Factuality Analysis of SNS Posts Containing Diverse Symptom Expressions for Public Health Surveillance | IEEE Conference Publication | IEEE Xplore

Factuality Analysis of SNS Posts Containing Diverse Symptom Expressions for Public Health Surveillance


Abstract:

We are developing a system that observes and analyzes the prevalence of infectious diseases and the onset of various symptoms, and visualizes them by prefecture and in ch...Show More

Abstract:

We are developing a system that observes and analyzes the prevalence of infectious diseases and the onset of various symptoms, and visualizes them by prefecture and in chronological order, as one of public health surveillance. The proposed system consists of a factuality analysis module that determines whether SNS posts containing symptoms are intended to represent the onset of the poster’s own symptoms, and a location estimation module that estimates the place of residence of users who have made posts related to diseases or symptoms. In this paper, we propose SVM, logistic regression, multilayer perceptron, and PubMedBERT models for factuality analysis of SNS posts containing 86 symptom expressions in 11 standard disease names (disease symptoms) defined in a dictionary consisting of expressions used by patients. From the experimental results to compare with the performance of each model, the average of the best F1 scores of each model for 11 disease symptoms were 0.667, 0.663, 0.681, and 0.721, respectively. We confirmed that the PubMedBERT model has the best classification performance for 9 of the 11 disease symptoms.
Date of Conference: 16-18 December 2023
Date Added to IEEE Xplore: 09 February 2024
ISBN Information:
Conference Location: Changsha, China

References

References is not available for this document.