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 language | English |
---|---|
Title of host publication | Advances in Knowledge Discovery and Data Mining - 22nd Pacific-Asia Conference, PAKDD 2018, Proceedings |
Publisher | Springer Verlag |
Pages | 183-195 |
Number of pages | 13 |
ISBN (Print) | 9783319930336 |
DOIs | |
Publication status | Published - 2018 Jan 1 |
Event | 22nd Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, PAKDD 2018 - Melbourne, Australia Duration: 2018 Jun 3 → 2018 Jun 6 |
Publication series
Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
---|---|
Volume | 10937 LNAI |
ISSN (Print) | 0302-9743 |
ISSN (Electronic) | 1611-3349 |
Other
Other | 22nd Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, PAKDD 2018 |
---|---|
Country | Australia |
City | Melbourne |
Period | 18/6/3 → 18/6/6 |
Fingerprint
Keywords
- Attention mechanism
- Classification
- Deep learning
- Spoiler alert
- Spoiler detection
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science(all)
Cite this
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 proceeding › Conference contribution
}
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
AN - SCOPUS:85049377704
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 -