Abstract

We propose a NE (Named Entity) recognition system using a semisupervised statistical method. In training time, the NE recognition system builds error-prone training data only using a conventional POS (Part-Of-Speech) tagger and a NE dictionary that semi-automatically is constructed. Then, the NE recognition system generates a co-occurrence similarity matrix from the error-prone training corpus. In running time, the NE recognition system constructs AWDs (Acyclic Weighted Digraphs) based on the co-occurrence similarity matrix. Then, the NE recognition system detects NE candidates and assigns categories to the NE candidates using Viterbi searching on the AWDs. In the preliminary experiments on PLO (Person, Location and Organization) recognition, the proposed system showed 81.32% on average F1-measure.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call