This paper presents a tool to debug knowledge base (KB) QA system to visualize steps to simplify error analysis. Knowledge base is divided into three steps: entity linking (1), semantic parsing (2) and answer retrieval (3). In the first step, entity is identified and linked to KB. This the most error generating step. Then in second step, entities are parsed semantically to connect with probable answers. The designed tool, which is comprised of back and front end, sends seperate requests to each steps. Front end helps to visualize each steps. they used existing tool to extract KB and used CNN to learn vector encoding for QA and semantic representation. Their interface shows entities, semantic graphs. This way it is possible to evaluate and tune models easily to correct steps of the model.
This paper presents a tool to debug knowledge base (KB) QA system to visualize steps to simplify error analysis. Knowledge base is divided into three steps: entity linking (1), semantic parsing (2) and answer retrieval (3). In the first step, entity is identified and linked to KB. This the most error generating step. Then in second step, entities are parsed semantically to connect with probable answers. The designed tool, which is comprised of back and front end, sends seperate requests to each steps. Front end helps to visualize each steps. they used existing tool to extract KB and used CNN to learn vector encoding for QA and semantic representation. Their interface shows entities, semantic graphs. This way it is possible to evaluate and tune models easily to correct steps of the model.