Evaluation of natural language processing algorithms used for knowledge graph generation
About the Project
A Knowledge Graph (KG) represents knowledge as a network that relates entities such as physical objects and concepts. In a KG, the objects and concepts are represented as nodes, while their relations are represented as edges of the graph. Due to the graph structure of this knowledge representation, a KG can be a potent mechanism for searching entities and their relations, discovering new information and enabling complex decision making.
The process of updating and expanding a KG can be manual and laborious and require ongoing collaboration with subject matter experts. Natural language processing (NLP) and text-mining algorithms could streamline this process by automatically extracting essential information about objects, concepts, and their relations. Once deployed, these algorithms could populate nodes and edges of a KG in an automated fashion based on incoming data.
Two crucial processes in generating nodes and edges of the KG are Named Entity Recognition and Linking (NER/NEL) and Relation Extraction (RE). NER/NEL automatically analyzes free text to recognize character substrings that identify entities of interest, while RE identifies entity pairs with certain relations.
As NER/NEL and RE are necessary steps in generating nodes and edges of the KG, it is critically important that these algorithms accurately extract entities and their relations from free text. To quantify performance of NER/NEL and RE, benchmarking datasets are often used. The entity and relation labels in these datasets are either annotated manually by subject matter experts or algorithmically using semi-supervised and/or rule-based approaches.
The focus of this project lies in identifying open-source NER/NEL and RE algorithms that show the highest performance in recognizing such entities as “company”, “molecule”, “protein” and their relations. A critical aspect of the work will be in generating a benchmarking dataset using publicly available data.
- Yujie Li
- Caroline He
- Sammy Suliman
- Safiya Alavi
- KunXiao Gao
- Maxim Ivanov, Amgen
- Bonnie Jin, Amgen
- Erika McPhilliips, UCSB
- Yan Lashchev, UCSB
Amgen is committed to unlocking the potential of biology for patients suffering from serious illnesses by discovering, developing, manufacturing and delivering innovative human therapeutics. This approach begins by using tools like advanced human genetics to unravel the complexities of disease and understand the fundamentals of human biology.
Our belief—and the core of our strategy—is that innovative, highly differentiated medicines that provide large clinical benefits in addressing serious diseases are medicines that will not only help patients, but also will help reduce the social and economic burden of disease in society today.
Amgen focuses on areas of high unmet medical need and leverages its expertise to strive for solutions that improve health outcomes and dramatically improve people's lives. A biotechnology innovator since 1980, Amgen has grown to be one of the world's leading independent biotechnology companies, has reached millions of patients around the world and is developing a pipeline of medicines with breakaway potential.