Extraction and integration of web sources with humans and domain knowledge

Abstract

The extraction and integration of data from many web sources in different domains is an open issue. Two promising solutions take on this challenge: top down approaches rely on a domain knowledge that is manually crafted by an expert to guide the process and bottom up approaches try to infer the schema from many web sources to make sense of the extracted data. The first solutions scale over the number of web sources, but for settings with different domains, an expert has to manually craft an ontology for each domain. The second solutions do not require a domain expert, but high quality is achieved only with a lot of human interactions both in the extraction and integration steps.We introduce a framework that takes the best from both approaches. The framework addresses synergically both extraction and integration of data from web sources. No domain expert is required, it exploits data from a seed knowledge base to enhance the automatic extraction and integration (top down). Human workers from crowdsourcing platforms are engaged to improve the quality and the coverage of the extracted data. The framework adopts techniques to automatically extract both the schema and the data from multiple web sources (bottom up). The extracted information is then used to bootstrap the seed knowledge base, reducing in this way the human effort for future tasks.