Proceedings of the 2012 IEEE ISMOT

Sentiment Analysis: A Literature Review

ZHU Nanli1,3, ZOU Ping1,2, LI Weiguo3, CHENG Meng4

1
Faculty of Management and Economics, Kunming University of Science and Technology, Kunming, China
2
Science and Technology Division of Yunnan Education Office, Kunming, China
3
School of Electron and Information Engineering, Ningbo University of Technology, Ningbo, China
4
Faculty of Chinese and Bilingual Studies, The Hong Kong Polytechnic University, Hong Kong, China
([email protected])

opinions on products, policies, people and events, etc.

Abstract - Sentiment analysis has aroused the interest of Sentiment analysis has grown into a hot research field in
many researchers in recent years, since subjective texts are natural language processing.
useful for many applications. In particular, sentiment analy- Sentiment analysis first created a sensation at Text
sis on online reviews has become a hot research field. Studies Retrieval Conference (TREC), and since 2006 related
on sentiment analysis mainly focus on framework and lexi-
evaluation tasks have appeared every year. The annual
con construction, feature extraction, and polarity determina-
tion. This paper presents a survey on the latest development multi-lingual opinion analysis task (MOAT) also started
in sentiment analysis, and makes an in-depth introduction of in 2006, including standard libraries in three languages,
its research and application in business and Blogsphere. The Chinese, English and Japanese. However, the research in
methods used in current research are especially emphasized China, particularly the sentiment analysis on Chinese has
and the existing problems of those studies are discussed. barely taken off. Initiated in 2008, Chinese opinion analy-
Finally, some possible future directions of research are sis evaluation (COAE) is the first sentiment analysis of
pointed out. the kind.
Keywords - Sentiment analysis, Information extraction
A. Research on Framework

Sentiment analysis research in China has seen a

I. INTRODUCTION steady growth since 2008. YAO Tian-fang et al. [2], and
Sentiment analysis, or opinion mining, aims at user’s HOU Feng et al. [3] approached from the perspective of
attitude and opinions by investigating, analyzing and ex- opinion mining, WANG Hui et al. [4], Ren Hongjuan and
tracting subjective texts involving users’ opinions, prefer- Zhang Zhiqiang [5] took the method of viewpoint mining,
ences and sentiment. Since Bo Pang [1] put forth this con- ZHOU Li-zhu et al. [6], Li Gang et al. [7], ZHAO Yan-
cept in 2002, the academics have undertaken a diverse Yan et al. [8], and WEI Wei et al. [9], however, adopted
range of related research, due to its practicality in opinion the sentiment analysis method. Li Xiaojun et al. [10] fol-
monitoring and business competitive intelligence. Senti- lowed the route of the sentiment polarity. All the above-
ment analysis on online reviews has become increasingly mentioned research makes distinctive summary and ex-
popular. A multidisciplinary research field in nature, sen- ploration of the language and technology support in this
timent analysis includes multiple fields such as natural field.
language processing (NLP), computational linguistics, Kaiser et al. [11] proposed a warning system for
information retrieval, machine learning and artificial intel- online market research which allows the identification of
ligence etc. critical situations in online opinion formation. Hasan and
As an astronomical quantity of sentimental subjective Adjeroh [12] proposed three proximity-based features,
texts appear on Internet, researchers put more emphasis namely, proximity distribution, mutual information be-
on complex sentimental sentences and texts instead of on tween proximity types, and proximity patterns. The exper-
words only. In light of text granularity, sentiment analysis iment results showed that proximity-based sentiment
is conducted on distinct research levels—word, phrase, analysis is able to extract sentiments from a specific do-
sentence, text and multi-text. main.
This paper presents an overview and the prospects of
the several major research fields in sentiment analysis. B. Lexicon Construction

