Session: DTM-02: Design Methods and Practice

Paper Number: 148503

148503 - Topic Evolution: Insights From Five Years of Idetc Conference Papers 

In the modern world, with a large amount of information accumulated, many companies and organizations have started to analyze data to explore the trends and to forecast emerging technologies and future innovations, thus helping them to make effective decisions such as marketing initiatives to make them competitive. One popular method to identify trends is by identifying topic relationships within large text corpora. For example, a study has been conducted to identify major academic branches to explore the research trends in academia [1]. There are also studies building topic evolutionary pathways to forecast future research directions [2]. Among these studies, the most common technique used is network analysis, such as co-word analysis, where the network is based on the co-occurrence of words [1]. However, researchers have realized that frequency is not the best way to identify the most important words in an article, so other methods are also developed to extract keywords. For example, Huang et al. transformed words into vectors and computed the word vector cosine similarities to construct keyword networks [2]. Both the co-word and keyword network analysis create a word network and apply clustering techniques to find topics within the documents. However, words have different meanings in different contexts, and disagreements can arise when defining a common label for a group of words. Therefore, instead of constructing a keyword network, this study develops a document network where each node in the network is a document instead of a keyword.

Presenting Author: Siyi Xiao Texas A&M University

Presenting Author Biography: PhD student in Mechanical Engineering Department at Texas A&M University.

Authors:

Siyi Xiao Texas A&M University
Daniel A. Mcadams Texas A&M University