<![CDATA[The phenomenon of information overload on digital platforms frequently leads to a skewed long-tail distribution, where user interactions concentrate on popular items while the majority remains undiscovered. This popularity bias is particularly acute in the book domain, where metaphorical titles defy traditional lexical matching, causing unique content to remain hidden in the long-tail area. While semantic models improve relevance, a significant research gap persists in mitigating the trade-off between accuracy and catalog exposure. This study proposes a novel stochastic framework by integrating Roulette Wheel Selection with hybrid rating and temporal decay weighting to enhance long-tail discoverability. To evaluate its generalizability, this mechanism was tested on the Goodreads Poetry dataset across four distinct content-based embedding models: SBERT-Tuned, SBERT, FastText, and Word2Vec. We conducted a rigorous comparative analysis between the proposed Stochastic (S) approach and Deterministic (D) Top-N selection. Experimental results demonstrate that the stochastic mechanism consistently improves catalog diversity across all baselines. SBERT-Tuned emerged as the most promising model, achieving a 461.71% increase in Catalog Coverage with a marginal 2.15% Recall reduction. Other models also exhibited significant gains, where the integration of Roulette Wheel Selection boosted coverage by 381.72% for FastText and 225.15% for Word2Vec, highlighting the framework's robustness. Furthermore, a consistent decrease in the Gini-index and an improvement in Novelty scores (ranging from +16% to +29%) confirm a more equitable distribution of item visibility. This research contributes a methodological advancement by proving that stochastic selection, guided by time-aware weighting, can effectively dismantle popularity bias without sacrificing relevance.]]>