As tagged Chinese opinion subject text facilitates

II. FRAMEWORK AND LEXICON CONSTRUCTION technologies, e.g. machine learning, to realize classifica-
tion, topic extraction, sentiment analysis and subject-
Natural human language carries two types of infor- sentiment relation recognition, so lexicon/corpus con-
mation: objective information about facts and critical in- struction is generally viewed as a prerequisite for senti-
formation with human subjective sentiment. The rapid ment analysis.
development of forum, BBS, Blog, and review websites Bai [13] proposed a heuristic search-enhanced Mar-
contributes to the exploding amount of such critical in- kov blanket model that is able to capture the dependencies
formation, which reflects users’ attitudes, viewpoints and among words and provide a vocabulary that is adequate

978-1-4673-4593-4/12/$31.00 ©2012 IEEE 572

 Proceedings of the 2012 IEEE ISMOT

for the purpose of extracting sentiments. Kanayama and and its perspectives are field theory in physics and
Nasukawa [14] introduced clause-level evaluation detec- multilinear algebra in mathematics. The tensor representa-
tion, which is a fine-grained type of opinion mining, and tions of documents and queries in TFM were constructed,
describes an unsupervised lexicon building method for presenting some key concepts such as term field, tensor
capturing domain-specific knowledge by leveraging the product of term array and term field constant.
similar polarities of sentiments between adjacent clauses.
Tan and Wu [15] conceived a random walk algorithm to
construct domain-oriented sentiment lexicon by simulta- IV. SENTIMENT POLARITY ANALYSIS
neously utilizing sentiment words and documents from
both old domain and target domain (or "new domain"). Sentiment polarity analysis prefers to the analysis and
For a specific domain, the sentiment resource can be re- judgment of the sentiment expressed in a subjective text,
garded as a lexicon, which contains a list of product fea- which is capable of distinguishing positive, negative, or
ture words and opinion expressions pair with polarity tag. even more subtle sentiment, such as happiness, anger,
To construct such a sentiment resource, Li et al. [16] de- grief or joy. It largely involves two tasks: the dichotomy
signed an automatic algorithm to extract opinion expres- of subjective/objective information, and the sentiment
sions and determine its polarity to existing feature words. classification of subjective information, including the
commonplace positive/negative dichotomy as well as
more exquisite n-multiple taxonomy [24].
III. FEATURE EXTRACTION There are some focuses of research on Chinese:
(1) in the granularity of words, Du Weifu et al. [25]
A. Topic Extraction proposed a novel scalable word semantic orientation
computing framework, in which the word semantic orien-
FAN Na et al. [17] noted a method of extracting Chi- tation computing is transformed into the function optimi-
nese sentiment topic sentences. Then, FAN Na et al. [18] zation. As an instance of the proposed framework, the
proposed a method based on mixture model for analyzing authors built an undirected graph in the use of word simi-
subject and sentiment in texts. Firstly texts in corpus were larity computing technology first, and then partition the
labeled as positive or negative and their subjects were word-to-word graph by the idea of ˃minimum-cut˃,
labeled too. The method is general and can be applied to thereby function optimization was adopted in this word
any text collections. semantic orientation computing framework and resolved
by using simulated annealing algorithm. A method of
B. Object Extraction collocation orientation identification based on hybrid lan-
guage information was proposed by WANG Suge and
One of the fundamental tasks of sentiment analysis YANG Anna [26]. Firstly, according to the characteristics
is to extract the object from review sentences, in order to of six kinds of collocation patterns, the probability latent
identify all the features of the events in a certain text. semantic models were determined for them. Then the ob-
Zhang and Liu [19] employed a learning-based meth- tained semantic models were used to identify the senti-
od called Bayesian Sets to extract opinions expressed on ment orientations of collocations. Lastly, for some collo-
entities and their attributes. To improve the performance cations containing a sentiment word, their previous tags
of opinion targets extraction, XU Bing et al. [20] pro- were modified by using some constructed rules. The ex-
posed to integrate shallow parsing features and heuristic periment result in the corpus of car reviews indicated that
position information for modeling of the training process the proposed method is superior to the method based only
without introducing domain lexicon. HUANG Yi-hua et on probability latent semantic model or rule for colloca-
al. [21] constructed a phrase tree structure to present the tion orientation identification. The technology of the Chi-
appraisal expression pattern, as well as a method based on nese word sentiment polarity judgment was discussed and
approximate convolution tree kernels to calculate the sim- analyzed by ZHANG Jing and JIN Hao [27]. The polarity
ilarity between these structures. was described by using the sentiment characteristics set.
The model of the sentiment polarity mutual information
C. Relation Extraction characteristics was created based on the bi-gram depend-
ency of POS tagging. The classifier was available by ma-
ZHANG Jian-feng et al. [22] presented a novel chine learning to automatically judge, compare and opti-
method to extract the pairs of opinion-bearing terms and mize the word sentiment polarity. All of these helped to
opinion targets as the candidate set, and then employed improve the properties, the highest accuracy of SVM
the maximum entropy model to combine lexical, part of reaches 95.47%, and the F value is up to 93.90%.
speech, semantic and positional features derived from text. (2) In the granularity of sentences, WANG Gen, and
The method incorporated relation extraction into opinion ZHAO Jun [28] proposed a new method called Multi-
mining and solves the problem of coreference and omit- redundant-labeled CRFs and applies it on sentence senti-
ting of opinion targets to some extent. ment analysis. This method can not only solve ordinal
Qiao et al. [23] noted that keywords are parallel to regression problems effectively, but also obtain global
each other. "Tensor Field Model" (TFM) was proposed, optimal result over multiple cascaded subtasks by merg-

