Efficient Knowledge Extraction from PDF Documents Using Graph-Based Representations