Design and Implementation of Efficient Cross

Abstract

newline Traditional information retrieval methods are designed to consider keyword-based matching. They cannot provide accurate outputs when words are not lexically similar, but semantically they are having the same meaning (like the words midday and noon ). In Intelligent retrieval of information, not only the word but the contextual meaning of it is also considered. Similarity measures are based on human judgment of relatedness of words or concepts. Human capacity to evaluate things has long been studied by cognitive science and psychology. The semantic similarity computing of two texts are needed in many natural language processing applications such as information retrieval, summarization, or textual entailment. newlineThis thesis describes various problems with existing information retrieval systems, and presents competent solutions based on cross-modal retrieval instead of single-modal retrieval. Due to the presence of huge multimodal data on the Internet, cross-modal retrieval has become the need of the hour, such as using a text query to search for images and image query to search for text. In cross-modal retrieval, the correlation of different modalities is used for the retrieval task. For instance, if someone is looking for a cricketer, they can use a text or image query and cross-modal retrieval systems will provide the results in the form of relevant multimodal data like images, text descriptions, videos, and more. newlineUnlike traditional retrieval techniques, cross-modal retrieval utilizes the widely available data from various modalities and meets the user s demands in receiving documents in different modalities from queries. To perform cross-modal retrieval, the key problem is to measure the semantic similarity between data in different modalities (like Text, Images, Audio and Video), because each modality has their own representation. So these varied representations need to be mapped into a single semantic space, from where multimodal query results can be produced. In this dissertation, the...