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

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

Fingerprint

Encoder
Feature Vector
Model
Qualitative Analysis
Baseline
Visualization
Vary
Neural Networks
Neural networks
Unit
Experimental Results

Keywords

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

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Chang, B., Kim, H., Kim, R., Kim, D., & Kang, J. (2018). A deep neural spoiler detection model using a genre-aware attention mechanism. In Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings (pp. 183-195). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10937 LNAI). Springer Verlag. https://doi.org/10.1007/978-3-319-93034-3_15

A deep neural spoiler detection model using a genre-aware attention mechanism. / Chang, Buru; Kim, Hyunjae; Kim, Raehyun; Kim, Deahan; Kang, Jaewoo.

Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings. Springer Verlag, 2018. p. 183-195 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10937 LNAI).

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

Chang, B, Kim, H, Kim, R, Kim, D & Kang, J 2018, A deep neural spoiler detection model using a genre-aware attention mechanism. in Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10937 LNAI, Springer Verlag, pp. 183-195, 22nd Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, PAKDD 2018, Melbourne, Australia, 18/6/3. https://doi.org/10.1007/978-3-319-93034-3_15
Chang B, Kim H, Kim R, Kim D, Kang J. A deep neural spoiler detection model using a genre-aware attention mechanism. In Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings. Springer Verlag. 2018. p. 183-195. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-93034-3_15
Chang, Buru ; Kim, Hyunjae ; Kim, Raehyun ; Kim, Deahan ; Kang, Jaewoo. / A deep neural spoiler detection model using a genre-aware attention mechanism. Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings. Springer Verlag, 2018. pp. 183-195 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{446e2cebc53e4620a8891a25736edd76,
title = "A deep neural spoiler detection model using a genre-aware attention mechanism",
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.",
keywords = "Attention mechanism, Classification, Deep learning, Spoiler alert, Spoiler detection",
author = "Buru Chang and Hyunjae Kim and Raehyun Kim and Deahan Kim and Jaewoo Kang",
year = "2018",
month = "1",
day = "1",
doi = "10.1007/978-3-319-93034-3_15",
language = "English",
isbn = "9783319930336",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "183--195",
booktitle = "Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings",

}

TY - GEN

T1 - A deep neural spoiler detection model using a genre-aware attention mechanism

AU - Chang, Buru

AU - Kim, Hyunjae

AU - Kim, Raehyun

AU - Kim, Deahan

AU - Kang, Jaewoo

PY - 2018/1/1

Y1 - 2018/1/1

N2 - 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.

AB - 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.

KW - Attention mechanism

KW - Classification

KW - Deep learning

KW - Spoiler alert

KW - Spoiler detection

UR - http://www.scopus.com/inward/record.url?scp=85049377704&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85049377704&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-93034-3_15

DO - 10.1007/978-3-319-93034-3_15

M3 - Conference contribution

SN - 9783319930336

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 183

EP - 195

BT - Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings

PB - Springer Verlag

ER -