Defining 2D-Trust Levels with Part-of-Speech Information: A Case Study with Rescue Requests

Abstract

In this paper, our goal is to extract rescue requests from the messages posted on SNS (X, formerly Twitter) during the Noto Peninsula earthquake in 2024. Specifically, we obtain two-dimensional trust values of trustworthiness and distrustworthiness from the message text and then convert them to Marsh et al.’s (one-dimensional) trust values. When the trust value exceeds the “trust threshold,” the message is regarded as a genuine rescue request. The two-dimensional trust value applies Oda’s Fuzzy-set Concurrent Rating (FCR) method to the information security field. The trustworthiness can be calculated from the degree to which privacy information (specifically, complete address information) appears in the text. In this paper, we propose a calculation method for the degree of distrust that focuses on the occurrence of specific parts of speech. Our findings reveal that messages in which the imperative form of a word appears without the hypothetical form of a word were more likely to be genuine requests for help. This paper also discusses a method to define the distrust level as a cumulative value.