Dr. Dmitry Ustalov

Post-Doctoral Researcher

B6, 26, Room B 1.19
D-68159 Mannheim

Email: dmitry (at) informatik.uni-mannheim.de

Research Group: Natural Language Processing and Information Retrieval


I joined the Data and Web Science group in November 2017. My research is mainly focused on different aspects of computational semantics, especially on word sense induction and disambiguation as well as on automatic thesaurus construction and evaluation using unstructured data and crowdsourcing. In February 2018 I defended my Kandidat Nauk (PhD) thesis.

I am working on the JOIN-T project funded by the Deutsche Forschungsgemeinschaft (DFG).

You can find my publications listed on Google ScholarScopusdblp, and arXiv.

My ORCID iD is https://orcid.org/0000-0002-9979-2188 and my ResearcherID is P-6307-2014.

Research Interests

  • Natural Language Processing
  • Computational Semantics
  • Crowdsourcing

Selected Results

  • Watset, an efficient meta-algorithm for fuzzy graph clustering. This algorithm creates an intermediate representation of the input graph that naturally reflects the “ambiguity” of its nodes. Then, it uses hard clustering to discover clusters in this intermediate graph. Watset shows excellent results on synset induction task for multiple languages as reported in our ACL 2017 paper: doi:10.18653/v1/P17-1145.
  • Hyperstar, a regularized projection learning approach that transforms hyponym word embeddings into the corresponding hypernym word embeddings. The asymmetry of the “is-a” semantic relation is enforced by adding a regularization term to the loss function. As the result, more accurate hypernyms are generated using the same data as reported in our EACL 2017 paper: doi:10.18653/v1/E17-2087.
  • Mechanical Tsar, an open source engine for microtask-based crowdsourcing. This highly customizable engine enables Web-based data annotation by inviting volunteers either from the Internet or from a private crowd. Mechanical Tsar automatically does task allocation, worker ranking, answer aggregation, and agreement assessment as described in the paper: doi:10.15514/ISPRAS-2015-27(3)-25.

Invited Talks and Tutorials

Related Projects

Recent Publications