سامانه همایش ها

صفحه اصلی / دومین کنفرانس ملی عصر انفجار تکنولوژی؛ هوش مصنوعی، تحولی در صنعت، تجارت و زنجیره تامین و دومین کنفرانس ملی علم داده در کاربردهای مهندسی

Recent Advances and Open Challenges in Explainable AI for Deep Learning-based Recommender Systems

نویسندگان :

Narjes Badpar¹ Azita Shirazipour² Seyed Javad Mirabedini³

1- دانشگاه آزاد اسلامی واحد تهران مرکزی 2- دانشگاه آزاد اسلامی واحد تهران مرکزی 3- دانشگاه آزاد اسلامی واحد تهران مرکزی

کلمات کلیدی :

Recommender system (RS)،Deep learning (DL)،Explainable Artificial Intelligence (XAI)،Deep learning-based recommender systems (DL-RS)،Machine Learning (ML)

چکیده :

The implementation of deep learning (DL) techniques within recommender systems (RSs) has enhanced their precision and ability to handle large datasets. Nevertheless, this enhancement comes at a cost—a lack of transparency, which raises significant issues about the interpretability of the model. Here, we offer a review that aims to analyze the state-of-the-art AI explanations provided for deep learning-based recommender systems. The review classifies the methods and frameworks into two primary types of explainability: intrinsic and post-hoc. It also addresses different explanation strategies including graph-based, example-based, and text-based techniques. In addition, we describe the common deep learning architectures applied in recommender systems like CNNs, RNNs, GNNs, and Transformers, and discuss how these models interact with various techniques of explainability. Besides, our review reveals other important gaps, including the balance between accuracy and interpretability, limits on scalability, social issues like bias and opacity, or transparency—among other ethical issues. Lastly, it focuses on the designed user-centered, universal and ethically aligned methods of explainability which are tailored to the needs of users. The goal of these insights is to aid researchers and practitioners in developing more trustworthy and transparent recommender systems.