Detecting Significant Events in Arabic Microblogs using Soft Frequent Pattern Mining

Abstract

nowadays, people use microblogs as a main platform to write about events that occur in their environment. Many researches have been conducted for event detection on the English language, but, Arabic context has not received much research. Furthermore, existing approaches rely on platform dependent features such as hashtags, mentions, or retweets, which make their approaches less efficient when these features are not presented. Further, some approaches which depend only on bursty or frequently used words, detect general viral topics instead of event related topics. In this work, we present a new approach for detecting events written in Arabic using frequent event triggers. The approach first identifies the part of speech tags of a sentence and then analyze them to extract event triggers. A soft frequent pattern mining method is applied to find co-occuring event triggers. The approach has been evaluated using a subset of the Evetar dataset. We divided the data into timely constrained windows to mimic the data stream behavior. Two experiments of different time intervals were conducted, 6-hours and one-day time intervals. We achieved an average F-meaure of 0.644 and 0.717. The results show that our approach outperformed some widely known approaches and it was comparable with others.