In this work, we present an innovative method for generating research paper titles from natural language processing for a given text using a Transformer Language Model GPT-3 that has been trained. The proposed method utilizes a localization strategy to select a suitable title from a pool of potential titles, which is then refined or de-noised to produce the final title. The pipeline of the method comprises three modules: Inception, Election, and Filtration, followed by a Labeling function. The Inception and Filtration modules are based on GPT-3, while heuristic algorithms are used in the Election module. Despite having limited training data, our model can generate appropriate titles since the Natural Language Inception skills are learned from an earlier stage of training, and this model only needs to master certain corpus and task-based skill features. The Election and Filtration modules ensure that the final titles are accurate in terms of semantics and syntactic structure and reflect the provided material. To refine our research article abstract model, we use the ArXiv and test it against various test sets. Our proposed method produces promising results when evaluated using the BLEU and ROUGE metrics against the test sets. We also perform human evaluation to validate the outcomes produced by our suggested technique.