Deteksi trending topic menggunakan tiga pendekatan yaitu berbasis textual content, social content dan hybrid. Ketiga metode tersebut memiliki permasalahan yang berbeda. Pertama, deteksi trending topic berbasis textual content memiliki permasalahan dalam menggunakan prapemrosesan yang kompleks. Kedua, deteksi trending topic berbasis social content belum mampu mendeteksi konten trending topic. Ketiga, deteksi trending topic berbasis hybrid sangat dipengaruhi oleh pengguna Twitter dengan jumlah follower yang besar (jumlah follower ribuan bahkan jutaan). Pada penelitian ini dilakukan penerapan metode baseline berbasis textual content yaitu BN-grams dan Doc-p pada tweet berbahasa Indonesia. Selanjutnya, BN-grams dilakukan modifikasi pada pembentukan klaster dan perangkingan topik menjadi metode Non Overlap Trending Topic (NOTT) dan Overlap Trending Topic (OTT). Kemudian, dilakukan penggabungan antara Link Anomaly (berbasis social content) dan deteksi burst Kleinberg (berbasis textual content) dengan hasil akhir berisi intersection interval waktu antara Link Anomaly dan deteksi burst Kleinberg serta menjadi metode baru yaitu Overlap Time Interval Trending Topic (OTITT). Metode NOTT dan OTT memiliki empat tahapan yang sama yaitu: prapemrosesan, pembentukan klaster menggunakan Frequent Term Based Clustering (FTC) atau Hierarchical Frequent Term Based Clustering (HFTC), perangkingan topik dan pemodelan topik. Metode OTITT terdiri dari empat tahapan utama yaitu: ekstraksi interval waktu dari Link Anomaly, ekstraksi interval waktu dari deteksi burst Kleinberg, intersection interval waktu Link Anomaly dan deteksi burst Kleinberg serta pemodelan topik. Keseluruhan metode usulan dilakukan pengujian dengan membandingkan hasil trending topic metode usulan dengan trending topic yang berasal dari media siber maupun Twitter. Berdasarkan hasil pengujian, metode BN-grams memiliki nilai topic recall lebih tinggi dibandingkan Doc-p dengan nilai 55%. Disisi lain, metode OTT memiliki nilai topic recall lebih tinggi dibandingkan NOTT, BN-grams dan Doc-p dengan nilai 33%. Metode OTITT memiliki nilai topic recall 47% dan tertinggi dibandingkan NOTT, OTT, BN-grams maupun Doc-p

Detection of trending topic using three approaches that are textual content, social content, and hybrid. The three methods have different problems. First, trending topic detection based on textual content has difficulties in using complex preprocesses. Secondly, trending topic detection based on social content has not been able to detect trending topic content. Thirdly, trending topic detection with the hybrid based method is influenced by Twitter users with considerable influence (number of followers thousands and even millions). In this research, we explore of textual content baseline method with BN-grams and Doc-p on Indonesian tweets. Furthermore, we modify BN-grams in the step of generating a cluster and topics ranking into Non-Overlap Trending Topic (NOTT) and Overlap Trending Topic (OTT) methods. Then, we combine between Link Anomaly (based on social content) and Burst Kleinberg detection (based on textual content) with the final result containing the intersection of time intervals between Link Anomaly and detection of Kleinberg bursts and becoming a new method of Overlap Time Interval Trending Topic (OTITT). The NOTT and OTT method has the same four steps, i.e., preprocessing, cluster formation using Frequent Term Based Clustering (FTC), Hierarchical Frequent Term Based Clustering (HFTC), topic ranking and topic modeling. The OTITT method consists of four main stages: extraction of time intervals from Link Anomaly, extraction of time intervals from detection of burst Kleinberg, the intersection of Link Anomaly time intervals and detection of Kleinberg bursts and topic modeling. We evaluate the entire proposed method by comparing the results of the proposed method trending topic with trending topics originating from cyber media and Twitter. Based on the experimental results, BN-grams method has a higher topic recall value than Doc-p with 55%. On the other hand, the OTT method has more topic recall value than NOTT, BN-grams, and Doc-p with 33%. The OTITT method has a 47% topic recall topic and is highest than NOTT, OTT, BN-grams or Doc-p.

Kata Kunci : OTITT,BN-grams,Link Anomaly,NOTT,OTT