Automatic Table Structure Identification and Content Interpretation

Tables are a widely-used structure for data presentation and summarization in documents. However, most of the tables are designed for human readers and their layout and logical structure are not well-defined for machine processing. This work focuses on designing table structure decoding systems and table content interpretation algorithms by analyzing various features within a complex table (e.g., layout, cell content, missing data, messy tables). The proposed method builds an offering to extract payment and rebate information from contract tables, and transfer it into AI/machine actionable data to identify performance gaps and cost savings opportunities.