文脈理解における再生課題正答率の眼球運動特徴による予測の試み

摘要

To determine the possibility of predicting the recall accuracy of reading and memorizing tasks using features of eye movements, a contextual understanding task experiment was conducted. The relationship between eye movement during memorization and recall performance was hypothesized. Viewer's eye-movements during the definition statements were observed, and twelve directional features of the eye-movements across a two-dimensional space were created. The prediction procedure was developed using Support Vector Regressions with Gaussian kernel. The correlation relationship between predicted and experimental recall accuracies was significant, therefore the hypothesis was supported. Also, it was found that the prediction performance depended on the combination of features of eye movements.%本報告では，文章提示に基づく記憶再生課題において，内容の理解と記憶を求める定義文に対する眼球運動の特徴情報から，再生課題の正答率を予測できるかを調べた．定義文を読んでいる間の眼球運動の特徴量は，注視位置，注視時間，サッカード長，サッカード時間のそれぞれについて，文章を提示した画面上での12方向，合計48で表現した．課題の難易度に相当する定義文の数や，再生課題の正答率で特徴量を比較したところ，一部の方位で有意な差が見られ，記憶時の眼球運動が正答率に関係することを確認した．定義文に対する眼球運動の特徴量と正答率との関係をサポートベクトル回帰を用いて表現した．眼球運動の特徴量から予測された正答率と実際の正答率との相関係数を調べた結果，相関関係が確認された．