=Paper=
{{Paper
|id=Vol-2293/jist2018pd_paper4
|storemode=property
|title=Dependency Parsing and Bidirectional LSTM-CRF for Aspect-level Sentiment Analysis of Chinese
|pdfUrl=https://ceur-ws.org/Vol-2293/jist2018pd_paper4.pdf
|volume=Vol-2293
|authors=Huichao Xiong,Hua Yan,Zhixian Zeng,Binhui Wang
|dblpUrl=https://dblp.org/rec/conf/jist/XiongYZW18
}}
==Dependency Parsing and Bidirectional LSTM-CRF for Aspect-level Sentiment Analysis of Chinese==
https://ceur-ws.org/Vol-2293/jist2018pd_paper4.pdf
Dependency Parsing and bidirectional LSTM-CRF for
Aspect-level Sentiment Analysis of Chinese
Huichao Xiong1, Hua Yan1, Zhixian Zeng1 and Binhui Wang1*
1Nankai University, Tianjin, China
{2120160362,1511466, 2120170549}@mail.nankai.edu.cn
wangbh@nankai.edu.cn
Abstract. Aspect-level sentiment analysis of Chinese is to extract, aggregate
a21 apply fine-grained aspect-level sentiment information from text for senti-
ment understanding, and it is useful in various application domains. It usually
first extracts aspect terms and sentiment words simultaneously, then pairs the
aspect terms with sentiment words, and lastly classifies the aspect-level senti-
ment. In this sense, we formulate this problem as a pipeline of aspect terms and
sentiment words extraction through sequence labeling, aspect-sentiment word
pairing, and aspect-level sentiment classification. In this paper, we use a bidi-
rectional LSTM-CRF model to extract aspect terms and sentiment words, some
syntax rules based on dependency parsing to pair aspect and sentiment words,
and mainstream classifiers to determine the sentiment polarities.
Keywords: Aspect-level sentiment analysis, dependency parsing, bidirectional
LSTM-CRF.
1 Introduction
Aspect-level sentiment analysis (ABSA) has been used in various application do-
mains, including online marketing, corporate public opinion monitoring, and govern-
ment opinion survey. Previous work used topic models for ABSA, while these meth-
ods learned topics are overly abstract and their work focuses on document-level sen-
timent classification [1, 2]. In addition, most research focused on the English lan-
guage, only few work aims at ABSA of Chinese [3]. To understand sentiments at the
level of aspects, there are two fundamental tasks, i.e., aspect term extraction and as-
pect-based sentiment classification. Since it is difficult to combine them in one step,
in this paper we first extract and pair the aspect terms with sentiment words, and then
classify for the aspect-level sentiment.
We first extract aspect terms and sentiment words using a BI-LSTM-CRF (Bidirec-
tional Long-Short Term Memory Conditional Random Field) model that utilizes both
the contextual information in bidirectional pathways and the sentence-level tags
through a CRF layer. We then pair the aspects with sentiments using some syntax
rules. Lastly, we use classifiers to classify sentiment polarity of aspect terms.
_________
*Corresponding author
2
2 Approach
Our approach includes three parts: aspect/sentiment extraction by BI-LSTM-CRF,
pairing aspect and sentiment words by syntax rules and gaining sentiment polarity of
aspects through classifiers.
2.1 BI-LSTM-CRF
BI-LSTM-CRF model includes bidirectional LSTM and CRF (shown in Figure 1). It
uses both the preceding input features and the future input features.
Fig. 1. A BI-LSTM-CRF Model [4]
2.2 Syntax Rules
After aspect terms and sentiment words are extracted from reviews, some syntax rules
can help find the relation between the aspect terms and their corresponding sentiment
words, to generate the aspect-sentiment pairs. Considering there may have some sen-
timent words without any toward aspect terms, but we need get these pairs if the sen-
timent words exist. So, we should first detect aspects and sentiment words then gain
the dependency structure shows which words depend on (modify or are arguments of)
which other words. In order to get the relation between aspects and sentiment words
and find the actual aspect terms lastly, we use the Chinese NLP tool jieba
(https://github.com/fxsjy/jieba) for word segmentation and LTP (https://github.com/HIT-
SCIR/ltp) for dependency parsing. As illustrated in Table 1, their relations usually are
ATT (attribute), SBV (subject-verb) and IOB (indirect-object). As the aspect usually
appear in front of sentiment, their relation of position in a sentence can be used.
