Intelligent News Aggregator for German with Sentiment Analysis - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 11

1.2.5 Sentiment Analysis
Sentiment analysis aims to detect and assess opinionated text. Often the synonym
“opinion mining” is used to refer to the same task. Journalistic content such as news
articles is considered to be a reliable information source and may serve as a basis
for various natural processing (NLP) tasks. News articles are objective reports and
ideally, the authors do not express their attitudes. But this does not mean that news
articles cannot contain any opinionated text. Usually, news texts reflect opinions and
sentiments of newsworthy people, which are directed toward topics or toward other
entities. The identification of passages that echo subjective statements, attitudes, or
views and distinguishing them from the objective parts reporting facts is the first step
in many sentiment analysis approaches before analyzing the identified passages.
Exploiting quotations is one way to tackle the problem of mining opinions from
news articles. Quotations are a trustworthy form of mirroring what people or organi-
zations have said in an original and genuine manner. In addition, given a quotation, its
speaker is usually the opinion holder, the entity that expresses the opinion. We pro-
pose a supervised two-stage approach where we first identify subjective quotations
and, second, classify the subjective quotations in either positive or negative.All
remaining quotations are regarded as neutral (Sect. 1.4.3 ). We explore a range of
features established in sentiment analysis of other text genres (e.g., product reviews,
tweets) and examine to what extent they are suitable to separate subjective from
neutral and positive from negative quotations (Sect. 1.4.6 ).
1.3 Quotation Extraction
Quotations report what persons or organization have said. In news articles they are
often used to confirm claims made by the author and indicate the importance of the
transported information. Thus, they may be an important piece of news article for var-
ious text processing tasks such as sentiment analysis or news summarization. In this
section we present our approach to quotation extraction, which covers both the extrac-
tion of direct and indirect quotations as well as the assignment of a quotation speaker.
1.3.1 Introduction
Quotations repeat a speech, text, or statement expressed by a speaker and can be
distinguished into direct and reported speech. We refer to reported speech also as
indirect speech and to a speaker also as quotation holder in the following. Quotations
are composed of a reporting and a reported clause. Following Krestel et al. [ 24 ]
Fig. 1.2 shows an example of the structure of both, a direct and an indirect quotation.
The reporting clause introduces the quotation. Besides the quotation holder, it may