Named entity extraction for food safety events monitoring

• Main Author: Lee, Zheng Han Nicholas
• Other Authors: Hui Siu Cheung
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/172106
• Institution: Nanyang Technological University

Due to the dynamic nature of the food supply ecosystem and the rising reports on food-borne disease outbreaks, effective food safety risk monitoring has become crucial for public health and socioeconomic development. The Singapore Food Agency (SFA) oversees the food safety and security of Singapore. Part of its functions is to monitor and detect potential food-borne incidents. Currently, SFA scientists rely on manual efforts to gather relevant information from various internal datasets to conduct food safety analysis, risk assessments, and decision making. This is not only time-consuming but also susceptible to human errors. To tackle this problem, SPIRIT NTU and SFA have collaborated to develop a predictive food safety risk monitoring system called SFA-FSM. SFA-FSM applies advanced analytics and AI technologies for monitoring online food safety news and local customer feedback. This thesis proposes a Named Entity Extraction for Food Safety Monitoring (NEE-FSM) framework as a crucial component for information extraction to support downstream functionalities of the SFA-FSM system. The NEE-FSM framework introduces a hybrid approach consisting of three modules: the Heuristic Entity Extraction (HEE) module, which constructs base dictionaries of commonly known food safety named entities from various knowledge resources; the Named Entity Recognition (NER) module, which iteratively augments and enhances the base dictionaries with new named entities from the food safety datasets; and the Named Entity Matching (NEM) module, which performs dictionary-based named entity extraction using the augmented dictionaries. This thesis focuses on discussing the various techniques to incorporate publicly available knowledge resources to effectively model the components of the NEE-FSM framework in low-resource settings. The proposed framework has shown to be effective for food safety named entity extraction, as demonstrated by the performance results of our experiments.