Ensemble BERT-BiLSTM-CNN Model for Sequence Classification

• Item type: SIG Technical Reports(1)
• 公開日: 2021-09-09
• タイトル: Ensemble BERT-BiLSTM-CNN Model for Sequence Classification
• タイトル / 言語: en
• タイトル: Ensemble BERT-BiLSTM-CNN Model for Sequence Classification
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: National Institute of Informatics
• 著者所属: National Institute of Informatics
• 著者所属(英): en
• 著者所属(英): National Institute of Informatics
• 著者所属(英): en
• 著者所属(英): National Institute of Informatics
• 著者名: Vuong, Thi Hong; Atsuhiro, Takasu
• 著者名(英): Vuong, Thi Hong; Atsuhiro, Takasu
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Ensemble methods use multiple learning algorithms to obtain better predictive performance. Currently, deep learning models with multilayer processing architecture are showed that the performance is better than the traditional classification models. Ensemble deep learning models combine the advantages of both ensemble learning and deep learning such that the final model has better performance. This paper presents a novel ensemble deep learning method, achieving robust and effective sequence classification facing sparse data. We use the BERT (Bidirectional Encoder Representation from Transformers) as the word embedding method. Then, we integrate the BiLSTM (Bidirectional Long Short-Term Memory) and CNN (Convolutional Neural Network) with an attention mechanism for sequence classification. We evaluate our ensemble models with two datasets with the different baseline methods. The first dataset is from IMDB and contains 50,000 movie reviews, labeled with two sentiment classes. The second dataset is the Toxic Comment Dataset with more than 150,000 comments for six classes. The experimental results show that our proposed method provides an accurate, reliable, and effective solution for sequence data classification.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Ensemble methods use multiple learning algorithms to obtain better predictive performance. Currently, deep learning models with multilayer processing architecture are showed that the performance is better than the traditional classification models. Ensemble deep learning models combine the advantages of both ensemble learning and deep learning such that the final model has better performance. This paper presents a novel ensemble deep learning method, achieving robust and effective sequence classification facing sparse data. We use the BERT (Bidirectional Encoder Representation from Transformers) as the word embedding method. Then, we integrate the BiLSTM (Bidirectional Long Short-Term Memory) and CNN (Convolutional Neural Network) with an attention mechanism for sequence classification. We evaluate our ensemble models with two datasets with the different baseline methods. The first dataset is from IMDB and contains 50,000 movie reviews, labeled with two sentiment classes. The second dataset is the Toxic Comment Dataset with more than 150,000 comments for six classes. The experimental results show that our proposed method provides an accurate, reliable, and effective solution for sequence data classification.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10112482
• 書誌情報: 研究報告データベースシステム（DBS）; 巻 2021-DBS-173, 号 3, p. 1-6, 発行日 2021-09-09
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-871X
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会