In this paper we present the RuSentRel corpus including analytical texts in the sphere of international relations. For each document we annotated sentiments from the author to mentioned named entities, and sentiments of relations between mentioned entities. In the current experiments, we considered the problem of extracting sentiment relations between entities for the whole documents as a three-class machine learning task. We experimented with conventional machine-learning methods (Naive Bayes, SVM, Random Forest).