Feature Selection Methods For Understanding Business Competitor Relationships
Rahul Gupta, Jay M. Pujara, Craig A. Knoblock, Shushyam M. Sharanappa, Bharat Pulavarti, Gerard Hoberg, Gordon M. Phillips · 2018
Understanding competition between businesses is essential for assessing the likely success of new ventures or products, for making decisions before investing capital in new businesses, and understanding the impacts of regulatory policy. One important resource for analyzing competitor relationships are business webpages, which can capture the mission, products, services, and key markets associated with a company. However, webpages also contain irrelevant, extraneous, or misleading text, hampering prediction. To address this challenge, predictive models use a process known as feature selection to identify only relevant terms. The diversity and specificity of business domains pose a challenge for automated approaches for feature selection. In this paper, we compare two approaches to feature selection: manually-curated lists of terms provided by experts and automated approaches to feature selection. We evaluate several approaches to feature selection and their impact on predicting competitor relationships, demonstrating that carefully designed automated feature selection approaches can surpass the performance of manually-curated word lists by 10%.