978-1-4673-4593-4/12/$31.00 ©2012 IEEE 573

 Proceedings of the 2012 IEEE ISMOT

ing subjective/objective classification, polarity classifica- el regularized topic modeling framework based on the
tion and sentimental strength rating into an integrated classical topic modeling through the analysis of the distri-
model, with each subtask maintaining its own feature bution of opinion words for different product features in
types. YAO Tian-fang and LOU De-cheng [29] presented reviews and also the assumption of topic dependency of
how to identify the topics as well as the relations between opinion words. In the new framework, they took into con-
the topics and the sentimental descriptive terms in a Chi- sideration the opinionated information by defining a
nese sentence, and how to compute the sentiment polarity regularizer based on the similarity in opinion word usage
of the topics. They extracted the topics and their attributes of different reviews. SHI Wei and WANG Hongwei [38]
from a sentence with the help of a domain ontology, then adopted a method that information quality framework is
identify the relations between the topics and sentimental used to evaluate the quality of the reviews. The study
descriptive terms based on parsing results, and finally treated the evaluation of review quality as a classification
determine the polarity of each topic in the sentence. problem, advanced the review features based on quality,
classified the reviews quality by using classification mod-
el, and got the reviews ranking by calculating reviews
V. BUSINESS APPLICATION quality and query-relevance scores, then the result was
returned to users. To enlarge the size of document,
With the development of Internet application in the WANG Bo [39] introduced reviews from other domains
last decades, more clients have started to browse the huge and evaluated system performance with them based on
quantity of online reviews to know about the common PMI and SI techniques. Mining product features and po-
opinions of products and services so as to improve deci- larity of customers’ opinion from Chinese online custom-
sion-making. Meanwhile, online client reviews serve as a er reviews helps manufactures and service providers im-
feedback mechanism to help producers and marketers to prove their products and services, and gain competitive
promote their products and services, and increase compet- leverages. LI Shi et al. [40] proposed an approach based
itiveness. As a newly emerging research area, online re- on Apriori algorithm which uses unsupervised feature
view sentiment analysis has attracted great attention from extraction algorithm and supervised sentiment analysis
a wide range of related disciplines such as computer sci- technology to realize the summery information of cus-
ence, economics, and management. Zhang Ziqiong [30] tomer reviews. And then ranked the work features and
conducted a summary and analysis of the present situation opinions based on customer attention strength.
and development of online reviews.
Kang et al. [31] proposed a new senti-lexicon for the
sentiment analysis of restaurant reviews. In order to miti- VI. BLOG-BASED SENTIMENT ANALYSIS
gate such problem, an improved Naive Bayes algorithm is
applied and a unigrams + bigrams was used as the feature. The robust growth of Web2.0 technology advances
Benton et al. [32] presented an approach that is able the prevailing idea of “User Generate Content” in internet
to collect a corpus of medical message board posts, de- era. Most of the website users are attracted by the blogs
identify the corpus, and extract information on potential and microblogs. However, this form of commentaries is
adverse drug effects discussed by users. Using a corpus of characterized as bulky, unformatted and full of spams. To
posts to breast cancer message boards, they identified facilitate a quick and effective analysis of the texts, the
drug event pairs using co-occurrence statistics. Sun et al. blog-based sentiment analysis has caught on rapidly [41].
[33] investigated the distinctive characteristics of custom- In the scope of blog-specific sentiment analysis, fea-
er opinions found on internet forums. A comparison of the ture extraction is an essential aspect of research. Previous
similarities and differences between customer reviews and research on feature-based opinion mining has not had
forum feedbacks was given. Finally, data mining tech- good results due to drawbacks, such as selecting a feature
niques based on concept-based representations were used considering only syntactical grammar information or
to analyze feedback from forums. Chen et al. [34] treated treating features with similar meanings as different. To
the evaluation of review quality as a classification prob- solve these problems, Jeong et al. [42] posed an enhanced
lem and employed an effective information quality feature extraction and refinement method called FEROM
framework to extract representative review features. Park that effectively extracts correct features from review data
and Lee [35] presented a framework for extracting cus- by exploiting both grammatical properties and semantic
tomer opinions from websites and transforming them into characteristics of feature words and refines the features by
product specification data. recognizing and merging similar ones. Fan and Chang
LI Shi [36] focused on product features mining from [43] addressed the concept of Blogger-Centric Contextual
reviews of Chinese network customers and proposes a Advertising, which refers to the assignment of personal
method based on Apriori algorithm which is an unsuper- ads to any blog page, chosen in accord with bloggers’
vised mining method. It extracted the candidate features interests. Traditional keyword extraction methods are
collection by Apriori algorithm, and took redundancy usually designed for formal documents such as news arti-
pruning and compactness pruning algorithms. To solve cles or scientific papers. Messages posted by microblog
the implicit product feature extraction task in product users, however, are usually noisy and full of new words,
opinion mining, QIU Guang et al. [37] constructed a nov- which is a challenge for keyword extraction. Liu et al.

