Information extraction of public complaints on Twitter text for bandung government

摘要

Tweets about public complaints from social networking site Twitter is growing significantly. This can be an opportunity for the government, such as Bandung Government, to obtain information that is important to improve public satisfaction. This research explore and analyze how the way to obtain the public complaints information from tweets. Classification-based approach is done through two main tasks, namely named entity recognition and relation extraction. Named entity recognition experiment achieves highest f-measure of 85.6% with 8 sets of features, namely lexical, NE tags, elements of Twitter, orthography, token kind, gazetteer, clue list and stopword list. Meanwhile, relation extraction experiment achieves highest f-measure of 77.2% with 5 sets of features, namely lexical, NE tags, number of words, one word before the first entity and clue list. Maximum f-measure on the two main tasks is obtained by Sequential Minimum Optimization algorithm.