Who Leads Whom: Topical Lead-Lag Analysis across Corpora

In this work, we study the problem of whether grant proposals lead academic publications in terms of generation of scientific ideas. This is an important computational social sciene question that can help us understand the dynamics of scientific innovation. We propose simple but scalable techniques for lead/lag estimation, based on LDA and time series analysis, that work on any unlabeled textual corpora with temporal information. We perform our analysis on about half a million Computer Science research paper abstracts and 20,000 successful NSF grant proposal abstracts that represent the entire field of Computer Science in the time span of 1991-2008. Our analysis, besides revealing interesting patterns, finds that the lead/lag of scientific papers with respect to grant proposals is highly topic specific.