Book Review

Sentiment Analysis: Mining Opinions, Sentiments, and Emotions

Bing Liu
(University of Illinois at Chicago)

Cambridge University Press, 2015, 381 pp.; hardcover, ISBN 9781107017894, $80

Reviewed by
Jun Zhao, Kang Liu, and Liheng Xu
Institute of Automation, Chinese Academy of Sciences

With the increasing development of Web 2.0, such as social media and online businesses,
the need for perception of opinions, attitudes, and emotions grows rapidly. Sentiment
analysis, the topic studying such subjective feelings expressed in text, has attracted
significant attention from both the research community and industry. Although we
have known sentiment analysis as a task of mining opinions expressed in text and
analyzing the entailed sentiments and emotions, so far the task is still vaguely defined
in the research literature because it involves many overlapping concepts and sub-tasks.
Because this is an important area of scientific research, the field needs to clear this
vagueness and define various directions and aspects in detail, especially for students,
scholars, and developers new to the field. In fact, the field includes numerous natural
language processing tasks with different aims (such as sentiment classification, opinion
information extraction, opinion summarization, sentiment retrieval, etc.) and these have
multiple solution paths. Bing Liu has done a great job in this book in providing a
thorough exploration and an anatomy of the sentiment analysis problem and conveyed
a wealth of knowledge about different aspects of the field.
Liu is a leading figure in this research area. Not only has he made important
contributions to the understanding of opinions and sentiments expressed in text, but
he also has significantly influenced the design of real-life sentiment analysis algorithms
and the building of practical sentiment analysis systems. This book has at least three
significant merits and meets the needs of different types of readers.

1. First, it is praiseworthy that the book gives detailed definitions of opinions

and probably all important pertinent concepts, including sentiment,
opinion target, time of opinion, opinion holder, opinion reason, opinion
qualifier, and the often-neglected opinion types such as comparative
opinions, fact-implied opinions, and so forth. The book explains and
illustrates these concepts clearly, which facilitates a comprehensive and
principled understanding of the sentiment analysis problem. In our
opinion, the definitions in this book for sentiment analysis are more
complete and profound than any other publication that we have seen.
2. This book not only presents the main sub-tasks of sentiment analysis,
such as sentiment classification at different discourse levels, opinion
summarization, opinion search, and emotion identification, but also covers

doi:10.1162/COLI r 00259

© 2016 Association for Computational Linguistics

Computational Linguistics Volume 42, Number 3

many emerging sentiment-related topics, such as sentiment analysis of

debates and discussions, mining of intentions, and deceptive opinion
detection. This shows that Liu has extensive research experience as well as
knowledge of industry applications and knows the importance of these
emerging tasks, which are in demand from the industry. Furthermore, this
book includes the latest methods and technologies for each related task
published in the mainstream conference proceedings and journals. What is
especially valuable is that it describes first-hand experience in building a
sentiment analysis system. Liu thus brings to readers a global view of the
whole area from research to practice, and meets the needs of different
readerships, from experienced scholars in natural language processing to
graduate students and industry practitioners interested in sentiment
analysis and social media analysis. This book thus serves as a
comprehensive yet in-depth survey of references in sentiment analysis.
3. It is also commendable that the book gives a balanced treatment of both
linguistic approaches and machine learning approaches to solving the
sentiment analysis problem. In recent years, the mainstream methods
in the research literature have been based on supervised learning with
elaborate feature engineering. State-of-the-art techniques also use deep
learning models to learn effective features directly from raw
data as an alternative to manual feature engineering. However, as
mentioned in this book, supervised learning methods provide no
linguistic interpretations and do not generate knowledge for linguists
or industry developers to gain insights of the problem. Practically,
when errors occur in an application, it is hard to know what is wrong
and how to fix it. Fortunately, this book provides a comprehensive
list of linguistic constructs and perspectives that are instrumental for
sentiment analysis, which make up for the deficiency of black-box
approaches using pure machine learning. Moreover, it also lists and
elaborates on many specific linguistic phenomena that are critical
for effective classification of sentiment such as negation (Chapter 5),
modality (Chapter 5), and comparison (Chapter 8). We believe that
this book will enable the reader to gain not just a comprehensive
understanding of the computation methods but also deep linguistic
insights of the sentiment analysis problem and its possible solutions.

