[2G1-GS-2d] 機械学習：シーケンシャルデータの処理

[2G1-GS-2d-01] Optimal transport based sequence matching with grouped elements.

〇Mitsuhiko Horie¹, Hiroyuki Kasai¹ (1. Waseda University)

Keywords:Sequence Matching, Optimal Transport

We propose a new sequence matching method for measuring the distance between two different sequence data. In addition to defining the calculation of sequence matching as an optimal transport problem, we seek to group multiple mutually-related elements within the target sequence, and perform group-wise matching instead of element-wise matching. The grouping process absorbs the difference of the time distribution of instances in the two different sequences, and the obtained matching takes the relevance of adjacent elements into account. We conducted evaluation experiments across three datasets to compare the proposed method with state-of-the-art ones, and show the effectiveness of the proposed method.