Project Name

Web Intelligence

Member

T. Ohbe, H. Yamazoe

Purpose

Supporting various problems using Web intelligence

Sub Theme

Supports article writing support, report evaluation support, weekly report writing support

Keyword

Web mining, natural language processing, polar analysis, Deep Learning, information recommendation, social media analysis, information visualization, document summarization, automatic scoring

Outline

The purpose of this project is to support the effective use of nonstructured and enormous information resources on the Web. As part of this project, we are developing a paper writing support system. In the paper writing, the title of the paper needs to accurately represent the point of claim, and determination of the title is important. In order for the primary scholar to prepare the title of the paper, support such as presentation of title candidates is preferable. In this research, we aim to realize a system that generates title candidates from the abstracts of articles and supports the writing of papers. Assuming that the word of the title of the paper is included in the abstract, the task of generating the title from the abstract of the paper can be considered as a kind of document summary. Application of the Encoder-Decoder language model used for machine translation has been successful as a document summarization method using deep learning. In this research, we are studying the extraction type method using Recurrent Neural Network (RNN) and the generation type method applying the Encoder-Decoder language model. Moreover, in the generation type method, in addition to the ordinary Decoder, we are studying the title generation that can correspond to unknown words using Pointer Networks.

With report evaluation support. We aim to realize a system that aims to reduce the burden on report evaluation by targeting short sentence reports for confirming understanding of lecture words. Specifically, focusing on the characteristics of the report, it has a function of semi-automatically guessing a report of an incorrect answer, a function of filtering a report that is not subject to evaluation, and a function of filtering reports based on the submitted report group, It is a system with the function of generating an average report answer. In addition, we implement these functions as an application that is easy to use for these functions and evaluators.