Student feedback is essential for teachers to improve their teaching and for schools to evaluate faculty performance, and in turn, helps student learning. The feedback also provides suggestions and information for other students to choose courses to take. Identifying valid, useful and multidimensional information from student feedback with is an important problem to solve. With digital tools, teaching evaluation data are more widely collected and effectively stored. Therefore, researchers are allowed to develop and apply mixed methods to analyze teaching evaluation data. This study aims to develop a framework for analyzing quantitative and qualitative teaching data with aspect-based sentiment analysis, deep learning models, and other state-of-the-art techniques to help researchers, teachers, students, and educators better understand teaching evaluation. With large-scale long-term student evaluation data, this dissertation is also able to discover longitudinal trends and detect potential gender bias of teaching evaluation.