Multimedia Tools and Applications Программное обеспечение
DeepInsight: a CNN-based approach for machine reading comprehension in query answering systems and its applications
Аннотация
Understanding and reading an unstructured text to answer queries on it, also called Machine Reading Comprehension (MRC), has been a hot topic amongst researchers worldwide in the past few years. There are many ways in which this problem has been tackled; each one has its perks and pitfalls. MRC is being thought about meticulously because of its various applications which are a pressing need due to ever-increasing data in this modern world. However, no work has been done focusing on the applications of MRC. We propose DeepInsight, an efficient CNN-based machine reading comprehension model inspired by QANet and an assemblage of three case studies. Each case study is about a specific application of DeepInsight, alongside the details of its implementation, results from the analysis, and shortcomings. These case studies can be referred to by developers around the world while building similar applications.The first case study is a fully implemented end-to-end android-based application that can upload documents to a server and receive different queries. Second, a mobile application that can help people with visual impairment to comprehend documents. Third, a video query application that can answer questions posed on video data, using the deep captioning model as a core for this application. DeepInsight has an EM/F1 score of 77.0 / 86.3 and performs better than the present state-of-the-art models while keeping the inference time of 0.535 seconds, justifiable for real-world applications.