Elaborative Trademark Similarity Evaluation Using Goods and Services Automated Comparison

Abstract

Trademark description comparison is an important problem during registering a new trademark. Manual search for the most similar trademark often faces the plenty of similar trademarks already registered and filed, so subjective evaluation of the trademark similarity imposes some deviations in the comparison. This paper discusses the methods that aim the automatic evaluation of the trademark’s similarity/identity based on the text description of goods and services. The analysis included trademarks covering legal services. Various models (count vectorizer, TF-IDF, word2vec, doc2vec) pretrained and trained by ourselves, and sets of vectors comparison approaches (cosine similarity, Tanimoto similarity) have been evaluated. Own simple greedy matching algorithm that can optionally include difference between size of sets which is useful for trademarks comparison has been proposed. The article shows that some combinations of model/similarity measure are successful in searching for correct hierarchy of trademarks, while some other are better in terms of human relevance. The quality and performance analysis of the discussed methods allowed us to choose the most effective of them according to the available data and technical requirements and implement them in practical applications to perform search for the similar or identical trademarks by example. We contribute to the study of methods for automating legal processes and offer a mean to reduce the risk of intellectual property disputes.