Topic Model-Based Freshness Estimation Towards Diverse Tweet Recommendation
Makoto Yokoyama, Qiang Ma · 2019
Most conventional tweet recommendation methods focus on matching tweets with user preferences. However, only considering user preferences may raise the risk of filter bubbles occurring, i.e., users may not be able to access diverse information. In this paper, we propose a freshness-oriented tweet recommendation method, aiming towards diverse information proliferation. We propose the notion of freshness and a corresponding topic model-based estimation method, to help users find novel information within a timeline by comparing previous tweets. Our experimental results demonstrate the superiority of our proposed method for helping users obtain fresh and useful information.