Multi-Hop Reasoning Question Generation and Its Application

This article focuses on the topic of multi-hop question generation (QG), which aims to generate the questions requiring multi-hop reasoning skills by understanding the semantics of the given text fully. These questions not only have valid syntax but also need to be logically correlated to the answers. Concretely, we first design a basic QG model based on the sequence-to-sequence framework. In order to improve the syntactic quality of the results, we customize several techniques to regularize the model's output. We then extract a reasoning chain heuristically from the given text and use the evidential relations to promote the logical correlations of the results. Considering that different samples have their own characteristics on the aspects of text contextual structure, the type of question, and logical correlation, it is difficult for such a one-size-fits-all model to generate the best results flexibly. Thus, we propose a new adaptive meta-learner to optimize the basic QG model according to the specific characteristic of the evaluated case. Each case and its similar samples are viewed as a pseudo-QG task. The similar structural contexts contained in the same task can be used as guidance to fine-tune the model robustly and produce the optimal results accordingly. Since each sample contains the text, question, and answer, with unknown semantic correlations among them, we propose a data-driven multi-level recognizer to measure the similarity of such structured inputs. The experimental results on two typical data sets in various domains show the effectiveness of the proposed approach. Moreover, we apply the generated results to the task of machine reading comprehension and achieve significant performance improvements. That demonstrates the capacity of multi-hop question generation in facilitating real-world applications.