978-1-4673-4593-4/12/$31.00 ©2012 IEEE 574

 Proceedings of the 2012 IEEE ISMOT

[44] combine a translation-based method with a frequen- information contained in blogs relates back to financial
cy-based method for keyword extraction to mine user measures, such as stock price is not established, but
interests via keyword extraction from microblogs. Xu et al. should be investigated.
[45] proposed a novel graphical model to extract and vis-
ualize comparative relations between products from cus-
tomer reviews, with the interdependencies among rela- ACKNOWLEDGMENT
tions taken into consideration, to help enterprises discover
potential risks and further design new products and mar- This paper is funded by the National Natural Science
keting strategies. Foundation —“Research on the evolutional mechanism of
enterprise organizational capability system based on or-
ganizational process” (No. 70961003) as a stage research
VII. CONCLUSION achievement.
Project supported by the Natural Science Foundation
This paper conducts an overall survey of the three of Zhejiang Province (Grant No. Y1100253).
major research fields in sentiment analysis: framework, Nanli Zhu thanks the special support to our research
feature extraction and sentiment analysis, making a sum- from the Educational Science Project of Ningbo City
mary and analysis of the present development, and giving (YGH034), and the Project of Association of Higher Edu-
a detailed introduction of its application in business and cation of Zhejiang Province (KT2011182).
Blogs. Despite the current immaturity of related research,
sentiment analysis of online review has taken its position
as an emerging research frontline, which takes advantage REFERENCES
of the achievements in many areas, such as text mining,
natural language processing, web mining, and machine [1] Pang B, Lee L. A sentimental education: Sentiment analysis
learning. But the related research did not take place until using subjectivity summarization based on minimum cuts.
recently, and semantic parsing and understanding exhibit In: Scott D, ed. Proc. of the ACL 2004. Morristown: ACL,
high complexity, the overall research in this field being in 2004, pp. 271−278.
its infantry. Still a lot of problems need further explora- [2] YAO Tian-fang, CHENG Xi-wen, XU Fei-yu, Hans
tion and solution as follows: USZKOREIT , WANG Rui. A Survey of Opinion Mining
for Texts[J]. Journal of Chinese Information Processing,
(1) Insufficient empirical language data and platform. 2008,(03) , pp. 71-80.
So far no experimental platform has been released for [3] HOU Feng, WANG Chuan-ting, LI Guo-hui. Survey on the
public use and cast widespread influence; experimental Opinion Mining, Summarization and Retrieval[J]. Comput-
public corpora, especially marked corpora are relatively er Science, 2009,(07) , pp. 15-19+51.
sparse. The available corpora for English sentiment analy- [4] WANG Hui, WANG Hui-yu, ZUO Wan-li. Survey on opin-
