With the growing availability of full-text scientific articles, how text mining researchers utilize them has become an important issue. Although abstract and title provide accurate and summary information of article, lots of details are inevitably lost for its short space. The primary goal of the study is to utilize the advantages of abstract and full-text to ease the burden of reading. Finding essential information from abstract, using this to search and to rank paragraphs in full-text, the proposed approach recommends significant paragraphs to user for saving time of perusing whole article. Finally we evaluated the performance of our system, it outperformed the baseline approach both in human ratings and ROUGE scores.