Effect of the training set on the word embeddings and similarity test set for Turkish

Word embedding, which is usually used in the literature especially for English, is a technique to associate each word to a mathematical vector representation under which some structural or semantic relations hold. There are some Turkish application of this technique. Despite being designed according to English, it is also satisfactory for Turkish. In this study, the performance of Turkish word embeddings is analysed against the convenience of the data to the goal of the embedding. For this study, a new test set based on subject similarity in Turkish is introduced. This set is used to measure the performance of the word embeddings. This set will be publicly available for academic purposes2. A subject classifier, which beats the state of the art performance, for Turkish labeled text corpus is also proposed.