WSD Team's Approaches for Textual Entailment Recognition at the NTCIR10 (RITE2).
Daiki Ito, Masahiro Tanaka, Hayato Yamana · 2013
In this paper, we describe the WSD team’s approaches to textual entailment recognition task (RITE) at NTCIR-10[1], a conference held in June 18-21, 2013, at NII in Tokyo, Japan, and present experimental results for three Japanese subtasks, called “Binary Class ” (BC), “Multi Class ” (MC) and “Entrance Exam BC” (ExamBC). Our approach employs two supervised learning techniques: support vector machine (SVM) and logistic regression (LR). For the binary classification subtasks (BC and ExamBC), we propose hand-coded rules to classify text into entailment or non-entailment categories, while for the multi-classification subtask (MC), we used the bidirectional features for the texts. The best performance in three runs achieved precision of 80.66 % in BC subtask, 69.53 % in MC subtask and 67.86 % in ExamBC subtask. We won second place for BC and MC, and third place for ExamBC. Categories and Subject Descriptors