Adaptive Cold-Start Thompson Sampling for Cold-Start Problems

The Thompson Sampling (TS) algorithm tends to malfunction for new users or items. This state is referred to as the cold start problem. Adaptive Cold Start Thompson Sampling (ACTS) is presented in this article to address this problem. Firstly, it uses basic data, relying on metadata to make smarter first guesses, and is more cautious in high-uncertainty situations. This paper also conducts testing of ACTS on open data, which models various cold start conditions. The findings show that ACTS experiences fewer errors earlier on and acquires the best possible option at a shorter period compared to the normal TS algorithm. The research looks at how various settings have affected the outcome and concludes that the algorithm has been able to deliver reliable results, which confirms that ACTS is a useful and efficient tool in the context of managing cold start issues in real-time learning systems.