Automated Text Mining Comparison Of Japanese And USA Multi-robot Research
Free (open access)
R. J. Watts, A. Porter & B. Minsk
Wouldn’t it be nice to automatically categorize documented research to distinguish organizational R&D emphases? For example, in the area of multirobot research, one could surmise that Japanese sources have less emphasis than expected on biological approaches and reconfigurable robots, and more emphasis than expected on human-robot interface, motion coordination and robot learning. Development and analysis of such a capability comprises the subject of this paper. Over the past decade, we have been developing a software tool for text analysis and classification, the Technology Opportunities Analysis of Scientific Information System (Tech OASIS). The Tech OASIS tool suite contains two information clustering algorithms, both of which evolved from principle components analysis (PCA). Based on the terms contained in an abstracts record field (e.g., abstract noun phrases, keywords, class codes, etc.), Tech OASIS generates topical clusters. We have now developed a capability to assess expectancy (i.e. binomial distribution likelihood) of occurrences. We demonstrate the derived analysis technique on the abstracts of 354 multi-robot research papers, principally, those from Japanese and USA sources. Keywords: text mining, expectancy measure, principal components analysis, factor mapping, principal components decomposition, multi-robot research, VantagePoint, visualization of statistical measures. 1 Introduction We first provide a brief background on information profiling and the software tool used to perform this analysis of multi-robot research. We then familiarize
text mining, expectancy measure, principal components analysis, factor mapping, principal components decomposition, multi-robot research, VantagePoint, visualization of statistical measures.