MAPPING AND END USER INTERFACE FOR TOPIC MODELS OF NIH GRANTS

点击翻译按钮获取中文摘要

In the fall of 2007, NINDS initiated a review of funded grants as part of a general strategic planning effort. The Institute was directed by a member of the NINDS advisory council to a web-based technique for analyzing the ~14,000 Annual Meeting abstracts of the Society for Neuroscience (http://scimaps.org/maps/neurovis/new/). This technique used a publically available topic modeling algorithm known as Latent Dirichlet Allocation to determine the topic distribution of individual SfN abstracts, and used a simulated annealing graphical layout algorithm known as VxINSITE (later incorporated into DrL - SAND2008-2936J: Sandia National Laboratories) to map the topical relationships between abstracts into a Google Maps application interface, in which users can select sets of abstracts from a two-dimensional color-coded map and access topical information regarding these abstracts via an underlying database. The team that developed this framework, Gully Burns at University of Southern California (USC), David Newman at University of California Irvine (UCI), and Bruce Herr at Indiana University (IU), agreed to apply it to a set of NIH grants in order to test its potential utility for NIH staff and extramural scientists. Subsequently the project was expanded to model and map the entire 2007 set of NIH grants, with a resulting color-coded map and topic database. The results of this work can be viewed at http://scimaps.org/maps/ninds/ and http://scimaps.org/maps/nih/2007/. For the upcoming project period, the Institute has enlisted the services of the Information Extraction and Synthesis Laboratory (IESL) at the University of Massachusetts to perform topic modeling of NIH grants. The basic grants data plus the results of the topic model analysis will be transferred to Chalklabs, which will integrate it into an interface for Topic Model Browsing and for mapping using the DrL algorithm.