The paper addresses the challenge of generating justifications for recommendations to enhance transparency and reliability. Existing approaches struggle to produce relevant and meaningful justifications, often relying on noisy or irrelevant review content. To tackle this, the authors propose a pipeline to identify and extract high-quality justifications from large review corpora, constructing personalized justification datasets. They introduce two models: (1) a reference-based Seq2Seq model with aspect-planning, which generates justifications covering different aspects using historical justifications as references; and (2) an aspect-conditional masked language model that generates diverse justifications from templates extracted from justification histories. Experiments on real-world datasets from Yelp and Amazon Clothing show that their models can generate convincing and diverse justifications, outperforming baselines in terms of relevance, informativeness, and diversity. The study also highlights the importance of aspect-planning in improving the quality and personalization of generated justifications.The paper addresses the challenge of generating justifications for recommendations to enhance transparency and reliability. Existing approaches struggle to produce relevant and meaningful justifications, often relying on noisy or irrelevant review content. To tackle this, the authors propose a pipeline to identify and extract high-quality justifications from large review corpora, constructing personalized justification datasets. They introduce two models: (1) a reference-based Seq2Seq model with aspect-planning, which generates justifications covering different aspects using historical justifications as references; and (2) an aspect-conditional masked language model that generates diverse justifications from templates extracted from justification histories. Experiments on real-world datasets from Yelp and Amazon Clothing show that their models can generate convincing and diverse justifications, outperforming baselines in terms of relevance, informativeness, and diversity. The study also highlights the importance of aspect-planning in improving the quality and personalization of generated justifications.