In this paper we present a new Russian paraphrase corpus derived from the news feed of the social network and conduct its primary analysis. Most media agencies post their news reports on their pages in social networks, and the headlines of the messages are often the same as those of the corresponding news articles from the official websites of the agencies. However, sometimes these pairs of headlines differ, and in such cases a headline from the social network can be considered a compression or a paraphrase of the original headline. In other words, such news feed from social networks is a rich resource of textual entailment, and, as it is shown in this paper, various linguistic phenomena, e.g., irony, presupposition and attention attracting markers. We collect the described pairs of headlines and construct the Russian social network news feed paraphrase corpus based on them. We test the paraphrase detection model trained on the other existing Russian paraphrase corpus, ParaPhraser.ru, collected from official news headlines only, against the constructed dataset, and explore its linguistic and pragmatic features.