Text Generation for Hindi

There is a rapid increase of multimodal data such as texts, images, etc. all over the internet and it is a strenuous task to obtain relevant and vital information from it. Hence, it became a challenging task to process this multimodal data for information retrieval. Thus, language generation model would be helpful for generating precise summary. In the present systems, language generation of textual data is majorly focused, but it’s been observed that images can also contribute to the language generation process. Hence, the proposed model is built on an integrated system to process texts and images of both English and Hindi language data and generates concise summary. Text rank algorithm and Term Frequency-Inverse Document Frequency algorithm is used for text generation for both Hindi and English Language. Additionally, a Neural Image Captioning Model is used to generate captions of images which are then correlated with the textual data for Language Generation. The approach of combining TF-IDF algorithm and Text rank algorithm with Image Captioning model aggregately was effective to generate precise summary while considering the information captured by the images. The text generation model built in this research is language independent and paved a way as a use case in generating text for smart composition of sentences, Question-Answering system on various websites, Search Engine Optimization, etc.