sis are MPQA news corpus and the movie critic corpus by ion mining[J].Application Research of Computers,
Pang. They are confined with limited arrange of tasks and 2009,(01) , pp. 25-29.
under-satisfied corpus quality. For lack of an open exper- [5] Ren Hongjuan, Zhang Zhiqiang.Towards Advances of
iment platform and standardized benchmark, it is difficult Opinion Mining Based Bibliometrics Methods——a Poten-
tial Research Specialty of Information Science in the Fu-
to assess the effectiveness of all the methods. ture[J]. Document,Information & Knowledge,2010,(02) , pp.
(2) No breakthrough in textual sentiment analysis. 55-63.
Deep sentiment analysis inevitably involves semantic [6] ZHOU Li-zhu, HE Yu-kai, WANG Jian-yong. Survey on
parsing, and sentiment transfers in texts moreover, so research of sentiment analysis[J]. journal of Computer Ap-
there is not much progress in deep-structure semantic sen- plications,2008,(11) , pp. 2725-2728.
timent analysis, and in textual sentiment analysis. [7] Li Gang,Cheng Yangyang ,Kou Guangzeng. Key Problems
(3) No research on the commercial value of online of Sentence Level Sentiment Analysis[J]. LIBRARY AND
product reviews. Online reviews cast a profound influence INFORMATION SERVICE,2010,(11) , pp. 104-107+127.
on the purchase behavior of consumers, but related re- [8] ZHAO Yan-Yan, QIN Bing, LIU Ting. Sentiment Analy-
sis[J]. Journal of Software, 2010, (08) , pp. 1834-1848.
search in China is still at the onset stage, only tentative [9] WEI Wei,XIANG Yang,CHEN Qian. Survey on Chinese
attempts by some domestic researchers using simple sta- text sentiment analysis[J]. journal of Computer Applica-
tistical economy model such as linear regression model. tions, 2011,(12) , pp. 3321-3323.
However the lack of control over other sales influential [10] LI Xiaojun , DAI Lin , SHI Hanxiao , HUANG Qi. Survey
factors besides online reviews leads to failure of drawing on sentiment orientation analysis of texts[J]. Journal of
consistent conclusions and regular patterns. Different al- Zhejiang University(Engineering Science) ,2011, (07) , pp.
gorithm for customers’ commentaries and discrepant used 1167-1174+1186.
data also give rise to differences in conclusion. As a result, [11] Kaiser, Carolin; Schlick, Sabine; Bodendorf, Freimut.
further research into the mechanism in which online re- Warning system for online market research - Identifying
critical situations in online opinion formation[J].
views affect product sales is urgent. Other problems con- Knowledge-Based Systems, v 24, n 6, August 2011, pp.
cern the supporting system of new product development 824-836.
(NPD), and how producers more efficiently gather, pro- [12] Hasan, S.M. Shamimul; Adjeroh, Donald A. Detecting
cess and exploit online reviews. Finally, it is not clear if human sentiment from text using a proximity-based ap-

