Today’s era is of asking questions Why, What, and How. There are many such application like Google, Quora, Stack overflow where we ask a variety of questions to find out satisfying answers. But due to huge collection of records in database data will become inconsistent to find out relevant output. This also leads to problem of duplication. Paraphrasing techniques are emerging methods in natural language processing (NLP). These paraphrasing techniques are used in a variety of applications. One such application is the question answering system. In this application many people ask same type of questions again and again. Sometimes we already have an answer stored into database for the asked question. But system cannot detect it and creates a new answer instead of searching in the database. So, to perform such activity it consumes a lot of time. To overcome this problem, we have proposed a system that will predict whether the asked question is duplicate or not using deep neural network and Universal Sentence Encoder (USE) embedding technique in natural language processing (NLP). The model is evaluated by considering specific epochs and the accuracy of the model implemented is 91.12% which is improvement over the existing systems.