Yan Wang, Yi Han, Lingfei Qian, Yueru He, Xueqing Peng, Dongji Feng, Zhuohan Xie, Vincent Jim Zhang, Rosie Guo, Fengran Mo, Jimin Huang, Yankai Chen, Xue Liu, Jian-Yun Nie
Conv-FinRe is a benchmark for evaluating financial recommendation models that focuses on rational decision quality rather than just mimicking user behavior.
In financial advising, it's important for recommendation systems to make decisions based on rational analysis rather than just copying what users have done in the past, which may be influenced by short-term market changes. The Conv-FinRe benchmark is designed to test recommendation models by having them consider long-term investment goals and individual risk preferences. This new benchmark allows researchers to see if models are making smart, utility-based decisions or just following noisy, short-term user behaviors. The benchmark uses real market data and controlled advisory conversations to evaluate the performance of state-of-the-art language models.
