Attention based bidirectional GRU hybrid model for inappropriate content detection in Urdu Language

Abstract

With the advancement in the scope of online discussion, the spread of toxic and inappropriate content on social networking sites has also increased. Several stud ies have been conducted in different languages. However, existing literature on inappropriate content detection lacks research in Urdu Unicode text language us ing deep learning techniques. Use of attention layer with deep learning model can help in handling the long-term dependencies and increase its efficiency. To explore the effect of attention layer, this study proposes an attention based Bidi rectional GRU hybrid model for identification of Inappropriate content in Urdu Unicode text language. Four different baseline deep learning models LSTM, Bi LSTM, GRU, and TCN are used to evaluate the performance of proposed model. The results of models are compared based on evaluation metrics, dataset size and impact of word embedding layer. The pre-trained Urdu word2vec embeddings are utilized for our case. Our proposed model BiGRU-A outperformed all other base line models by yielding 84% accuracy without using pre-trained word2vec layer. From our experiments we have established that attention layer improves the ef ficiency of model and pre-trained word2vec embedding does not work well with inappropriate content dataset.