10 April 2025 in ACM Transactions on Recommender Systems
In recommender systems, explanations are essential for supporting users’ decision-making processes. While many studies have focused on explanation content or user interface, the expression of textual explanations has been largely overlooked. The expression refers to textual styles such as formal or humorous, which we call tone in this article. Although tone contributes to smooth human communication, its impact on users’ perceptions of recommender systems remains largely unexplored. In particular, it is unclear whether the perceived effects of explanation tone differ by domain or user attributes. Therefore, we investigate the effects of explanation tones through two online user studies considering domains and user attributes. In the first study with 470 participants, we generated datasets using a large language model to create fictional items and explanations with six tones across three domains: movies, hotels, and home products. The participants evaluated two explanations for an item, each presented in a different tone, and rated 10 metrics. In the second study with 103 participants, we used a real-world dataset from the hotel domain and incorporated a simple personalized recommender system to examine effects of tone in a more realistic setting. The results revealed that the perceived effects of tones differ by domain and are significantly influenced by user attributes such as age and personality traits. Our findings suggest that appropriately adjusting the tone of explanations according to domains and user attributes can enhance the perceived effects of recommender systems.
1 April 2019 in IEEE/ACM Transactions on Computational Biology and Bioinformatics
Engineering stable proteins is crucial to various industrial purposes. Several machine learning methods have been developed to predict changes in the stability of proteins upon single point mutations. To improve accuracy of the prediction, we propose a new unsupervised descriptor for protein sequences that is based on a sequence-to-sequence (seq2seq) neural network model combined with a sequence-compression method called byte-pair encoding (BPE). Our results exhibit that BPE can encode a protein sequence into a sequence of shorter length, thereby enabling efficient training of the seq2seq model. Furthermore, we implement a basic predictor using the proposed descriptor, and our experimental results demonstrate that the predictor achieved state-of-the-art accuracy in case of tests for proteins that are not included in the training data.
1 July 2024 in IEEE Access
Preferential Bayesian optimization (PBO) is a framework for human-in-the-loop optimization to maximize black-box human preference functions such as seeking perceptually good visual designs. It is advantageous when consistently providing a degree of preference is challenging, but selecting the best option among multiple choices is relatively more straightforward. However, similar to conventional BO methods, PBO suffers from high dimensional problems, that is, finding a good solution becomes increasingly difficult as the dimensionality of the search space increases. In this paper, we focus on slider-based PBO (S-PBO), which is a variant of PBO generating a set of candidates aligned on a one-dimesional line segment, and propose novel acquisition strategies for high dimensional problems. Specifically, we propose two acquisition functions that have a different balance of local/global search abilities. In addition, we propose a simple yet effective randomized strategy that balances the local/global search abilities provided by the two proposed acquisition functions. Through empirical evaluation, we assess the effectiveness of our proposed strategy in improving high-dimensional problems.
6 May 2020 in Lecture Notes in Computer Science