controversial news issues draw much interest from the public. But it
is not simple for an ordinary user to search and contrast the
opposing arguments and have complete understanding of issues.
Disputant relation based method classifies the opposing views of the
news/issues which can help readers to easily understand the issue.
For classifying news articles on contentious issues disputant
relation-based method is used. It is known that the disputants of a
contention are an important feature for understanding the
conversation so, the disputant relation based method performs
unsupervised classification on news articles based on disputant
relations, and helps readers naturally view the articles through the
opponent- based frame and attain balanced understanding, free from a
specific biased viewpoint.
method is performed in three stages: disputant mining, disputant
separation and article classification. Also a modified version of
HITS algorithm is used in disputant partitioning and an SVM
classifier is used for article analysis.
document analysis, text mining.
today’s world, news issues become an important part of every one’s
life. It is essential to know the information of the surrounding as
covering of news issues work is done by journalism. The covering of
controversial issues of the surrounding is an important function of
journalism as controversial issues grow in various areas such as
business, politics, sports, etc and these issues include various
participants and their various
It is notice that sometimes news articles get biased and unable to
present conflicting views of the issue. That’s why it is very
difficult for ordinary readers to analyze the conflicting views and
understand the controversy. Readers generally make their views
through single articles advanced news presentation models are need to
increases the awareness on differing views. In this paper, we survey
on a disputant relation based method for classifying news articles on
controversial issues. it is observed that the disputants i.e. the
participants who take his appropriate position and participate in the
controversy such as businessmen, politician, sportsmen, experts, news
writer, and so on are an important aspects for understanding the
conversation. News producers primarily shape articles on a contention
by selecting and covering specific disputants 2. Readers also
naturally understand the controversy by identifying who the opposing
Disputants are. The method helps readers naturally view the news
articles through the ‘opponent-based frame’ 1. It performs
classification in an unsupervised mode: it dynamically identifies the
opposing disputant groups and classifies the articles according to
their positions. As such, it successfully helps readers compare
articles of a controversy and achieve balanced understanding, free
from a specific biased viewpoint. The surveyed method differs from
those used in related tasks as it aims to achieve classification
under the opponent based frame. Most research on sentiment
classification and debate position appreciation takes a
topic-oriented view, and attempts to perform classification
the ‘positive versus negative’ frame for the given topic, for
example, positive versus negative about television. However, news
articles of a controversy are hardly classified under such frames.
a suitable argument frame is important. It provides a framework that
allows readers to naturally understand the controversy. It also
determines how classification methods should classify articles of the
Work In Document Classification
et al., and Pang et al., have been made researched on sentiment
classification in document-level. It aims to automatically recognize
and classify the sentiment of documents into positive or negative.
Opinion summarization aims a similar goal, to identify special
opinions on a topic and generate summaries of them 2, 3.
et al., developed an unsupervised method for generating summaries of
contrastive opinions on a common topic. Many of these works make a
number of assumptions that are difficult to apply to the conversation
of controversial news issues. They usually apply a single static
classification frame, ‘positive versus negative,’ to the topic
et al., had proposed that a number of works deal with debate attitude
recognition, which is a closely related task. They try to identify a
position of a dispute, such as ideological. This debate frame is
often not appropriate for controversial issues for similar reasons as
the positive/negative frame. In contrast, surveyed method does not
assume a fixed debate frame, and rather develops one based on the
opponents of the controversy at hand 5.
et al., and Agrawal et al., had proposed that the several works have
used the relation between speakers or authors for classifying their
debate stance. However, these works also assume the same debate frame
and use the debate mass, e.g., floor debates in the House of
Representatives, online dispute forums. Their approaches are also
supervised, and require training data for relation analysis, e.g.,
voting records of congress people 6, 7.
et al., had proposed the conversation of controversial issues in news
articles shows different characteristics from that studied in the
sentiment classification tasks. First, the opponents of a
controversial issue often discuss different topics. Second, the frame
of dispute is not fixed as, positive vs. negative 9.
Relation-Based Classification Method
disputant relation-based method implements the opponent based frame
for classification. It attempts to recognize the two opposing groups
of the matter at hand, and determines whether an article replicates
the position of a definite side more. The method is based on the
observation that there are usually two opposing groups of disputants,
and the groups compete for news coverage. They strive to influence
readers’ understanding, estimate of the issue, and grow support from
them 2. In this challenging process, news articles may give more
chance of speaking to a detailed side, explain or detailed them, or
supply facts helpful of that side. The surveyed method is performed
in three stages:
first step, disputant extraction, mines the disputants appearing in
an article set.
second step, disputant partition, divisions the mined disputants into
two opposing groups.
the news classification step classifies the articles into three
categories, i.e., two for the articles influenced to each group, and
one for the others.
method assumes polarization for controversial issues. This statement
was valid for most of the tested issues. For a few issues, there were
some participants who do not belong to either side; however, they
usually did not take a particular position nor make strong arguments.
