In educational settings, teachers often need to formulate some reading comprehension questions for a passage instead of a single sentence. Thus, how to obtain and exploit an answer-aware and context-aware passage-level representation is a challenge.

To solve the problems, a research team led by Qi LIU published their new research on 15 October 2024 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.

The team proposed a hierarchical answer-aware and context-aware network for question generation that focuses on how to construct and exploit the passage-level representation for generating questions. Compared with the existing research results, the proposed method can effectively utilize passages for generating questions.

In the research, the network mainly contains two modules, i.e., Hierarchical Passage Encoder (HPE) and Hierarchical Passage-aware Decoder (HPD). Specifically, HPE is composed of two-level answer-aware and context-aware attention with multi-hop reasoning, which aims to generate a better answer-aware and context-aware passage-level representation. Meanwhile, HPD is composed of a passage-aware decoder and a three-way copy mechanism to determine how much passage-level information is needed during decoding and copies rare words from the answer-specific sentence or the passage.

Extensive experimental results and case studies demonstrate that the hierarchical answer-aware and content-aware network proposed in this paper can effectively utilize passages for generating questions.

For the future direction, one is to expand to multilingual questions. Another is to optimize for different types of questions across various subjects.

DOI: 10.1007/s11704-023-2246-2