Towards Question-based Recommender Systems

Conversational and question-based recommender systems have gained increasing attention in recent years, with users enabled to converse with the system and better control recommendations. Nevertheless, research in the field is still limited, compared to traditional recommender systems. In this work, we propose a novel Question-based recommendation method, Qrec, to assist users to find items interactively, by answering automatically constructed and algorithmically chosen questions. Previous conversational recommender systems ask users to express their preferences over items or item facets. Our model, instead, asks users to express their preferences over descriptive item features. The model is first trained offline by a novel matrix factorization algorithm, and then iteratively updates the user and item latent factors online by a closed-form solution based on the user answers. Meanwhile, our model infers the underlying user belief and preferences over items to learn an optimal question-asking strategy by using Generalized Binary Search, so as to ask a sequence of questions to the user. Our experimental results demonstrate that our proposed matrix factorization model outperforms the traditional Probabilistic Matrix Factorization model.

Introduction. Online shopping on Internet platforms, such as Amazon, and eBay, is increasingly prevalent, and helps customers make better purchase decisions [44]. The high demand for online shopping calls for taskoriented conversational agents which can interact with customers helping them find items or services more effectively [34]. This greatly stimulates related research on conversational and questionbased recommender systems [34, 41]. Traditional recommender systems infer user preferences based on their historical behaviors, with the assumption that users have static preferences. Unfortunately, user preferences might evolve over time due to internal or external factors [31]. Besides, the quality of traditional recommendations suffers greatly due to the sparsity of users’ historical behaviors [33]. Even worse, traditional recommendation systems fail to generate recommendations for new users or new items, for which the historical data is entirely missing: the cold-start problem [33].

Discussion / Conclusion. In this paper, we propose a novel question-based recommendation method, Qrec, which directly queries users on the automatically extracted entities in relevant documents. Our model is initialized offline by our proposed matrix factorization model QMF and updates the user and item latent factors online by incorporating the modeling of the user answer for the selected question. Meanwhile, our model tracks the user belief and learns a policy to select the best question sequence to ask. Experiments on the Amazon product dataset demonstrate that the effectiveness of the Qrec model compared to existing baselines. In this work, the questions asked to users are based on the presence or absence of entities in the target items, following past work. Richer type of questions could be constructed by using other sources such as categories, keywords, labelled topics [47, 48], structural item properties, and domain-specific informative terms.