LFKQG: A Controlled Generation Framework with Local Fine-tuning for Question Generation over Knowledge Bases

Proceedings of the 29th International Conference on Computational Linguistics, pages 6575–6585 October 12–17, 2022. 6575 LFKQG: A Controlled Generation Framework with Local Fine-tuning for Question Generation over Knowledge Bases Zichu Fei1,2, Xin Zhou1,2, Tao Gui3, Qi Zhang1,2, Xuanjing Huang1,2,3 School of Computer Science, Fudan Unviersity1 Shanghai Key Laboratory of Intelligent Information Processing, Shanghai, China2 Institute of Modern Languages and Linguistics, Fudan University, Shanghai, China3 {zcfei19, xzhou20, qz, tgui,xjhuang}@fudan.edu.cn Abstract Question generation over knowledge bases (KBQG) aims to generate natural questions about a subgraph that can be answered by a given answer entity. Existing KBQG models still face two main challenges: (1) Most models often focus on the most relevant part of the answer entity, while neglecting the rest of the subgraph. (2) There are a large number of out-of-vocabulary (OOV) predicates in real- world scenarios, which are hard to adapt for most KBQG models. To address these challenges, we propose LFKQG, a controlled generation framework for Question Generation over Knowledge Bases. (1) LFKQG employs a simple controlled generation method to generate the questions containing the critical entities in the subgraph, ensuring the question is relevant to the whole subgraph. (2) We propose an optimization strategy called local fine-tuning, which makes good use of the rich information hidden in the pre-trained model to improve the ability of the model to adapt the OOV predicates. Extensive experiments show that our method outperforms existing meth- ods greatly on three widely-used benchmark datasets SimpleQuestion, PathQuestions, and WebQuestions . 1 Introduction Question generation (QG) aims to endow machines with the ability to ask relevant and to-the-point questions for a given form of data such as text (Du et al., 2017a; Song et al., 2018a), image (Li et al., 2018), table (Bao et al., 2018) and knowledge bases (KB) (Elsahar et al., 2018). KBQG aims to generate natural language questions given a subgraph in the KB, i.e. a set of connected triples of the form <subject, predicate, object>. KBQG is an effective approach to generating high- quality QA pairs that can significantly address the data scarcity issue for Knowledge Base Question Answering (KBQA). In addition, KBQG can be applied for educational purposes by producing Figure 1: The examples of two main challenges in KBQG where the questions are generated from the subgraph and the red texts are the answer entity. Q1 only focus on the second triple but neglect the first.Q2 does not contain any semantic of discoverer or inventor, which is an OOV predicatge. practice assessments (Heilman and Smith, 2010) and can help dialog systems have more engaging conversations (Mostafazadeh et al., 2016). Current KBQG systems follow an attention- based sequence-to-sequence structure (Elsahar et al., 2018; Kumar et al., 2019). To make use of the rich structure information hidden in the subgraph, (Chen et al., 2020) proposes the graph- to-sequence framework. However, these models face two main challenges: (1) Most models often focus on the most relevant part of the answer entity, while neglecting the rest of the subgraph (Bi et al., 2020; Chen et …

Content truncated to 500 words