Query Performance Prediction and Topic Shift in Microblog Retrieval

In Microblog retrieval, a system’s ability to know when to ”shut up” and how many results to return for a given query can have huge impact on its performance [1]. In addition, since relevant but redundant tweets will be deemed as irrelevant, it is also important to detect whether a tweet contain novel information or not. Therefore, in this year’s Real-Time Summarization Track, we focus on estimating result cut-off thresholds and redundancy thresholds. Query performance prediction techniques [2, 3] can be used to predict the performance of a query and therefore would be helpful in deciding both thresholds. Moreover, we define topic shift of a query as the change of subtopics or aspects discussed or mentioned on Twitter over time. We suggests that using it could help us further refine the thresholds since it reflects how much new information emerges on Twitter.