Recommendation systems are widely applied in many fields, such as online customized product searches and customer-centric advertisements. This research develops the methodology for a patent recommender to discover semantically relevant patents for further technology mining and trend analysis. The proposed recommender adopts machine learning (ML) algorithms for natural language processing (NLP) to represent patent documents in vector space and to enable semantic analyses of the patent documents. The ML approach of neural network (NN) language models, trained by domain patent documents (text) as a training set, convert patent documents into vectors and, thus, can identify semantically similar patents using document similarity measures. In particular, the proposed recommender is deployed to in-depth case studies for advanced patent recommendations. The case domain of smart machinery is used to better enable smart manufacturing by incorporating innovative technologies, such as intelligent sensors, intelligent controllers, and intelligent decision making. The research uses six sub-domains in smart machinery technologies as the case studies to verify the superior accuracy and efficacy of the recommender system and methodologies.