Text representations and evolutionary based Intelligent information retrieval

Abstract

Abstract newlineThe rapid adoption of digitization in recent years has made it possible to easily access newlinea large number of documents; however, digging through such voluminous data to newlinedeliver timely and relevant information to user-specific needs for decision-making is newlinea complex and time-consuming process. The exponential surge in knowledge sources newlinehas intensified the difficulties in understanding the meaning of information in order to newlineretrieve appropriate outcomes from large datasets. In this thesis, we re-address an information newlineretrieval task using recent technological breakthroughs. Therefore, we propose newlinethree different IR frameworks to handle query and document representation for relevant newlineinformation retrieval. The research has investigated the existing state-of-the-art newlinemethodologies in the domain of information retrieval and proposed a swarm optimized newlinecluster-based framework, a phrase embedding-based query expansion framework, and a newlinetransformer-based deep semantic representation framework to effectively and efficiently newlineretrieve information. newlineThe decomposition of large datasets into small groups enables systems to gain a deeper newlineunderstanding of the context of information in less time. This emphasizes the need of newlinetopical search strategies that perform data categorization for searching. Unsupervised newlinetechniques such as clustering can be used to perform the decomposition of unstructured newlineand unlabeled data when the context is unclear. Dividing large datasets into small newlinegroups allows for faster and more accurate access to information. Therefore, we propose newlinea swarm optimized cluster-based framework with frequent pattern mining techniques to newlineretrieve user-specific knowledge from the extensive document collections. The preprocessing newlinetask is divided into two sub-tasks namely, document clustering and frequent newlinepattern mining. The first applies the proposed bio-inspired K-Flock clustering algorithm newlineto decompose document collection into small groups, and the second extract frequent newlinepatterns from the decomposed groups using a memory