The aim of the project is to develop methods based on multimodal and multitasking machine learning and image analysis to extract all information relevant for real estate valuation. The task is to combine very heterogeneous data, structured and unstructured text data as well as images, proprietary and non-proprietary data (e.g. purchase contracts from the land register, images, geographical information, etc.), extract them and make them available for further analysis.
The successful modeling of complementary information across the various data sources (modalities) is the key to the robust extraction of real estate-related information and to achieving high data quality. The participating research institutions (University of Applied Sciences Kufstein Tirol and St. Pölten University of Applied Sciences) have launched the interdisciplinary research initiative "ImmoPixel", which deals with innovative approaches to the development of new real estate-related information sources through multimodal analysis, modeling and machine learning. The many years of experience of the academic partners in combination with the rich multimodal data of the partner Data Science Service (DSS) offers a unique opportunity to take automated real estate valuation to a new level.
