カンファレンス (国内) 情報推薦における透明性がユーザに与える影響

日暮 立, 山口 修司, 坪内 孝太, 友成 愛, 大島 みゆき

2022年度 人工知能学会全国大会(第36回)


In this study, we clarified the effect of the transparency of information recommendation.In recent years, with the introduction of neural networks and the rise of nonlinear models in learning, attention has begun to focus on the dangers of entrusting decision-making to models that cannot be interpreted by the user, and the interpretability of models has come into focus.The interpretability of a model should be able to be interpreted from the main perspective of adopting the model and making a decision, and from the perspective of the user receiving recommendations from the model.The greater the interpretability of the model itself, the more clearly it will be able to communicate and present to the user "why this kind of information is recommended to me".No one has investigated the transparency of the model due to such interpretability, whether it has a positive impact on the user, and whether it contributes to improving the performance of the service.Therefore, to conduct a survey of users with reasons for information recommendation in the content of information recommendation, and to discuss the pros and cons of transparency in information recommendation.

Paper : 情報推薦における透明性がユーザに与える影響 (外部サイト)