Abstract
In social media, code mixed data is getting exceptionally mainstream due to which there is an enormous development in noisy and inadequate multilingual content. Automation of noisy social media text is one of the existing research areas. This work focuses on extracting sentiments at word level for movie-related code mixed English-Telugu bilingual Roman script data extracted from Twitter. Initially, data was cleaned and slang words were replaced. Next, named entities and language of each word were identified. Further Telugu words were back transliterated to the native script. Finally, the text was classified either positive, negative or neutral sentiments based on the count of corresponding positive and negative lexicons present in each sentence and achieved an average accuracy of 79.9%.