Sentiment annotations for reviews: an information quality perspective

DOIhttps://doi.org/10.1108/OIR-04-2017-0114
Pages579-594
Date10 September 2018
Published date10 September 2018
AuthorHeng-Li Yang,August F.Y. Chao
Subject MatterLibrary & information science,Information behaviour & retrieval,Collection building & management,Bibliometrics,Databases,Information & knowledge management,Information & communications technology,Internet,Records management & preservation,Document management
Sentiment annotations for
reviews: an information
quality perspective
Heng-Li Yang and August F.Y. Chao
Department of Management Information Systems,
National Cheng-Chi University, Taipei, Taiwan
Abstract
Purpose The purpose of this paper is to propose sentiment annotation at sentence level to reduce
information overloading while reading product/service reviews in the internet.
Design/methodology/approach The keyword-based sentiment analysis is applied for highlighting
review sentences. An experiment is conducted for demonstrating its effectiveness.
Findings A prototype is built for highlighting tourism review sentences in Chinese with positive or
negative sentiment polarity. An experiment results indicates that sentiment annotation can increase
information quality and users intention to read tourism reviews.
Research limitations/implications This study has made two major contributions: proposing the
approach of adding sentiment annotation at sentence level of review texts for assisting decision-making;
validating the relationships among the information quality constructs. However, in this study, sentiment
analysis was conducted on a limited corpus; future research may try a larger corpus. Besides, the annotation
system was built on the tourism data. Future studies might try to apply to other areas.
Practical implications If the proposed annotation systems become popular, both tourists and attraction
providers would obtain benefits. In this era of smart tourism, tourists could browse through the huge amount
of internet information more quickly. Attraction providers could understand what are the strengths and
weaknesses of their facilities more easily. The application of this sentiment analysis is possible for other
languages, especially for non-spaced languages.
Originality/value Facing large amounts of data, past researchers were engaged in automatically
constructing a compact yet meaningful abstraction of the texts. However, users have different positions and
purposes. This study proposes an alternative approach to add sentiment annotation at sentence level for
assisting users.
Keywords Information quality, Sentiment analysis, Tourism, Chinese review analysis,
Sentiment annotation
Paper type Research paper
1. Introduction
Since user-generated content (UGC) has become the mainstream paradigm of internet, there
are a number of personal travel experiences available online. Such data have become a
crucial reference for travelers while planning their vacations (Pudliner, 2007). These data are
different from the notes listed on public websites; instead these opinions are distributed
across Web 2.0 sites like blogs and social network sites. This valuable customer feedback is
also important to business because it generates electronic word-of-mouth (e-WOM)
promotion. Litvin et al. (2008) pointed out that if a tourist is recognized by his on-line peers
as experienced and reliable, his comments become valuable e-WOM and would have
significant impacts on purchase decisions of other travelers. However, there are some
problems for consumers while searching for these UGCs. A search conducted according to
designated keywords would return a number of results presented in a variety of formats
(Xiang and Gretzel, 2010). Some are in structured type, such as numeric ratings, but most of
others are in unstructured type, such as textual comments (Zhang et al., 2016). Further, some Online Information Review
Vol. 42 No. 5, 2018
pp. 579-594
© Emerald PublishingLimited
1468-4527
DOI 10.1108/OIR-04-2017-0114
Received 9 April 2017
Revised 13 July 2017
Accepted 9 April 2018
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/1468-4527.htm
The authors would like to thank the National Science Council, Taiwan, for financially supporting this
research under contract NSC 101-2410-H-004-015-MY3.
579
Sentiment
annotations
for reviews
online reviews are manipulative rather than authentic (Banerjee and Chua, 2014). Therefore,
reviewers have to spend huge amount of time to read each posted article thoroughly for
sifting the possible useful information. It would cost overwhelming mental efforts and might
not contribute to the decision-making.
For mitigating the burden of readers, this study suggests that review websites present
sentiment annotation for assisting the capture of useful information. In literature,
researchers (e.g. Huang et al., 1999; Chen and Huang, 2014) have put forward methods to
automatically construct a compact yet meaningful abstraction of the texts (e.g. news). Text
summary is helpful to give readers an overall picture of the document. However, because
users have different positions and purposes, some useful and fine granularity of information
might be missing if only providing abstraction. Thus, this study proposes adding sentiment
annotation at sentence level to assist users to browse through large amounts of data.
We further suggest an add-on annotation system for review websites to highlight review
sentences with positive or negative sentiments. The Chinese tourism review corpus was
used as an example and a keyword-based annotation prototype system was built.
In literature, some researchers (e.g. Zhang et al., 2011; Schmunk et al., 2014) have tried to
identify the positive and negative sentiment embedded in the tourist experiences, which
may be non-spaced languages (i.e. there is no space separated between words). However,
there is no such an annotation system. After completing the system, we also conducted an
experiment to demonstrate the feasibility and advantages of the proposed system.
We evaluated the proposed system using the perspective of information quality, and
suggested that the proposed system could improve information quality so as to facilitate
users understanding the reviews.
The research framework of this study is as shown in Figure 1. We first applied a review
sentiment analysis that contains three major steps: data preparation, word-level sentiment
analysis and sentence-level sentiment analysis. After analyzing texts, we built a prototype
to add sentiment annotations to each sentence and then conducted an information quality
experiment. The purpose is to demonstrate that our annotation approach could be accepted
by users and assist them browsing texts effectively.
REVIEW SENTIMENT ANALYSIS
DATA PREPARATION Search for
Synonym
Group in Chilin
NTUSD
Collected Texts from
IPEEN.com.tw
SINICA CKIP
Part-of-Speech
tools
Negation
Process
Texts with
Negation Notation
Calculating
Sentiment Polarity
Sentiment Polarity
Wordlist
Calculating
Sentence Sentiment
Sentiment Sentences
Build Sentiment
Annotation Prototype
Conduct Information
Quality Experiment
SENTENCE-LEVEL SENTIMENT ANALYSIS
Extended
NTUSD
WORD-LEVEL
SENTIMENT
ANALYSIS
Notes: NTUSD, National Taiwan University Sentiment Dictionary. SINICA (Academia Sinica) is
the most preeminent academic institution in Taiwan, and CKIP is its Chinese Knowledge and
Information Processing group. Chilin is a Chinese Synonym Forest. IPEEN.com.tw is a website
gathering user’s experiences of food tasting, traveling, etc. in Taiwan
Figure 1.
Research framework
580
OIR
42,5

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT