Fast and Accurate Patent Classification in Search Engines

This article presents a new approach to large scale patent classification. The need to classify documents often takes place in professional information retrieval systems. In this paper we describe our approach, based on linguistically-supported k-nearest neighbors. We experimentally evaluate it on the Russian and English datasets and compare modern classification technique fastText. We show that KNN is a viable alternative to traditional text classifiers, achieving comparable accuracy while using less additional hardware resources.