Effective Identification of Distinctive Wordmarks
Sujatha Das Gollapalli, Jung‐Jae Kim · Companion Proceedings of the Web Conference 2020 · 2020
A wordmark or a logotype string is a stylized text used by companies or businesses for the purposes of identification and branding products. Pie face under the product category “baked goods and pastries” and airbus smarter fleet under the product category “data processing and aircraft maintenance software” are examples of wordmarks. The wordmark strings are manually examined by patent officers for their distinctiveness and uniqueness to be accepted as “protected intellectual property” for specific businesses. We address the problem of automatically identifying acceptable English wordmarks based on their textual content. Different from most text mining tasks, our problem involves the classification of a small set of words (often less than five) in context of a specific product description. We handle this sparsity challenge by designing a range of features for characterizing wordmarks using the syntactic and linguistic properties of words as well as by incorporating co-occurrence and similarity information from external resources such as WordNet, Wikipedia, and Word Embeddings. We investigate machine learning models for this novel task and study their classification effectiveness on a large dataset of about 71K wordmarks.