本系列文章主要针对方面级分类，该项任务的本质仍然是分类任务，但不同于一般的文本分类和情感分析任务，其还涉及对aspect word或者target word的信息提取。
本系列文章主要针对方面级分类(aspect-level sentiment classification)、target-dependent sentiment classification.另外还会记录了三篇与该任务不直接相关的文章阅读笔记：cross-domain sentiment classification、cross-lingual sentiment classification、document-level multi-aspect sentimnet classification.
文章阅读笔记记录顺序：
1. 2016 COLING——Effective LSTMs for Target-dependnt sentiment classification.
propose
TD-LSTM
and
TC-LSTM
;
2.2016 EMNLP——Attention-based LSTM for Aspect-level sentiment classification.
propose
AE-LSTM
and
AT-LSTM
and
ATAE-LSTM
;
3.2017 IJCAI——Interative attention networks for aspect-level sentimnet classification.
4.2016 EMNLP——Aspect-level sentiment classification with deep memoy network.
5.2017 ACL——attention modeling for targeted sentiment
6.2017 EMNLP——Recurrent attention on memory for aspect sentiment analysis.
7.2018 ACL——Exploiting Document knowledge for aspect-level sentiment classification.
8.2019 ACL——Progressive self-supervised attention learning for aspect-level sentiment aanalysis.
9. 2019 ACL ——Context-aware embedding for target aspect-based sentiment analysis.
相关任务paper：
10.2016 EMNLP ——Learning sentence embedding with auxiliary tasks for cross-domain sentiment classification(Yes)
11.2016 EMNLP——Attention-based LSTM network for cross-lingual sentiment classification(Yes)
12.2017 EMNLP——Document-level Multi-aspect sentiment classification as machine comprehension.(Yes)
附：
1)与aspect level情感分析一样，不同的上下文信息对于判断不同的target的情感倾向贡献不同，对于target-dependent情感分析来说，主要问题就是将target word和相应的context words的语义信息结合起来，由此推断不同上下文对判断不同的target的情感倾向的影响，分析相应target的情感倾向
2)cross-domain:source domain-target domain
3)cross-lingual:source language-target language
4)document-level multi-aspect 情感分析相较于简单的aspect-level sentiment classification 更复杂一点。