Abstract: The applicability of the existing technology opportunity mining results is relatively low owing to the small sample size and the lack of evaluation on the technology application prospects in the mining process. In order to solve this problem, with the goal of improving the applicability of mining results, based on the existing research, this paper proposes a three-dimensional patent prediction model by taking into account a large number of patents and adding an assessment of the prospects for technology applications. Using the PLSA algorithm in machine learning and combining with the MapReduce computing framework under Hadoop, it uses patent text mining to construct the technology and function dimensions of the patent prediction model, adopts entropy weight and TOPSIS method to construct the value dimension of the patent prediction model, and fills the element items in the patent forecasting model based on the MapReduce computing framework. Then, it applies a patent prediction model to 133508 patent texts in the titanium field in the DII database from 1999 to 2018. The results show that the model has identified a total of 3 priority and 2 secondary technology opportunities in the titanium field, and these technology opportunities can be developed in order of priority. This model enriches the method for technology opportunity mining and provides a more accurate and prospective technology research and development direction for innovation subjects.

Key words: machine learning algorithms, patent text mining, technology opportunity, titanium