Extracting Database Data from Free Text Messages

One potential mechanism for capturing data is their input in the form of simple free text messages, such as electronic mail. The goal is to seek statements concerning data which should be in the database and to generate database update statements from these.

The approach taken uses pattern matching. A series of template sentences are provided, each parameterised by meta-data and data placeholders. From each of these is generated one pattern for each combination of metadata and these patterns are then sought in the message. The data is matched against the context built up from previous sentences. When found an SQL update is produced.

Example

• Template: "The <PropName> of this <TypeName> is <<PropValue>>.”
• Pattern: "The date of this book is <DateValue>.”
• Recognised Sentence: "The date of this book is 1816.”
• Update: "Update Book set date = 1816 where ISBN = '123-456-7890'”