978-1-4673-4593-4/12/$31.00 ©2012 IEEE 575

 Proceedings of the 2012 IEEE ISMOT

proach[J]. Journal of Digital Information Management, v 9, [30] ZHANG Zi-qiong , YE Qiang , LI Yi-jun. Literature review
n 5, October 2011, pp. 206-212. on sentiment analysis of online product reviews[J]. Journal
[13] Bai, Xue. Predicting consumer sentiments from online of Management Sciences in China,2010,(06) , pp. 84-96.
text[J]. Decision Support Systems, v 50, n 4, March 2011, [31] Kang, Hanhoon ; Yoo, Seong Joon; Han, Dongil. Senti-
pp. 732-742. lexicon and improved Naïve Bayes algorithms for senti-
[14] Kanayama, Hiroshi; Nasukawa, Tetsuya. Unsupervised ment analysis of restaurant reviews[J]. Expert Systems with
lexicon induction for clause-level detection of evalua- Applications, v 39, n 5, April 2012, pp. 6000-6010.
tions[J]. Natural Language Engineering, v 18, n 1, January [32] Benton, Adrian; Ungar, Lyle; Hill, Shawndra; Hennessy,
2012, pp. 83-107. Sean; Mao, Jun; Chung, Annie; Leonard, Charles E.;
[15] Tan, Songbo; Wu, Qiong. A random walk algorithm for Holmes, John H. Identifying potential adverse effects using
automatic construction of domain-oriented sentiment lexi- the web: A new approach to medical hypothesis genera-
con[J]. Expert Systems with Applications, v 38, n 10, Sep- tion[J] . Journal of Biomedical Informatics, v 44, n 6, De-
tember 15, 2011, pp. 12094-12100. cember 2011, pp. 989-996.
[16] Li, Zhichao; Yin, Hang; Zhang, Min; Ma, Shaoping. Ex- [33] Sun, Jie; Loh, Han Tong; Yeo, Aik Siang; Liu, Ying. Opin-
traction and polarity determination for opinion expres- ion comparison between internet forums and customer re-
sion[J]. Journal of Computational Information Systems, v 7, views[J]. International Journal of Computer Applications in
n 16, December 2011, pp. 5776-5784. Technology, v 40, n 1-2 SPEC.ISSUE, February 2011, pp.
[17] FAN Na, CAI Wan-dong, ZHAO Yu, LI Hui-xian. Extrac- 107-113.
tion of sentiment topic sentences of Chinese texts[J]. jour- [34] Chen, Chien Chin; Tseng, You-De. Quality evaluation of
nal of Computer Applications, 2009,(04) , pp. 1171- product reviews using an information quality framework[J].
1173+1176. Decision Support Systems, v 50, n 4, March 2011, pp. 755-
[18] Fan Na, Cai Wandong ,Zhao Yu. Mixture model-based 768.
subject-sentiment analysis in texts[J]. JOURNAL OF [35] Park, Yongtae; Lee, Sungjoo. How to design and utilize
HUAZHONG UNIVERSITY OF SCIENCE AND TECH- online customer center to support new product concept gen-
NOLOGY.NATURE SCIENCE,2010,(01) , pp. 31-34. eration[J]. Expert Systems with Applications, v 38, n 8,
[19] Zhang, Lei; Liu, Bing. Entity set expansion in opinion doc- August 2011, p 10638-10647.
uments[C]. HT 2011 - Proceedings of the 22nd ACM Con- [36] LI Shi. Research on Pruning Algorithm of Product Feature
ference on Hypertext and Hypermedia, 2011, pp. 281-289. Mining in Chinese Review[D]. Computer Engineering,
[20] XU Bing,ZHAO Tie-Jun,WANG Shan-Yu, ZHENG De- 2009.
Quan. Extraction of Opinion Targets Based on Shallow [37] QIU Guang, ZHENG Miao, ZHANG Hui, ZHU Jian-ke,
Parsing Features[J]. Acta Automatica Sinica, 2011,(10) , pp. BU Jia-jun, CHEN Chun, HANG Hang. Implicit product
