🧑‍🔬 Tim Finin

Tim Finin's papers here sit where natural language processing meets structured knowledge. They ask how a system can tell when two pieces of text mean the same thing, and how a name in a document can be tied to the right entry in a large reference resource. The methods mix corpus statistics with curated lexical and knowledge resources, and supervised learning with careful feature design. They touch on Information extraction, Named entity recognition and Knowledge graphs & ontologies.

The most cited work, UMBC_EBIQUITY-CORE: Semantic Textual Similarity Systems (2013), describes a team's entries to a shared task on scoring sentence pairs for similarity. All the entries rest on a common word-similarity measure. It derives latent semantic analysis scores from a cleaned web corpus and corrects them with WordNet relations. One entry aligns words across sentences without any training, while the others combine many features in a regression model. Close behind is Entity Disambiguation for Knowledge Base Population (2010). It links mentions of people, organisations and places to a knowledge base through fast candidate filtering followed by a learned ranker. In that ranker, "no matching entry" competes as one more option.

Spanning 2010 to 2013, these papers move from entity linking toward measuring meaning at the level of words and sentences. Finin works at the University of Maryland, Baltimore County, with earlier affiliations at Unisys and the University of Pennsylvania. Co-authors on the entity work include Mark Dredze, Paul McNamee, Delip Rao and Jonathan Weese.

Tim Finin has 2 papers in our library, in 2010–2013, most of them on Named entity recognition, Information extraction, Knowledge graphs & ontologies; often with Delip Rao, Jonathan Weese, Mark Dredze.