Popularity-based temporal relevance estimation for micro-blogging retrieval

Finding relevant information among the vast amounts of data generated continuously by modern micro-blogging platforms has opened new challenges in information retrieval. Recent studies on time-based retrieval have shown that identifying the relevant time periods to be incorporated into the retrieval process is promising; by relevant time period we mean the peak time of a query that satisfies the temporal needs of the user's query. Or in other words, a time period at which the potential to find accurate matches for the query in a set of retrieved documents is relatively high when compared with other time periods. We refer to this as temporal relevance. In this paper, using data collected from Twitter, we propose a new temporal relevance estimation technique based on tracking documents published by the popular users, who have high indegree (i.e., number of followers). In this study we concentrate on queries that are short (one word) and popular, i.e., constantly consumed by micro-bloggers. We choose the simple frequency-based technique to estimate the relevant time period as a baseline against which we evaluate our technique. The results of our technique either match or suggest a better time period as the most relevant one, when compared with the baseline. In fact, for the type of queries in our study, narrowing our focus to the documents published by popular users produces a query-to-documents matching pattern that uncovers some information about temporal relevance that might otherwise be hidden. Also, our matching pattern reflects the nature of the real-world news events that are related to the user's query more so than the baseline, thus revealing the important time frames more clearly.