Abstractive Summarization: Graph Representations and Platform Integration
Text summarization has always been a compelling problem concerning natural language processing and machine learning. There exist two basic types of summarization: extractive and abstractive. Extractive methods identify relevant pieces of sub-text, subsequently concatenating them together and dropping the rest. Abstractive summarization on the other hand requires the machine to reason at a high level to generate an entirely new output summary.
Related to summarization is the concept of knowledge graphs. These are labeled graphs where nodes can represent ideas, concepts, people, etc., and edges describe relationships between nodes. These graphs can be queried for Q&A systems or be used to build up summaries, and in general are a form of memory that can be leveraged in all types of systems.
Our goals for this project are to:
- Generate a comprehensive model that abstractively summarizes text
- Visually represent the information in a high-level graph
- Implement integration to platform that real-world clients can use