VFF: Visualization of Duke Research via Analysis of Faculty Publications

We will present a new interactive visualization for exploring a network of Duke authors based on data from the Scholars@Duke system, along with the novel algorithm that was used to generate the ¿topic model¿ the network is based on. The model is an automatic method for information extraction from data consisting of "multiway" relationships. A multiway data set consists of multiple set of entities and describes the relationships among these entities. For example, when analyzing a corpus of research publications, the multiway relationships may correspond to the co-occurrence statistics of three sets of entities ¿ authors, words, and publication venues (e.g., author "Alice" used the word "Genes" 5 times in her papers published at venue "Nature"). Our method takes this multiway data and infers "themes" or "topics" in the data as well as the strength of association of entities with each of these topics. These topics can be used as attributes of the entities. For example, each author can now be described by the topics she uses the most when writing her paper. These attributes can then be used for solving various problems of interest, such as grouping authors based on the author-author similarity, or matching authors to the publication venues that they will most likely be interested in.