The book is organized into 13 chapters. The first two chapters introduce the basics
and define the sentiment analysis problem. Chapters 3–9 discuss the core sentiment
analysis tasks (e.g., sentiment classification, aspect analysis, and opinion summariza-
tion) and their current solution methods. Chapters 10–13 investigate the emerging
themes from recent research and applications (e.g., analysis of debates, intentions, fake
opinions, and review quality).
Specifically, Chapter 1 motivates and gives an overview of the whole book. It
describes the expression of sentiment as one of the most important and complicated
phenomena of human language. The goal of sentiment analysis is to computationally
extract sentiments, opinions, and emotions expressed in text, which is different from the
goal of traditional linguistic studies aimed at understanding the human language. Tech-
nically, analysis of sentiment can be divided into several levels according to different

596
 Book Review

discourse granularity, such as document, sentence, and aspect or sentiment-and-target

levels. Liu took a structured approach to write this book.
Chapter 2 gives the definition of sentiment analysis, along with discussions of many
key concepts such as subjectivity, affect, emotion, and mood. An opinion can either be
defined as a quadruple or quintuple, and sentiment involves the type, orientation, and
intensity about an opinion. Many succinct examples are also given to make the concepts
easily understandable.
Chapter 3 discusses document-level sentiment classification, which is probably the
most studied problem in sentiment analysis. It usually involves direct applications of
(supervised or unsupervised) machine learning algorithms. Although the problem is
called sentiment classification, it could also be taken as a regression problem. Most
sentiment classification approaches categorize documents into only positive, negative,
and objective types, whereas emotion classification involves more categories with over-
lapping meanings and is more difficult to perform accurately.
Document-level sentiment classification is, however, too coarse for practical appli-
cations. More elaborate analysis is applied to sentence and aspect levels, as introduced
in Chapters 4 and 5. These two problems are the most practically useful research topics
of sentiment analysis. In particular, aspect-level analysis forms the core of applications
of sentiment analysis as it aims to identify the atomic unit of information contained
in sentiment, opinion, and emotion expressions, which is the pair of sentiment and its
target. For sentiment identification, the book shows that beyond positive and negative
orientations of sentiment expressions, many sophisticated language phenomena also
need to be considered in analysis (e.g., conditional expressions, sarcasm, sentiment
composition, and negation). To deal with such sophisticated language phenomena,
term-level features are no longer sufficient. One has to incorporate deeper knowledge
of syntax, semantics, and discourse. Although significant efforts have been made, the
problem is still far from being solved.
Chapters 6 and 7 describe aspect extraction and sentiment lexicon generation,
respectively. The two topics are inter-related because sentiment lexicons are often used
to express opinions on aspects or targets. Therefore, many researchers have studied the
two tasks together. Furthermore, aspects and sentiment expressions also have common
characteristics, such as, for example, the fact that they are both domain-specific.
Although the two problems are challenging, these chapters provide valuable resources
and practical algorithms for their solutions.
Chapter 8 studies comparative expressions related to sentiment. The task is also
challenging because of the flexible usage of comparatives and the difficulty of identify-
ing the preferred entity set. Limited research attention has been paid to this problem,
but it is very useful in applications.
Chapter 9 provides several approaches to generating opinion summaries. Opin-
ion summarization is quite different from conventional single-document or multi-
document summarization, because it is centered on opinion targets and produces
quantitative sentiment ratings for targets. The output of opinion summarization can be
interactively visualized, which gives really interesting and easily understandable results
that can be seen by users in a single glance.
Chapters 10–13 introduce several related topics to sentiment analysis, namely, the
study of debates, intentions, deceptive opinions, and quality of reviews. These are
emerging themes from recent research. We believe there will be significant research
activities and applications on these topics in the years to come.
As a whole, this book serves as a useful introduction to sentiment analysis along
with in-depth discussions of linguistic phenomena related to sentiments, opinions, and

597
Computational Linguistics Volume 42, Number 3

emotions. Although many sentiment analysis methods are based on machine learning
as in other NLP tasks, sentiment analysis is much more than just a classification
or regression problem, because the natural language constructs used to express
opinions, sentiments, and emotions are highly sophisticated, including sentiment shift,
implicated expression, sarcasm, and so on. Liu has described these issues and problems
very clearly. Readers will find this book to be inspiring and it will arouse their interests
in sentiment analysis.

Jun Zhao is a professor in the National Laboratory of Pattern Recognition, Institute of Automation,
Chinese Academy of Sciences. His primary research focus is information extraction and question
answering. Zhao’s e-mail address is [email protected]. Kang Liu is an associate professor
in the National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy
of Sciences. His primary research focus is opinion mining, information extraction, and machine
learning. Liu’s e-mail address is [email protected]. Liheng Xu received a Ph.D. degree from
the National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of
Sciences. His primary research focus is opinion mining and deep learning. Xu’s e-mail address is
[email protected].

598