Paper: Learning A Scanning Understanding For Real-World Library Categorization

ACL ID A92-1043
Title Learning A Scanning Understanding For Real-World Library Categorization
Venue Applied Natural Language Processing Conference
Session Main Conference
Year 1992
Authors

This paper describes, compares, and evaluates three dif- ferent approaches for learning a semantic classification of library titles: 1) syntactically condensed titles, 2) com- plete titles, and 3) titles without insignificant words are used for learning the classification in connectionist re- current plausibility networks. In particular, we demon- strate in this paper that automatically derived feature representations and recurrent plausibility networks can scale up to several thousand library titles and reach al- most perfect classification accuracy (>98%) compared to a real-world library classification.