Identifying and Predicting Patent Trends in the Artificial Intelligence Domain

This paper determines underlying trends in a patent dataset to identify top topics in the Artificial Intelligence (AI) domain. A lot of inventors are investing in AI-based portfolios. Owing to the time lag from filing to the grant of the application, the innovators prefer predicting patent topics before innovating. In this paper, four Natural Language Processing (NLP) algorithms were used to understand crucial topics of AI in the abstract column of the original dataset. The NLP algorithms were Latent Dirichlet Allocation (LDA), Bag of Words (BoW), Term Frequency – Inverse Document Frequency (tf-idf), and Global Vectors for Word Representation (GloVe) algorithms. The abstract is chosen since it is a replica of the first claim, which is important enough to cover the necessary topics along with the legal aspect that is required by the inventors. Further, NLP-based analysis is performed by segregating the abstract for the top five assignees. The LDA and BoW algorithms performed best as processing time was reduced without trading the accuracy of the results for both the datasets. However, only the results of LDA model were chosen as it identifies relevant context of the underlying topics of the documents, while BoW focuses only on word frequency without capturing context. Finally, the LDA algorithm’s outcome of the top topics was used for prediction in the upcoming five years, since it was the best performing model among the four tested NLP algorithms for the original patent dataset.