Class Strength V2: An Adaptive Multilingual Tool for Tweet Classification

摘要

In this paper we present the second version of our multilingual tweet classification tool. ClassStrength v2 classifies tweets into 14 categories (Sports, Music, News&Politics etc.) using a distant supervision approach. The new version extends the initial set of five languages to ten (English, French, German, Chinese, Japanese, Arabic, Russian, Spanish, Portuguese and Polish). In addition, the classification models of each language get automatically updated every month to allow accurate classification over time. Our experimentation showed that the larger the time gap between the tweet and the data used for training the model, the worse the performance, which motivated for creating an adaptive version of ClassStrength that get its models updated periodically.