Automatic Expansion of a Social Network Using Sentiment Analysis

In this paper we present an approach for automatic learning of a signed social network from online news articles. The vertices in this network represent people and the edges are labeled with the polarity of the attitudes among them (positive, negative and neutral). Our algorithm accepts as its input two social networks ex-tracted via unsupervised algorithms: 1. A small signed network labeled with atti-tude polarities (see Tanev 2007). 2. A quotation network, without attitude polarities, consisting of pairs of people where one person makes a direct speech statement about another person (see Pouliquen et al. 2007). The algorithm which we present here finds pairs of people who are connected in both networks. For each such pair (P1, P2) it takes the corresponding attitude polarity from the signed network and uses its polarity to label the quotations of P1 about P2. The obtained set of labeled quotations is used to train a Naïve Bayes classifier which then labels part of the remaining quotation network and adds it to the initial signed network. Since the social networks taken as the input are extracted in an unsupervised way, the whole approach including the acquisition of input networks is unsupervised. We carried out experiments with online English-language news and succeeded to increase the number of edges of the input signed network by 386%. The accuracy (62%) would require further improvements, but could be judged as good, if we take into account that the whole process is completely unsu-pervised.

TANEV Hristo;  POULIQUEN Bruno;  ZAVARELLA Vanni;  STEINBERGER Ralf; 

2011-01-24

Springer

JRC63159

978-1-4419-6286-7,   

1934-3221,   

http://www.springer.com/business,    %26,    management/business,    information,    systems/book/978-1-4419-6286-7,    https://publications.jrc.ec.europa.eu/repository/handle/JRC63159,   

10.1007/978-1-4419-6287-4,