1241-1247. feature extraction through regularized topic modeling[J].
[21] HUANG Yi-hua,PU Xiao-jia,YUAN Chun-feng,WU Gang- Journal of Zhejiang University(Engineering Science),
shan. Appraisal expression extraction based on parse tree 2011,(02) , pp. 288-294.
structure[J]. Application Research of Computers, 2011,(09) , [38] SHI Wei, WANG Hongwei. Study on Retrieval System of
pp. 3229-3234. On-line Product Reviews[J]. Journal of Information,
[22] ZHANG Jian-feng,ZHANG Qi, WU Li-de, HUANG Xuan- 2011,(09) , pp. 149-154.
jing. Subjective Relation Extraction in Chinese Opinion [39] WANG Bo. Sentiment Analysis for Fund Reviews Based
Mining[J]. Journal of Chinese Information Processing, on Cross Domain Knowledge[J]. Journal of Information,
2008,(02) , pp. 55-59+86. 2011,(02) , pp. 44-47.
[23] Qiao, Ya-Nan; Yong, Qi; Di, Hou. Tensor Field Model for [41] O'Leary, Daniel E. Blog mining-review and exten-sions:
higher-order information retrieval[J]. Journal of Systems "From each according to his opinion"[J]. Deci-sion Support
and Software, v 84, n 12, December 2011, pp. 2303-2313. Systems, v 51, n 4, November 2011, pp. 821-830.
[24] XU Lin-hong, LIN Hong-fei, ZHAO Jing. Construction and [40] LI Shi, YE Qiang, LI Yi-jun, LUO Si-qing. Mining product
Analysis of Emotional Corpus[J]. Journal of Chinese In- features and sentiment orientation from Chinese customer
formation Processing, 2008,(01) , pp. 116-122. reviews[J]. Application Research of Computers,2010,(08) ,
[25] Du Weifu, Tan Songbo, Yun Xiaochun, and Cheng Xueqi. pp. 3016-3019.
A New Method to Compute Semantic Orientation[J]. Jour- [42] Jeong, Hana; Shin, Dongwook; Choi, Joongmin. FEROM:
nal of Computer Research and Development,2009,(10) , pp. Feature extraction and refinement for opinion mining[J].
1713-1720. ETRI Journal, v 33, n 5, October 2011, pp. 720-730.
[26] WANG Suge , YANG Anna. A Method of Collocation [43] Fan, Teng-Kai; Chang, Chia-Hui. Blogger-centric contextu-
Orientation Identification Based on Hybrid Language In- al advertising[J]. Expert Systems with Ap-plications, v 38,
formation[J]. Journal of Chinese Information Pro- n 3, March 2011, pp. 1777-1788.
cessing,2010,(03) , pp. 69-74. [44] Liu, Zhiyuan; Chen, Xinxiong; Sun, Maosong. Min-ing the
[27] ZHANG Jing, JIN Hao. Study on Chinese Word Sentiment interests of Chinese microbloggers via key-word extrac-
Polarity Automatic Estimation[J]. Computer Engineering, tion[J]. Frontiers of Computer Science in China, v 6, n 1,
2010,(23) , pp. 194-196. February 2012, pp. 76-87.
[28] WANG Gen, ZHAO Jun. Sentence Sentiment Analysis [45] Xu, Kaiquan; Liao, Stephen Shaoyi; Li, Jiexun; Song,
Based on Multi-redundant-labeled CRFs[J]. Journal of Chi- Yuxia. Mining comparative opinions from customer re-
nese Information Processing, 2007,(05) , pp. 51-55+86. views for Competitive Intelligence[J]. Decision Support
[29] YAO Tian-fang, LOU De-cheng. Research on Semantic Systems, v 50, n 4, March 2011, pp. 743-754.
Orientation Analysis for Topics in Chinese Sentences[J].
Journal of Chinese Information Processing,2007,(05) , pp.
73-79.

978-1-4673-4593-4/12/$31.00 ©2012 IEEE 576