Discriminative Sentence Modeling for Story Ending Prediction

TL;DR

This paper introduces Diff-Net, a neural network that improves story ending prediction by modeling differences at multiple semantic levels, significantly outperforming existing methods on the Story Cloze Test dataset.

Contribution

The paper presents a novel neural network, Diff-Net, which effectively discriminates story endings at multiple semantic levels, enhancing prediction accuracy.

Findings

Diff-Net outperforms previous models on SCT dataset

Ablation studies reveal the importance of multi-level semantic modeling

BERT-based models show interesting comparative results

Abstract

Story Ending Prediction is a task that needs to select an appropriate ending for the given story, which requires the machine to understand the story and sometimes needs commonsense knowledge. To tackle this task, we propose a new neural network called Diff-Net for better modeling the differences of each ending in this task. The proposed model could discriminate two endings in three semantic levels: contextual representation, story-aware representation, and discriminative representation. Experimental results on the Story Cloze Test dataset show that the proposed model siginificantly outperforms various systems by a large margin, and detailed ablation studies are given for better understanding our model. We also carefully examine the traditional and BERT-based models on both SCT v1.0 and v1.5 with interesting findings that may potentially help future studies.