3
Table 1. Examples of Relation between Aspects and Sentiment Words
Sentences Aspects Sentiment Relation
words
这是一款漂亮的衬衫。 衬衫 漂亮 ATT
(This is a beautiful shirt.) (shirt) (beautiful)
iphone销量提升。 销量 提升 SBV
(The sales of iphone rises.) (sales) (rise)
抵制苹果品牌来支持Nokia。 销售 抵制 IOB
(Resist the selling of iphone to support Nokia.) (selling) (resist)
2.3 Classifiers
We form a lexicon from training set and use it to prepare the features for both training
and testing sets. Due to the known “no free lunch theorem” in supervised learning, we
use traditional classifiers, including Naïve Bayes (NB), Logistic Regression (LR) and
Support Vector Machines (SVM) to determine the polarity of sentiment words that
doesn’t appear in the lexicon.
3 Experiments
To evaluate the performances of the models on the two sub-tasks of ABSA: 1) aspect
term extraction 2) aspect-based sentiment classification. The metrics for the two sub-
tasks are precision (P), recall (R), and F1-score (F1). We compare the performance
across multiple models in Table2. We find that the models with CRF components
perform better than without, and that bidirectional LSTM works better than LSTM in
one direction.
Table 2. Comparison of ABSA Performance for Various Models
Models Aspect Extraction Sentiment Classification
P R F1 P R F1
LSTM+LR 59.35 65.74 62.38
LSTM+NB 46.36 52.75 49.34 58.48 65.40 61.75
LSTM+SVM 48.60 61.10 54.14
BI-LSTM+LR 71.08 83.95 76.98
BI-LSTM+NB 76.87 77.10 76.99 67.05 83.15 74.24
BI-LSTM+SVM 51.91 79.25 62.73
CRF+LR 69.79 83.99 76.23
CRF+NB 83.20 69.74 75.88 65.55 83.13 73.30
CRF+SVM 54.42 80.36 64.89
LSTM-CRF+LR 64.87 80.78 71.95
4
LSTM-CRF+NB 82.16 80.01 81.07 61.51 79.94 69.53
LSTM-CRF+SVM 53.66 77.66 63.47
BI-LSTM-CRF+LR 71.64 83.68 77.19
BI-LSTM-CRF+NB 83.55 80.24 81.86 67.85 82.92 74.63
BI-LSTM-CRF+SVM 56.95 80.30 66.64
4 Conclusion
This paper reviews and applies the bidirectional LSTM-CRF model in the product
reviews in Chinese and systematically compares the performances across multiple
state-of-the-art models that cast aspect term and sentiment word extraction to se-
quence labeling. Then, we use mainstream classifiers to gain the corresponding sen-
timent polarity of aspect terms. Experimental results show that the BI-LSTM-CRF
with LR outperforms other counterparts. As one future work, we will try to incorpo-
rate convolutional neural networks into the BI-LSTM model with some attention
mechanism [5] for CRF input to discover the implicit aspects.
Acknowledgments. This work is supported by Tianjin Social Science Planning Top-
ics TJXC15-002.
References
1. Lin C., He Y.: Joint sentiment/topic model for sentiment analysis. Proceedings of the 18th
ACM conference on Information and knowledge management. 375-384(2009).
2. Titov I., McDonald R.: Modeling online reviews with multi-grain topic models. Proceed-
ings of the 17th international conference on World Wide Web. 111-120(2008).
3. Lipenkova, J.: A system for fine-grained aspect-based sentiment analysis of Chinese. Pro-
ceedings of the 53rd Annual Meeting of the Association for Computational Linguistics.
55–60(2015).
4. Huang Z., Xu W., Yu K.: Bidirectional LSTM-CRF models for sequence tagging. arXiv
preprint arXiv:1508.01991(2015).
5. Bahdanau D., Cho K., Bengio Y.: Neural machine translation by jointly learning to align
and translate. arXiv preprint arXiv:1409.0473(2014).