An Algorithm for Automatic Text Annotation for Named Entity Recognition using spaCy Framework

Text Annotation is the process of adding metadata in the text and used in various tasks like natural language processing (NLP) and machine learning models. Named entity recognition (NER) is one of the interesting and challenging tasks of NLP and is being used extensively in many domains. The application of NER will also be useful in handling documents, queries, reports and research articles related to agriculture in identifying pests affecting crops. SpaCy, a free and open source library is being used for NER that requires the text data in a complex annotated format. The process of manual annotation is difficult and time-consuming task. Therefore, to streamline the process of text annotation, we developed an algorithm and a tool for automatic annotation of text data. Approximately 3.6 million queries were collected from “Kisan Call Centre” , a helpline service to farmers by Government of India and plant protection queries of Paddy and Wheat crops were extracted from this database. These queries were annotated with the help of developed tool and annotated corpus was created. The annotated corpus is used to develop NER models and trained for crops and associated pests identification in agriculture domain. Further, the performance of the model is enhanced by reducing features using plural to singular conversion and synonym substitution. The model achieved an F1-score of 97.20%, demonstrating a significant improvement of 3.01% compared to the performance with original queries.