A Cross-Media Method of Stakeholder Extraction for News Contents Analysis

摘要

We are studying contents analysis of multimedia news as a solution to the issue of bias, which multimedia news, as a reflection of the real world, is also facing. For the contents analysis, we use a stakeholder model representing descriptions of different stakeholders, which are defined as the main participants in an event. In this paper, we propose a method of detecting stakeholders as the core component of the stakeholder-oriented analysis. In our work, a stakeholder is assumed to appear in the video clips and be mentioned in the closed captions frequently. Given a series of video clips and their closed captions reporting the same event, we extract stakeholder candidates from both textual and visual descriptions. After that, we calculate the degree of exposure for each candidate to identify stakeholders. We also present experimental results that validate our method.