Review-based Question Generation with Adaptive Instance Transfer and Augmentation

Abstract

While online reviews of products and services become an important information source, it remains inefficient for potential consumers to exploit verbose reviews for fulfilling their information need. We propose to explore question generation as a new way of review information exploitation, namely generating questions that can be answered by the corresponding review sentences. One major challenge of this generation task is the lack of training data, i.e. explicit mapping relation between the user-posed questions and review sentences. To obtain proper training instances for the generation model, we propose an iterative learning framework with adaptive instance transfer and augmentation. To generate to the point questions about the major aspects in reviews, related features extracted in an unsupervised manner are incorporated without the burden of aspect annotation. Experiments on data from various categories of a popular E-commerce site demonstrate the effectiveness of the framework, as well as the potentials of the proposed review-based question generation task.

Pretrained Transformers for Text Ranking: BERT and Beyond

“…The expanded collection can be indexed and used exactly as before-either by itself or as part of a multi-stage ranking architecture. Perhaps due to its simplicity and effectiveness, doc2query has been adapted and successfully applied to other tasks, including scientific document retrieval [Boudin et al, 2020], creating artificial in-domain retrieval data [Ma et al, 2020], and helping users in finding answers in product reviews [Yu et al, 2020b].…”

Learning to Diversify for Product Question Generation

“…Zhang et al [25] have proposed KPCNet -a seq2seq model that attended on selected product description keywords to improve question specificity. Several other works [13,20,22] have utilized adversarial learning (e.g., GANs) to improve question generation "quality" by using an additional discriminator model. Yet, training such a discriminator requires additional labeled data with question answers.…”

Evaluating BERT-based Rewards for Question Generation with Reinforcement Learning

“…Some works [7,11,26,46,58] take QG as a generator to harvest question-answer pairs from passages, and use this harvested data to pre-train QA models, which subsequently resulted in improved QA model effectiveness. QG is also widely used in IR tasks, such as improving search system effectiveness by generating clarifying questions [57], or generating questions from e-commercial customers reviews [55].…”

“…Question generation (QG) systems aim to generate natural language questions that are relevant to a given piece of text (the so-called context-typically a sentence or a paragraph), and can usually be answered by just considering the context. As an important natural language processing task, QG can be used to improve questionanswering [11,58], conversational systems [48], and information retrieval (IR) [55,57]. As a concrete example of the latter, QG has been employed to improve the retrieval effectiveness of search systems by expanding documents with generated questions that the document might answer [35,36].…”