Discovering promising research themes in a scientific domain by evaluating semantic information extracted from bibliometric databases represents a challenging task for Natural Language Processing (NLP). While existing NLP methods generally characterize the research topics using unique key terms, we take a step further by more accurately modeling the research themes as finite sets of key terms. The proposed approach involves two stages: identifying the research themes from paper metadata using LDA topic modeling; and, evaluation of research theme trends by employing a version of the Mann-Kendall test that is able to cope with multivariate time series of term occurrences. The results obtained by applying this general methodology to Information Security domain confirm its viability.INDEX TERMS LDA topic modeling, multivariate Mann-Kendall test, natural language processing, paper metadata, research theme, research trend.