Unsupervised Spectral Sparse Regression Feature Selection using Social Media Datasets

Recent innovations in the social media sites such as Google Plus, Face book, Thomas cook, Twitter and Flickr have increased more and more subscribers to participate in social online events just as connecting with friends, families and other like-minded subscribers, posting statuses and blogs, sharing and communicating information etc. As a result usage of social media by the users has increased and generated unprecedented volume of social data such as linked, labeled and unlabeled data. Mining the data produced by the social media has its ability to dig out actionable patterns that could be valuable for government, business, users, and consumers. However, Social media documents are vast, dynamic, high dimensional, unstructured and forcible thereby open to new challenges. At present, in data mining Supervised Feature Selection method outperforms effectively in manipulating labeled and high dimensional data thereby providing an effectual mechanism for learning. However, there is no single method proven to be effective for mining the high volume of data produced by social media. An Unsupervised Feature Selection method employing spectral analysis taking social media dataset have been proposed. This paper also aims at conducting experiments with the data sets from different social media to evaluate its accuracy and performance.