Unseen Named Entity Recognition for Thai Language

Most common challenges in Named-entity recognition (NER) require suitable and annotated datasets, tailored to the specific domain of the documents under analysis, for example, legal, medical, and financial documents each require distinct datasets. Reliable identification of named entities within the training set is crucial. Annotations is a vital role in building an effective mechanism. However, creating training sets demands significant human effort, contributing to the high costs associated with NER. High quality training materials are indispensable for evolving machine-learning algorithms, underscoring the importance of dependable training data. Thus, the primary challenge lies in acquiring and managing data rather than in the algorithms themselves. To overcome most common situations of finding new unseen names, this paper proposes a mechanism to detect a new undiscovered name in Thai language based on sentence contexts. The method uses BI-LSTM model to learn sentence structures and predict a new unseen name from its context. Our framework comprises two main tasks. The first one is word entity extraction. The second task is using the model to identify unrecognized named entities. Evaluation of the proposed method is described with a successful rate of accuracy in discovering newly identified entities. This contributes to the expansion of entity recognition capabilities in Thai language processing