Assessment of Pharmaceutical Patent Novelty with Siamese Neural Networks

Patents in the pharmaceutical field fulfil an important role as they contain details of the final product that is the culmination of years of research and possibly millions of dollars of investment. It is crucial that both patent producers and consumers are able to assess the novelty of such patents and perform basic processing on them. In this work, we review approaches in the literature in patent analysis and novelty assessment that range from basic digitisation to deep learning-based approaches including natural language processing, image processing and chemical structure extraction. We propose a system that automates the process of patent novelty assessment using Siamese neural networks for similarity detection. Our system showed promising results and has a potential to improve upon the current patent analysis methods, specifically in the pharmaceutical field, by not just focusing on the task from a Natural Language Processing perspective, but also, adding image analysis and adaptations for chemical structure extraction.