We develop a novel approach to study overeducation by extracting pre-match information from online recruitment platforms using word segmentation and dictionary building techniques, which can offer significant advantages over traditional survey-based approaches in objectiveness, timeliness, sample sizes, area coverage and richness of controls. We apply this method to China, which has experienced a 10-fold expansion of its higher education sector over the last two decades. We find that about half of online job-seekers in China are two or more years overeducated, resulting in 5.1% pay penalty. However, the effect of overeducation on pay varies significantly by college quality, city type, and the match of college major with industry. Graduates in STEM (Science, Technology, Engineering and Mathematics) or LEM (Law, Economics and Management) from Key Universities are much less likely to be overeducated in the first place, and actually enjoy a significant pay premium even when they are in the situation.