Intelligent News Aggregator for German with Sentiment Analysis - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 33

Table 1.4 Overview of the sentiment analysis features and values
Feature Name

Feature Value

Bow-of-Words (uni- and bigrams) Occurrence flag, tf, tf-idf, tf-delta-idf

Parts-of-Speech (uni- and bigrams) Occurrence flag, tf, tf-idf, tf-delta-idf

SentiWS words Occurrence flag, tf, tf x polarity-value

SentiWS pos/neg/all Occurrence, count, count x polarity-value

Valence shifters Occurrence flag, tf

Valence diminishers/intensifiers/shifters Occurrence, count

Discourse markers Occurrence flag, tf

1.4.5 Corpus

The corpus for evaluating our sentiment analysis approach consists of 851 quotations
extracted from a dataset of German news articles dated from February 23, 2012
to May 21, 2012. The manually annotated quotations of the quotation extraction
corpus described in Sect. 1.3.4 served as the basis for the sentiment corpus. We
asked four annotators to tag each quotation as either neutral, positive, negative,
or mixed and if possible to mark the opinion target. 25 In order to obtain a consistent
corpus we defined a set of annotation rules, which we describe in detail below. In
general, personal attitudes of the annotators, their moral perceptions, or political
views must not influence the tagging. The task was to determine the opinion of the
quotation speakers. If a quotation appeared to be incomplete, the annotators also had
the possibility to tag a quotation as don't know.

1.4.5.1 Neutral Quotations

A quotation should be regarded as neutral if the statement serves solely to transport
facts. Information, announcements, or intentions of the speaker without any personal
assessment by the speaker are considered to be neutral. It is not relevant whether the
fact itself is positive or negative from a moral, political, or any other point of view. In
the remainder of this section, we provide examples of different types of quotations.
The content of the quotation in Example 1.1 solely reports fact:

Example 1.1 Alisade berichtete, dass bald 500 weitere KFC-Filialen lan-
desweit eröffnen würden. (Alisade reported, that soon 500 more KFC-stores
would open countrywide.)

25 We do not use the annotated opinion targets yet, but describe them here for the sake of complete-
ness.