A proposal for an information retrieval system that provides diverse perspectives

Abstract

The recent development of search algorithms and recommendation functions, as well as the rise of social media, has caused the problem of shaping users' thinking in a biased direction. However, it is difficult to say that society as a whole is tackling the proposal of system methodologies, with the exception of some research. In this study, we propose a search system that provides diverse viewpoints by using community extraction in complex network analysis and deep learning. The system has a function to display objective contextual information, information relationships and bird's-eye view information. Context information is extracted from sentences using two LSTM-based models, with an average of 80% and 74% correct answers, and the Louvain method and cos-similarity are applied to extract communitiesis, achieving high cluster coefficients and modularity. In an evaluation experiment, 70% of the total number of subjects indicated that they felt a variety of viewpoints in the whole system.