Learning Framework for Pronoun Resolution Using Convolution Neural Network

The task of understanding grammatical entities in sentences is a very important part of preprocessing in natural language processing tasks. Named entity recognition is a subtask of information extraction. The fundamental goal of NER is to induce and classify the defined categories such as person names, organizations, locations, and other entities which might be the requirement of the application. Parts of speech are another important aspect of preprocessing required by many tasks in NLP. The main challenge in most reference resolution systems is having a pre-labeled dataset that has entities that are rich in features such as NER and POS tags.This research study proposes a learning-based outcome resolution of grammatical entities in a self-curated data set in this study. The suggested model, which includes six classes, learns from a hand-annotated corpus and determines different classes of input entities. This system is the first learning-based model for the provided corpus to approach comparable performance. It pledges and achieves a high performance as compared to a non-learning outcome. The proposed model's the main challenge is to anticipate the references among the entities which are the base for understanding the relationship between each other. Towards this, the proposed model curate sentences that have entities that require references to be resolved to help build the mentions for coreference resolution systems. To test the proposed hypothesis, a learning-based approach with word embedding and position embedding techniques is proposed to classify various types of nouns and pronouns. The pronouns in the dataset are linked with the nouns and their types by following the approach of the binding theory. The evaluation of the model's accuracy is robust, with an F1 score of 97.45, recall of 99.20, and precision of 95.75, and it identifies the correct references and compare it to a state-of-the-art model.