For example, in the second issue Entrance of retailers to supermarket
business the government was in the middle between the big retail
companies and the small store owners. The government commented that
this is a difficult problem to solve and did not support a specific
side. Based on this observation, the method is designed to identify
opposing two groups of disputants and recognize articles biased to a
this step, the participants who participate in the controversy have
to be extracted/ mined. We expand that many disputants appear as the
subject of quotes in the news article set. The articles actively
quote or cover their action to deliver the controversy actively. We
used straight-forward methods for extraction of the subjects. The
methods were efficient in practice as quotes of articles often had a
regular pattern. The subjects of direct and indirect quotes are
mined. The sentences including a statement inside double quotes are
considered as direct quotes. The sentences that express a statement
without double quotes, and those describing the action of a disputant
are considered as indirect quotes (see the example 1 below). The
indirect quotes are identified based on the morphology of the ending
word. The ending word of the indirect quotes often has a verb as its
root or includes a verbalization suffix. Other sentences, typically,
those describing the reporter’s explanation or comments are not
considered as quotes (see example sentence 2).
that there would not be any talks unless Pakistan apologizes for the
is that a demanding response is the only solution for the current
opponent-based partitioning” method is developed for disputant
partitioning step. The method initially identifies two key opponents,
each representing one side, and uses them as a pivot for partitioning
other disputants 1. The other disputants are divided according to
their relation with the key opponents, i.e., which key opponent they
stand for or against.
partitioning it is a meaningful task to explore for more optimized
methods. The presented method is our first round solution to the
task, based on the observation of the criticizing structure that is
frequent in news article sets of controversial issues, i.e.,
existence of the key opponents who actively criticize and are
criticized by others, and existence of other minor disputants who
commonly criticize the key opponents but are not criticized often.
The presented key opponent-based partitioning is an initial algorithm
that operates this criticizing structure. The perception behind the
method is that there frequently exists key opponents who represent
the controversy, and many participants disagree about the key
opponents, whereas they rarely identify and talk about minor
key opponents: To identify the key opponents of the issue, we search
for the disputants who frequently criticize, and are also criticized
by other disputants. As the key opponents get more news coverage,
they have more chance to clear their argument, and also have more
chance to face counter-arguments of other disputants. This is done in
two steps. First, for each disputant, it is to be analyzing whom he
or she criticizes and by whom he or she is criticized. The method
goes through each sentence of the article set and investigates for
both disputant’s criticisms and the criticisms about the disputant.
Based on the criticisms, it analyzes the associations among
On the other hand, if the disputant is not the subject but
demonstrates in the quote, the sentence is considered to convey a
criticism about the disputant from another disputant. Second,
a modified version of the HITS graph algorithm to uncover major
disputants. For this, the criticizing relationships obtained in the
first step are represented in a graph. Each disputant is modeled as a
node, and a connection is made from a criticizing disputant to a
3.2: Modified HITS algorithm.
the HITS algorithm is designed to rate WebPages regarding the link
structure. The feature of the algorithm is that it separately models
the value of outlinks and inlinks. Each node, i.e., a webpage, has
two scores: The authority score, which reflects the value of inlinks
toward itself, and the hub score, which reflects the value of it’s
outlinks to others. The hub score of a node increases if it links to
nodes with high authority score, and the authority score increases if
it is pointed by many nodes with high hub score. The HITS algorithm
is adopted due to above feature. It enables us to separately measure
the significance of a disputant’s criticism (using the hub score) and
the criticism about the disputant (using the authority score). The
aim is to find the nodes that have both high hub score and high
authority score; the key opponents will have many links to others and
also be pointed by many nodes.
adapted HITS algorithm is shown in Fig. 3.2. Some adaptation is to
make the algorithm reproduce the disputants’ uniqueness. The initial
hub score of a node is set to the number of quotes in which the
matching disputant is the subject. The initial authority score is set
to the number of quotes in which the disputant appears but not as the
subject. In addition, the weight of each link (from a criticizing
disputant to a criticized disputant) is set to the number of
sentences that convey such criticism.
