Trademark Design Code Identification Using Deep Neural Networks

Abstract

Trademark review and approval is a complex process that involves thorough analysis and review of the design components of the marks including the visual characteristics as well as the textual mark description data specifying the significant aspects of the mark. One of the crucial aspect in review of the trademark application is determining the design codes of the trademarks based on their mark description. Currently, the process of identifying the design codes for a trademark is performed manually in the United States Patent and Trademark Office (USPTO) and takes substantial amount of time. Recently, word embeddings and deep neural networks (DNNs) have demonstrated excellent performance in computer vision and various natural language processing (NLP) tasks such as machine translation, speech recognition, sentence and document classification etc. to name a few. In this study, we explored fastText and different neural networks such as Convolution Neural Networks (CNN), Long Short Term Memory (LSTM), bidirectional versions of both LSTM and Gated Recurrent Unit (GRU) and Recurrent Convolutional Neural Network (RCNN) to automate trademark design code classification based on their mark description. Overall, it was found that the trademark word embeddings with RCNN model outperformed other models. Our study thereby seeks to provide a solution towards the time intensive and laborious process of identifying design codes of the trademarks.