Identification of promising inventions considering the quality of knowledge accumulation: a machine learning approach

The identification of promising inventions is an important task in technology planning practice. Although several studies have been carried out using patent-based machine learning techniques, none of these have used the quality of knowledge accumulation as an input for identifying promising inventions, and have simply considered the number of backward citations as the link with previous knowledge. The current study therefore aims to fill this research gap by predicting promising inventions with patent-based machine learning, using the quality of knowledge accumulation as an important input variable. Eight criteria and 17 patent indicators are used as input variables, and patent forward citations are employed as the output variable. Six machine learning techniques are tested on 363,620 G06F patents filed between January 1990 and December 2009, and the results show that the quality of knowledge accumulation is the most important variable in predicting emerging inventions.