Generating tailored summary of scientific paper from Rhetorical Document Profile

The purpose of reading a scientific paper is to look for specific information. Generating Rhetorical Document Profile (RDP) of a scientific paper has yielded a frame of important information of the paper. In the RDP, all sentences has been grouped into fifteen rhetorical slots stating the intention to be conveyed to the reader by the author, such as aim, other related work, method, previous research, etc. However, RDP in the form of frame is not user friendly to be understood easily by various end user. A paper summary could be a solution. RDP can be processed to provide a tailored summary, the summary in accordance with the reader´s task, background, and the length of the summary. Therefore, this study aims to build a system that can generate tailored summary from RDP. Generating tailored summary relies heavily on building plan that contains the information template of a summary based on the needs defined by the user. A building plan contains the number of sentences and rhetorical slots of the sentences that will be compiled to be a summary. Because this study uses RDP with AZ15 scheme, building plan is defined by adapting the building plan designed to RDP with AZ7 scheme. To obtain candidates of summary sentences, Maximal Marginal Importance - Multi Sentence (MMI-MS) is applied to select rhetorical sentences for each slot contained in the building plan. All candidates are arranged according to the order of the sentences in original paper. Last, surface repair uses template-based paraphrasing to generate a comprehensible summary. This system is also flexible with the additional facilities for defining the building plan by the Tailored summary is very capable of generating summary based on the users needs. From the evaluation done in this paper, MMI-MS gives a good result in doing sentence selection. The best result of sentence selection evaluation has 0.61 recall and 0.61 precision. The result is given by tf and idf weigthing method. And the result of surface repa- - ir acceptance is 93%.