Pembobotan Fitur Tekstual dengan Inferensi Metaheuristik untuk Pengurutan Jawaban (original) (raw)

Abstrak — In this paper, an approach to construct a dynamic feature weighting in question answering system is proposed. In order to find a set of optimum weights for a question and its answer (candidate) pair(s), a combination of genetic algorithm and logistic regression technique are introduced in our approach. As part of our experimental study, a standard dataset of non-factoid question and answer pairs, called UIUC, is used as a testing-case. The experiment results show that our approach can produce a set of feature weights which leads to a better performance compared to the conventional static weighting. Keywords— metaheuristik, algoritma genetika, regresi logistik sistem tanya jawab, pengurutan jawaban. I. PENDAHULUAN Sistem tanya jawab (STJ) adalah sebuah penerapan temu balik informasi yang diharapkan dapat memberikan sebuah kandidat jawaban secara langsung. Ditinjau dari sisi artitektur perangkat lunak, STJ memiliki empat komponen utama [1], yaitu: 1. analisis pertanyaan; 2. ...