Presentation #240.02 in the session Catalogs — iPoster Session.
What publications mention a given observatory? How many refer to a concrete telescope or instrument? Quickly answering these questions can boost the efficiency of astronomers, librarians, and administrative personnel when writing reports, applying for new funds, or collecting institutional statistics. The NASA Astrophysics Data System (ADS) has more than 16 million bibliographic records for which it has indexed the full-text content of more than 6 million scientific articles (more than 1.3 million of which correspond to astronomical publications). In this ocean of data, the NASA ADS allows users to quickly string match words and sentences against the indexed full-text content but this does not solve the word sense disambiguation problem (e.g., “Planck” can refer to the person, the mission, the constant or several institutions). The NASA ADS is preparing a dataset of manually tagged entities which is being used to train Deep Learning models to automatically recognize and disambiguate facilities. This poster details our efforts and lessons learned in our adventure to improve the NASA ADS search capabilities using Deep Learning techniques.