Bengali Context–Question Similarity Using Universal Sentence Encoder

In natural language, the similarity between the two texts is judged by their similarity score. Some of the recent NLP application such as text summarization, question answering, text generation, and text mining are depended on the machine provided text. Accuracy of response text is measured by the similar with corresponding text or human given text. Comparing by two texts and measuring the similarity defines that the two texts are lexical or semantically similar. If two texts are related to each other with the word or character, this text is lexically similar. Also, if the texts are related in meaning but not in word or character level that are semantically similar. In this research, we measure the similarity of context and question for your question answering system. Then we find the most similar answer for the corresponding question. We used universal sentence encoder for embedding and measure the similarity using cosine distance of the text. We used deep averaging network for find the best similar text. For evaluation of similarity model, we calculate the Pearson correlation value for our dataset and achieve 0.41 coefficient.