Extracting and Learning Social Networks out of Multilingual News

Various kinds of social networks can be derived from the analysis of news articles. We present here our experience in building social networks by the extraction of relationships between entities all automatically derived from multilingual news articles. Unqualified relationships between persons can be extracted through simple co-occurrence statistics. Qualified relationships can be extracted using linguistic patterns. Our highly redundant sources (50,000 daily articles in 40 languages) are used to both validate our algorithms and strengthen pertinent relationships. Due to the amount of data we process these social networks provide a complex challenge for their useful visualization and navigation.