Machine Learning and Context-Based Approaches to Get Quality Improved Food Data

Abstract

The provision of high-quality food data presents challenges for developers of health apps. There are no standardized data sources with information on all food products available in Europe. Commercial data sources are expensive and do not allow long-term storage, whereas open data sources from communities often contain inconsistent, duplicate, and incomplete data. In this thesis, methods are presented to load data from multiple sources via extract, transform, and load process into a central food data warehouse and to improve the data quality. Data profiling is used to detect inconsistencies and duplicates. With the help of machine learning methods and ontologies, data is completed and checked for plausibility using similar datasets. Via a specific API, an usage context can send to the central food data warehouse together with the search word to be queried. The API send a response with the food data results which were checked based on the context and provides further information as to whether the quality of the result data is sufficient in the respective context. All developed methods are tested using linear sampled test data.