Real-World Deep Learning for NLP

Deep Learning is the most promising field of research these days, one can read about amazing achievements on every day. However, implementing and using those promising results into an existing system is far from trivial and the arising problems and costs cannot justify the gain in precision and other measures. We present two fairly simple deep learning tools built for solving real-world problems we faced during building the backend for Profession.hu, one of the leading job search sites in Hungary. First, we introduce a variant of Deep Spell, a spell checker optimized for search terms. Second, we discuss using word2vec for finding synonyms for building a job thesaurus. Finally, we discuss the pros and cons of the two solutions.

Varjú Zoltán

Research Lead, Precognox

Zoltan Varju is the Research Lead at Precognox. When he is not working on NLP and search projects, Zoltan is organizing the Hungarian Natural Language Processing Meetup, writing for nyest.hu, a Hungarian popsci magazine.