have not been returned by any other previous systems.