Given the two key opponents, we have to partition the rest of
disputants based on their relations with the key opponents. For this,
we identify whether each disputant has a positive or negative
relation with the key opponents. The disputant is classified to the
side of the key opponent with whom the disputant shows a more
positive relation. If the disputant shows a negative relation, the
disputant is classified to the opposite side.
are the four features to capture the positive and negative
relationships between the disputants:
Quote Rate (PQRab): Given two disputants (a key opponent a, and a
minor disputant b), the feature measures the ratio of positive
quotes between them.
Quote Rate (NQRab): This feature is an opposite version of PQR. It
measures the ratio of negative quotes between the two disputants.
of Standing Together (FSTab).: This feature attempts to capture
whether the two disputants share a position.
of Division (FDab): This feature is an opposite version of the FST.
It counts how many times they are not collocated in the sentences.
news article of the set is classified by evaluating which side is
prominently enclosed. The method classifies the articles into three
categories, either to one of the two sides or the category “other”.
is observed that the major components that shape an article on a
controversy are quotes from disputants and journalists’ commentary.
Thus, this method believes two points for classification: First, from
which side the article’s quotes came; second, for the rest of the
article’s text, the correspondence of the text to the arguments of
for the quotes of an article, the method computes the amount of the
quotes from each side based on the disputant partitioning step’s
result. As for the rest of the sentences, a similarity analysis is
conducted with an SVM classifier 8. The SVM classifier receives a
sentence as input, establishes its class to one of the three types,
i.e., one of the two categories, or other. It is qualified with the
quotes from each side. The related number of quotes from each side is
used for training. It is automatically obtained based on the
partitioning result of the earlier stage.
to survey, an article is classified to a precise side if more of its
quotes are from that side and more sentences are similar to that
side: Given an article a, and the two sides b and c,
a to b if,
a to c if,,
a to other, otherwise, where
Number of all sentences of the article
Number of quotes from the side i.
Number of quotes from either side i or j.
Number of sentences classified to i by classifier.
Number of sentences classified to either i or j.
and Future Work
this paper the problem of classifying news articles on controversial
issues is studied. It involves new challenges as the conversation of
controversial issues is complex, and news articles show different
characteristics from commonly studied amount, such as product
reviews. The conversation involves many topics and the arguments
often do not fit the ‘positive versus negative’ frame.
suggests that opponent-based frame is a clear and effective frame for
understanding controversial issues. The frame does not require
articles to cover a common topic nor the arguments to explicitly
express positive or negative sentiments. In this study, it can be
find that the participants easily identified the opposing disputants,
and classified articles written for or against the disputants.
Instead of taking a topic-oriented view, the disputant relation-based
method focuses on the disputants of the controversy.
better performance of Disputant relation based method, ‘Naive Bayes
algorithm’ can be used in article classification step as; SVM
classifier cannot classify the issue that involves three or more
Park, Jungil Kim, Kyung Soon Lee, and Junehwa Song, Disputant
Relation Based Classification for Contrasting Opposing Views of
Contentious News Issues, IEEE
Transactions On Knowledge And Data Engineering, Vol. 25, No. 12,
Turney, Thumbs Up or Thumbs Down? Semantic Orientation Applied to
Unsupervised Classification of Reviews, Proc.
40th Ann. Meeting Assoc. Computational Linguistics (ACL ’02), pp.
Pang, L. Lee, and S. Vaithyanathan, Thumbs up? Sentiment
Classification Using Machine Learning Techniques”, Proc.
Conf. Empirical Methods Natural Language Processing (EMNLP ’02), pp.
Paul, C. Zhai, and R. Girju, Summarizing Contrastive Viewpoints in
Opinionated Text, Proc.
Conf. Empirical Methods in Natural Language Processing (EMNLP), pp.
Somasundaran and J. Wiebe, Recognizing Stances in Ideological Online
NAACL HLT Workshop Computational Approaches Analysis and Generation
Emotion in Text (CAAGET ’10),
pp. 116-124, 2010.
Thomas, B. Pang, and L. Lee, Get Out the Vote: Determining Support
or Op- position from Congressional Floor-Debate Transcripts, Proc.
Conf. Empirical Methods Natural Language Processing (EMNLP ’06), pp.
Agrawal, S. Rajagopalan, R. Srikant, and Y. Xu, Mining Newsgroups
Using Networks Arising from Social Behavior, Proc.
12th Int’l Conf. World Wide Web (WWW ’03), pp.
Joachims, Making Large-Scale SVM Learning Practical, Advances
in Kernel Methods-Support Vector Learning,
B. Scholkopf, C. Burges, and A. Smola, eds., MIT Press, 1999.
Schon and M. Rien, Frame
Reection: Toward the Resolution of Intractable Policy