A deep neural spoiler detection model using a genre-aware attention mechanism

Buru Chang, Hyunjae Kim, Raehyun Kim, Deahan Kim, Jaewoo Kang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Citations (Scopus)

Abstract

The fast-growing volume of online activity and user-generated content increases the chances of users being exposed to spoilers. To address this problem, several spoiler detection models have been proposed. However, most of the previous models rely on hand-crafted domain-specific features, which limits the generalizability of the models. In this paper, we propose a new deep neural spoiler detection model that uses a genre-aware attention mechanism. Our model consists of a genre encoder and a sentence encoder. The genre encoder is used to extract a genre feature vector from given genres using a convolutional neural network. The sentence encoder is used to extract sentence feature vectors from a given sentence using a bi-directional gated recurrent unit. We also propose a genre-aware attention layer based on the attention mechanism that utilizes genre information for detecting spoilers which vary by genres. Using a sentence feature, our proposed model determines whether a given sentence is a spoiler. The experimental results on a spoiler dataset show that our proposed model which does not use hand-crafted features outperforms the state-of-the-art spoiler detection baseline models. We also conduct a qualitative analysis on the relations between spoilers and genres, and highlight the results through an attention weight visualization.

Original languageEnglish
Title of host publicationAdvances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings
PublisherSpringer Verlag
Pages183-195
Number of pages13
ISBN (Print)9783319930336
DOIs
Publication statusPublished - 2018 Jan 1
Event22nd Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, PAKDD 2018 - Melbourne, Australia
Duration: 2018 Jun 32018 Jun 6

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10937 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other22nd Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, PAKDD 2018
CountryAustralia
CityMelbourne
Period18/6/318/6/6

Keywords

  • Attention mechanism
  • Classification
  • Deep learning
  • Spoiler alert
  • Spoiler detection

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'A deep neural spoiler detection model using a genre-aware attention mechanism'. Together they form a unique fingerprint.

Cite this