Identification & Mitigation of Trademark Infringement is critical to protecting an organization’s trademark. Typically trademark Infringement has been identified using manual methods.

Trademark Infringement is the unauthorised use of a trademark in such a way that it causes confusion and leads to deception among customers. It is not to be taken lightly as it can cost companies loss of millions of dollars’ worth of revenue. Thereby resulting in an inefficient, costlier, and time-consuming legal proceedings. Increase in E-commerce had led to exponential rise in Trademark Infringement.

In this study, participants automated Trademark Infringement detection and legal relevant information extraction from past judgements. They aimed to reduce the manual effort & time taken by both the Legal research team and the Trademark Search using advanced Machine Learning Tools by deploying Logomark Similarity Search, Wordmark Infringement Detection, E-commerce Infringement Detection and Past Judgement Analysis.

They built a one-stop model for wordmark infringement - ‘Ensemble Match’ that combined similarity algorithms. The team experimented with multiple Deep Learning and traditional approaches to retrieve similar logo marks and combined them using simple yet effective Inverse Rank Position approach for logomark similarity search. The team was able to identify 15 matched past trademark infringement judgements using the latest state-of-the-art SBERT model for semantic search.


The business value of the project was to provide a one-stop solution to look for Wordmark Infringement which provides better potential infringement detection and few more similarity searches which are not considered currently by WIPO but considered as infringement by the Indian courts. This new and innovative solution will help in reducing manual effort and time involved in “Trademark Search”.

The researchers used the following techniques for the project: Search Wrapper, Infringement Classification, Rule based Classification Model, Random Forest, Roberta, Glove, SBERT, Cosine Similarity distance metric, LexNLP Model, RegEx, LSTM decoder-encoder, Attention Model, Custom NER model.


References:

1) Kumar S., Boorugu A., Jalasutram N., Mittal N., Jethwani N., Jain T., ‘Trademark Infringement Lookout, Identification and Stoppage Automation using Machine Learning (TILSIM)