One practical and low-risk approach to product planning for technology-based firms is to identify applicationproducts based on their existing product portfolios. Previous studies, however, have tended toneglect the current product development capabilities of target firms and to apply the technical data ofspecific fields to their methods, thereby failing to quantify a way of identifying various product opportunities.As a remedy, this paper proposes a new multi-step approach to product recommendation. Thesteps include (1) generating assignee–product portfolio vectors using text mining on a large-scale sampleof patents, (2) recommending untapped products for a target firm by using latent Dirichlet allocation andcollaborative filtering, (3) producing a visual map based on the promise and domain heterogeneity of therecommended products. To validate the practicability, we applied our approach to a Korean high-techmanufacturer by using all of the patents registered in the United States Patent and Trademark Officedatabase during the period of time from 2009 to 2013. This study contributes to the systematic discoveryof new product opportunities across various domains using the existing product portfolios of firms, andcould become the basis for a future product opportunity analysis system.