A novel method of evaluating tweets is introduced in this paper. Most retrieval algorithms do not differentiate the structure of tweets. The authors show convincingly that it does have an impact on retrieval effectiveness. In their study, two types of tweets are evaluated separately, those containing only plain text, and those containing any URLs.

The key is that in ranking the tweets with URLs, the ranker considers the content of the page(s) pointed to by the URLs in addition to the tweets. After ranking the two types of tweets, a support vector machine-based classifier with 18 features is used to evaluate the relevance between the tweets and the query. If a tweet is time-sensitive, the temporal information of both the tweet and its parent is taken into consideration.

Data from TREC 2011, TREC 2012, and TREC Tweets 2011 are used to evaluate the algorithm. The results indicate that (1) it is more effective to rank the two types of tweets separately and then merge them; (2) incorporating temporal information yields further improvements; and (3) the proposed “method compares favorably with state-of-the-art methods in retrieval effectiveness.” The novelty of the proposed method is that it offers the ability to rank the tweets with and without URLs separately and to incorporate the temporal information in ranking the tweets.

The paper is well written and self-contained. Many examples illustrate the concepts discussed. The readers can explore the topic further using the abundant references provided.