Creating AI-powered Trademark Search for Legal Firm
Vostok AI team was approached by large Intellectual Property bureau (est. 1988) to create innovative solution that will help to check online if their customer’s trademarks were duplicated. The solution is planned to be used by experts in IP management, creative agencies and companies planning to expand in a new market and registering their trademark.
Customer Challenge.
Our customer wanted to be able quickly check the logo for plagiarism and find if any duplicates. Before project implemented all was done via Google Images search, thus resulting in many iterations and lots of manual work. This solution wasn’t fast and accurate enough to meet our customer requirements, thus they asked to develop special Neural Network to be able to find duplicated logos and trademarks. Keeping in mind that customer database consisted of 1.5 million trademarks, it was essential for them to get working solution fast.
Solution.
Whole development process took almost 4 months from start to finish. It was divided in a few stages:
Our team working with customer has selected 100x trademarks;
Each trademark was combined with 50x “similar” logos and images;
We managed to train our Neural Network on this dataset of 5,000 images;
Using trained Neural Network we were managed to work with existing database of 1.5 million trademarks. The system can identify logo that has visual similarities.
Result.
With all simplicity the project was quite a complicated one, as we need to create most of datasets on our own. The first iteration of algorithm has demonstrated next results:
82% accuracy in terms of identifying plagiarism;
Average “search per trademark” take between 5 to 7 seconds;
Most relevant images were shown in a first 100 logos results;
More than 5,000 hours of patent attorneys were saved.