Vietnamese Legal Question Answering with combined features and deep learning

Luu Hoai Linh, Hai-Long Nguyen, Nguyen Hai Yen, Thi-Hai-Yen Vuong

Abstract: Legal Question Answering is an arduous problem that is divided into certain phases, each with its own set of challenges. In this work, we have accomplished three tasks given by the ALQAC 2021 competition, which are aimed at addressing the aforementioned problem, by proposing the combined features (cosine similarity of TF-IDF, an average of word embedding; and Jaccard distance) accompanied by a classification model for task 1; ensemble learning multiple deep learning models for task 2. Finally, we employed a specifically modified mechanism for long documents to undertake task 3. All three methods perform satisfactory results and have profuse potential improvements.