A text summarization system generates short and brief summaries of original document for given user queries. The machine generated summaries uses information retrieval techniques for searching relevant answers from large corpus. This research article proposes a novel framework for generating machine generated summaries using reinforcement learning techniques with Non-deterministic reward function.  Experiments have exemplified with ROUGE evaluation metrics with DUC 2001, 20newsgroup data. Evaluation results of proposed system with hypothesis of automatic summarization from given datasets prove that statistically significant improvement for answering complex questions with f- actual vs. critical values.

Document summarization is capable of generating user query relevant, precise summaries from the original document for user needs. To reduce the response time summary generation, QA corpus is built for similar questions and answer with help of learning model. It has been trained and tested by Quora duplicate and Yahoo! Answer datasets. The large QA corpus has been dynamically clustered with semantic features paves a way for efficient document’s retrieval. Answers are produced from datasets or generate summaries for unanswerable from the available sources. Results obtained from statistical significance test with hypothesis testing and evaluation with standard metrics proves the significant improvement in generating text summarization using QA corpus. The outcome is better in the producing close proximity of answers for the given user query.