Hierarchical Types Constrained Topic Entity Detection for Knowledge Base Question Answering

The Architecture of HTTED


    09/2017 - 12/2017


    Yunqi Qiu, Manling Li, Yuanzhuo Wang, Yantao Jia, Xiaolong Jin, Xueqi Cheng



    Knowledge Base Question Answering (KBQA) aims to answer questions with simple facts in KB. Generally, KBQA systems have two key components: (1) topic entity detection, which detects topic entities in questions and links them to KB; (2) answer selection, which identifies KB relations linked to topic entities and selects correct answers. Topic entity detection is the foundation of getting correct answers. However, traditional methods ignore the information of entities, especially entity types and their hierarchical structures, restricting the performance.


    Proposed Hierarchical Types constrained Topic Entity Detection (HTTED) to increase topic entity detection accuracy. It is a neural model to match questions and entities by learning the representation of question context, entity hierarchical types and entity relations.

My Work

  • Proposed to formulate the representation of parent types as the composition of child types to model the interplay of child and parent types. This formulation enables the model to share parameters between parent and child types, so that it helps learn embeddings of child types, which suffers from dearth of training data.
  • Co-authored a short paper submitted to WWW 2018.