Researchers have analyzed how generative news recommenders can inadvertently amplify popularity bias. Using an instruction-tuned Llama 3 8B model trained on browsing data from a Japanese news service, they found that the model's generated titles were more concentrated and less diverse than actual user browsing patterns. This suggests that models may reproduce popular content not just due to user preferences, but because they memorize frequently encountered text strings from their training data. The study indicates that this memorization pathway has significant implications for information diversity on news platforms. AI
IMPACT Suggests generative models may inadvertently narrow information diversity by reproducing popular content due to memorization.
RANK_REASON The cluster contains an academic paper detailing a memorization-based analysis of generative news recommendation. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →