The amount of information is rapidly increasing in the information age. Paradoxically, it is becoming difficult to find credible information. This phenomenon can have various negative effects. The associated social costs are increasing, and the digital divide can become worse. Vulnerable users with low levels of domain expertise might not be able to select credible information on their own. Thus, one of the primary goals of this dissertation research is to create a more reliable information retrieval environment for general users by developing an intelligent system that can automatically predict the credibility of information. This study combined both social science and technical approaches to an interdisciplinary problem. From the human side, this study examined the criteria people use to judge the credibility of health information on social media by conducting a content analysis of randomly selected health information. From the system side, the study operationalized these criteria as features for machine-learned models. Finally, the dissertation study evaluated the credibility models based on the ground-truth data that was annotated by experts and crowd workers. This study has contributed to our knowledge of credibility assessment of health information in various ways. The study discovered an extensive set of credibility factors, operationalized them, and verified their discriminative powers in predicting credibility. The study illustrated how different credibility factors were associated with particular contextual factors (topic, prior knowledge, and type of social media) in which people judge the credibility of information. This study also empirically examined the application of the Elaboration Likelihood Model to the context of credibility assessment. The study findings have implications for label creation for supervised learning, theory development, and future work. In terms of label creation, there are implications for the criteria to select coders and to give proper weights to their judgments and integration of human’s feedback to reinforce those criteria. This study showed the potential to test a theory-based model using Data Science approaches by integrating crowdsourcing, feature categorization, and feature ablation study. Finally, this study provides meaningful insights for the next phase of credibility research, in which adaptive and personalized credibility models needs to be investigated.