Published Paper Details:

The extremely high pace at which data increases in domains such as health and sports makes it increasingly apparent that there is a need for systems that can quickly extract and process domain-specific information. Often such traditional approaches would have different models for each of the domains, which brings even more complex and resource-consuming setup processes. This work introduces a joint pipeline towards extracting information across domains in such a situation where transfer learning is efficient. The understanding that makes it even more compelling is that it uses fine-tuning on multi-domain tagged datasets like healthcare and sports to eliminate the need for separate domain-specific models by pre-trained transformer models available like BERT. Architecture provides domain classification as well as named entity recognition (NER) and relationship extraction. A new prioritization mechanism will evaluate the relevance of the individual lines read in the input text by the number of extracted entities and their associations. Lines having greater relevance will be sequenced for the generation of queries and Extraction of information; thus, efficient and meaningful information Extraction is realized. This system will effectively incorporate the query formulation and Extraction components for accurate and relevant data Extraction in healthcare and sports applications. Experimental results show that the unified approach produces a simplified architecture, reducing resource consumption while improving scalability for flexible and efficient use in multi-domain applications of natural language processing.