Named Entity Recognition for Weather Domain in the Marathi Language

A vast volume of content is being published on the internet every second as a consequence of the advancement of digital media and the growing breadth of journalism. Computationally, this unstructured data is opaque. Named Entity Recognition (NER) is a two-step process of finding and then classifying crucial textual data. Most Named Entity Recognition work has already been addressed in English literature compared to other Indian languages. There are restrictions on research in the Marathi language because of scarce resources, and regional and morphologically complicated language. This work is concerned with the Weather Entity Recognition framework in the Marathi language. For predicting the sequences, Conditional Random fields(CRFs) and bidirectional LSTM (BiLSTM) models are used. To represent the word vectors, fastText and Transformer based mBERT embeddings are used. The result observed in the Bi-LSTM CRF model with FastText word embedding has an accuracy of 97.45%.