The unprecedented growth of scholarly literature published every year has affected many aspects of our lives. Despite the extensive studies of scholarly impact, there are broader impacts across society that remain underexplored. This thesis aims to predict the societal impact of research using information from a wide range of sources not limited to academic sources like citations. It identifies factors best suited to recognize scientific works that are most likely to be of interest to society. This has been achieved by building machine learning models that use three indicators of online attention: (1) whether a research article will be cited in public policy and the number of citations it is likely to receive (2) if a research article will be found newsworthy and the number of mentions it is likely to receive (3) public understanding of the research paper. This research also explores new approaches to measure the general public's understanding of scientific outcomes thereby enabling more accurate measurements of scientific literacy. Models were used to study relationships between public understanding of scientific outcomes and textual features extracted from scholarly text like average word length and average sentence length that are indicative of the text complexity.