Towards Defining Trust And Distrust Levels for Disaster Relief Messages based on Parts of Speech

Abstract

We examine the posts on SNS (X, former Twitter) during the Noto Peninsula earthquake in 2024 and attempt to judge disaster relief requests based on the occurrence of specific parts of speech. Specifically, we extracted messages that were candidates for disaster relief requests based on the occurrences of privacy information (precisely, the inclusion of complete address information). Then, we classified messages based on hypothetical and imperative forms of words. Messages in which the imperative form of the word appeared without the hypothetical form of the word tended to contain messages that seemed to be genuine rescue requests. Based on the results, we investigate how to define two-dimensional trust values (trustworthiness/distrust-worthiness pairs), particularly how to define distrust-worthiness as a cumulative value.