Graphsage-Based Named Entity Recognition for Malayalam

The primary goal of a Named Entity Recognition (NER) system is to gather relevant information from the given text. NER in Malayalam text is challenging due to the agglutinative structure of the language. Due to the increasing presence of the Internet, more people use smartphones, leading to an increase in Malayalam search queries. A major portion of these searches contains Named Entities which makes it reasonable to study methods to identify them effectively. Words can have relations with other words in a sentence. If these relationships are considered as edges of a graph, the sentences could be processed as a graph instead of in a sequential manner. Graph Neural Networks (GNN) have been used for processing graphs in recent years. In this work, graphs for sentences in the Malayalam language are created using a simple nearest word method and processed in a GraphSAGE (Graph Sample and Aggregated)-based NER system.