Recommendation systems emerged with the expansion of e-commerce shopping websites, addressing the concern of information overload. Most recommendation algorithms focus on predicting ratings or identifying relevant key features in reviews to gain insights into user preferences and product quality. However, these methodologies often neglect the effectiveness and user adoption (persuasiveness) of the recommended results. Explainable recommendations provide not only the recommendation outcomes but also a deeper rationale for the proposal, enhancing user satisfaction, trust, and transparency. The existing explainable recommender systems often face three critical challenges: (1) lack adequate extraction of both direct and indirect product features, (2) inability to effectively model complex feature interactions, and (3) clarity and coherence of the automatically generated sentence justifying the recommendation is lacking. To address these gaps, this paper presents the development of an Explainable Product Recommendation systems (XPRs) that leverages the comprehensive multi-task framework. This framework integrates two distinct bidirectional learning mechanisms, namely Bidirectional Encoder Representation from Transformers (BERT) for extracting interactive product features for rating prediction and the Bidirectional Long Short-Term Memory (Bi-LSTM) model for generating personalized, high-quality, and meaningful textual explanations. A key novelty of this approach lies in the hierarchical attention mechanism incorporated into the explanation generation model, which enhances the interpretability and persuasiveness of recommendations. Additionally, we introduce a fine-grained feature interaction module that explicitly models how different product attributes influence user preferences, improving recommendation accuracy. The performance of the proposed framework on highlighted datasets exhibits superior results compared to baseline frameworks in terms of recall and precision-oriented evaluation metrics.