Identifying Valuable Knowledge Topics in Innovation Communities Using Innovation-LDA

Abstract

Researchers and practitioners have recognized that user-generated content in the innovation community plays an important role. However, it is challenging to automatically identify valuable knowledge from these unstructured texts. Thus, in this study, we propose an efficient model for extracting innovation-oriented topics and, simultaneously, for assigning discovered topics to each post in the online innovation community. Specifically, we introduce a variant of the latent Dirichlet allocation (LDA) topic model, called the Innovation-LDA model, which comprehensively considers users’ interests (reflected by pageviews and replies) and the structure of threads (e.g., header or body) to generate the valuable topics. We access the quality of discovered information through statistical fit as well as substantive fit. Based on our experimental results, we can conclude that our proposed method exhibits better performance than that of the contrasted method and can locate more meaningful innovation topics; that is, our innovation-LDA model is capable of not only identifying more rigorous topics for each thread by utilizing the text structure but is also capable of learning more semantic and coherent themes from user interests. This investigation expands topic identification research by providing both a new theoretical perspective and useful guidance for enterprises in product innovation.