Trending topics detection of Indonesian tweets using BN-grams and Doc-p

Researches on trending topics detection, especially on Twitter, have increased and various methods for detecting trending topics have been developed. Most of these researches have been focused on tweets written in English. Previous researches on trending topics detection on Indonesian tweets are still relatively few. In this paper, we compare two methods, namely document pivot and BN-grams, for detecting trending topics on Indonesian tweets. In our experiments, we examine the effects of varying the number of topics, n-grams, stemming, and aggregation on the quality of the resulting trending topics. We measure the accuracy of trending topics detection by comparing both algorithms with trending topics found in local news and Twitter trending topics. The results of our experiments show that using ten topics produces the highest topic recall; that using trigrams in BN-grams results in the highest value topic recall; and that using aggregation reduces the quality of trending topics produced. Overall, BN-grams has a higher value of topic recall than that of document pivot. Keywords: Trending topics detection, Twitter, BN-grams, Document pivo