Complex question answering & generation over knowledge graphs

Complex questions are those that involve discrete, aggregate operators that operate on numbers (min, max, arithmetic) and sets (intersection, union, difference). Recent advances in complex question answering take a neural-symbolic approach and combine meta-learning and reinforcement learning techniques [1,2,3]. On the other hand, the generation of complex questions, the dual problem, is less explored. Recent works on knowledge graph question generation [4,5] have mainly focussed on multi-hop questions.

This project aims at developing novel methods that jointly address the challenging, dual problem of answering and generation of complex questions over knowledge graphs.

1. Hua, Yuncheng; Li, Yuan-Fang; Haffari, Reza; Qi, Guilin.
   Few-shot Complex Knowledge Bases Question Answering via Meta Reinforcement Learning.
   Conference on Empirical Methods in Natural Language Processing (EMNLP'20).

2. Hua, Yuncheng; Qi, Daiqing; Zhang, Jingyao; Qi, Guilin; Li, Yuan-Fang.
   Less is More: Data-efficient Complex Question Answering over Knowledge Bases.
   Journal of Web Semantics, 2020.

3. Hua, Yuncheng; Li, Yuan-Fang; Haffari, Reza; Qi, Guilin; Wu, Wei.
   Retrieve, Program, Repeat: Complex Knowledge Base Question Answering via Alternate Meta-learning.
   29th International Joint Conference on Artificial Intelligence (IJCAI 2020).

4. Bi, Sheng; Chen, Xiya; Li, Yuan-Fang; Wang, Yongzhen; Qi, Guilin.
   Knowledge-enriched, Type-constrained and Grammar-guided Question Generation over Knowledge Bases.
   28th International Conference on Computational Linguistics (COLING'20).

5. Kumar, Vishwajeet; Hua, Yuncheng; Qi, Guilin; Ramakrishnan, Ganesh; Gao, Lianli, Li, Yuan-Fang.
   Difficulty-controllable Multi-hop Question Generation From Knowledge Graphs.
   18th International Semantic Web Conference (ISWC'19), Springer.