Matching Innovation Descriptions with Organization Profiles

Abstract

Open Innovation platforms are becoming increasingly common to share knowledge. Organizations searching for innovative ideas and providers of innovations have to sign up on these platforms to participate in the market. Establishing a match between the two parties can be challenging and even a matter of luck. This thesis proposes a practitioner-oriented take on how this process can partially be automatized by making use of techniques from the area of Natural Language Processing and Information Retrieval. A prototypical implementation was built in the Java programming language to validate the premise that such a system can be built and that it yields reasonable results. The document compares and evaluates methods to extract relevant keywords and other information from the websites of organizations as well as from textual innovation descriptions. The resulting organization and innovation proles can then be matched with the help of an enterprise search engine. The most promising methods were evaluated using the prototype. A qualitative review highlights the strenghts and weaknesses of each approach. It is shown that keywords can indeed be extracted from websites and innovation descriptions, albeit with mixed outcomes. While the results of certain methods such as Latent Dirichlet Allocation were rather underwhelming, other approaches such as keyword assignment with a controlled vocabulary yielded keyword candidates comparable to humans.