The Stranger Algorithm: A Novel Diversity-Centric Recommendation Model for New Product Survival in the Big Data Era

The advent of Big Data has amplified the challenges faced by traditional Recommender Systems (RSs), particularly concerning product Cold Start and user entrapment within the Filter Bubble. This paper introduces the Stranger Algorithm (STRIDE); a hybrid recommendation model designed to inject measurable diversity and ensure the initial visibility and survival of new products. We define a stranger user based on their Diversity Score(𝑆𝑐𝑜𝑟𝑒𝑆𝑡𝑟𝑎𝑛𝑔𝑒𝑟), calculated from Categorical Entropy, Novelty Propensity, and Anti-Popularity Score. This component is dynamically integrated into a hybrid scoring function. The optimization of the scoring weights(𝑤1, 𝑤2, 𝑤3) is managed by a Recurrent Neural Net-work (RNN), trained with a weighted loss function to ensure high reliability. Results from the MovieLens 1M dataset of Amazon demonstrate that the RNN-optimized Stranger Algorithm increases the New Product Survival Rate by +133.3%, achieving Precision of 89% and Recall of 90% for targeted Cold Start interactions.