Inferring Attitudinal Spaces in Social Networks

Ideological scaling methods have shown that behavioral traces in social platforms can be used to mine opinions at a massive scale. Current methods exploit one-dimensional left-right opinion scales, best suited for two-party socio-political systems and binary social divides such as those observed in the US. In this article, we introduce a new method to overcome limitations of existing methods by producing multidimensional network embeddings and align them with referential attitudinal for a few nodes. This allows us to infer a larger set of opinion dimensions from social graphs, embedding users in spaces where dimensions stand for indicators of several social dimensions including (in addition to left-right cleavages) attitudes towards elites, or ecology among many other issues. Our method does not rely on text data and is thus language-independent. We illustrate this approach approach on a Twitter follower network. Finally, we show how our method allows us to analyze the opinions shared within various communities of social networks. Our analyses show that communities of users that have extreme political opinions are also more homogeneous ideologically.