Open magicpieh28 opened 5 years ago
Author: Xin Li, Lidong Bing, Piji Li, Wai Lam Link: https://arxiv.org/pdf/1811.05082.pdf
boundary情報を用いて感情分類とキーワード抽出を同時にこなすunified modelを提案
評価：SemEval ABSA challenges - Laptop, restaurant/ Twitter 訓練：SemEval ABSA 2014, 2015, 2016の訓練データ
auxiliary task: ターゲットboundary予測タスク・ターゲット単語探索（Lexicon使用）
CRF
NN-CRF: feature extractors
TNet, HAST HAST-TNet: state-of-the-art
LSTM-unified
LSTM-CRF-1
LSTM-CRF-2
LM-LSTM-CRF
pre-trained GloVe.840B.300d: out-of-vocabulary words are sampled from the uniform distribution U(-0.25, 0.25)(Kim 2014) 重み初期化：Glorot Unirom strategy (Glorot and Bengio 2010)/ U(-0.2, 0.2) 最適化：50 epoch, beta1=beta2=0.9, 学習率10^{-3}, decay rate(Lample et al. 2016), dropout0.5 最高精度：F1 隠れ層次元：50 最大proportionε：0.5 window s: 3(1~5)
https://github.com/lixin4ever/E2E-TBSA
Li et al. 2018a. Transformation networks for target-oriented sentiment classification. In ACL Mitchell et al. 2013. Open domain targeted sentiment. In EMNLP Pontiki 2016. Aspect based sentiment analysis. In SemEval Zhang, Zhang, and Vo 2015. Neural networks for open domain targeted sentiment. In EMNLP
about this paper
Author: Xin Li, Lidong Bing, Piji Li, Wai Lam Link: https://arxiv.org/pdf/1811.05082.pdf
boundary情報を用いて感情分類とキーワード抽出を同時にこなすunified modelを提案
What they want to do
problems
how to solve
experiment
datasets
評価：SemEval ABSA challenges - Laptop, restaurant/ Twitter 訓練：SemEval ABSA 2014, 2015, 2016の訓練データ
auxiliary task: ターゲットboundary予測タスク・ターゲット単語探索（Lexicon使用）
compared models
Mitchell et al. 2013
CRF
Zhang, Zhang, and Vo 2015
NN-CRF: feature extractors
Li et al. 2018a and Li et al. 2018b
TNet, HAST HAST-TNet: state-of-the-art
-
LSTM-unified
Lample et al. 2016
LSTM-CRF-1
Ma and Hovy 2016
LSTM-CRF-2
Liu et al. 2018
LM-LSTM-CRF
settings
pre-trained GloVe.840B.300d: out-of-vocabulary words are sampled from the uniform distribution U(-0.25, 0.25)(Kim 2014) 重み初期化：Glorot Unirom strategy (Glorot and Bengio 2010)/ U(-0.2, 0.2) 最適化：50 epoch, beta1=beta2=0.9, 学習率10^{-3}, decay rate(Lample et al. 2016), dropout0.5 最高精度：F1 隠れ層次元：50 最大proportionε：0.5 window s: 3(1~5)
results
code
https://github.com/lixin4ever/E2E-TBSA
next
Li et al. 2018a. Transformation networks for target-oriented sentiment classification. In ACL Mitchell et al. 2013. Open domain targeted sentiment. In EMNLP Pontiki 2016. Aspect based sentiment analysis. In SemEval Zhang, Zhang, and Vo 2015. Neural networks for open domain targeted sentiment. In EMNLP