Khmer Named Entity Recognition Based on LSTM-CRF Model

Named Entity Recognition(NER), as a significant research content of information extraction, is broadly used in Natural Language Processing fields such as information retrieval, machine translation, question answering system, and so on. Named Entity Recognition in Khmer plays a supportive part in the study of Chinese–Khmer bilingual understanding. However, the differences between languages make it hard to transfer the common Chinese and English Named Entity Recognition into Khmer. Aimed at the problem that the output of long short-term memory neural network model does not consider the sequence of tags, the output of long short-term memory neural network model and Khmer entity feature is used to input features of conditional random field model to extract Khmer corpus entity. The result of experiment shows that the precision of the conditional random field model is increased by 1.32%, recall rate by 2.55% and the F1 value which is used to measure the precision and recall rate by 1.92% after using long short-term memory neural network as the inputted feature of conditional random field model, which unites the output of neural network model with the entities in Khmer. The outcome of experiment proves the effectiveness of this method.