Design and empirical research of simulation algorithms for business decision support with BERT and ISAC integration

In the rapidly evolving digital economy, business decision-making faces increasing uncertainty due to dynamic market environments, nonlinear interactions between variables, and human behavioral complexity. Traditional decision-support models rely heavily on linear optimization or rule-based systems, which are insufficient for modeling adaptive strategies in complex business ecosystems. This paper proposes a BDSSF-SAC based on an ISAC algorithm, enhanced by the integration of the BERT model for natural language processing. The proposed framework integrates deep reinforcement learning with business simulation to enable adaptive policy generation, probabilistic risk modeling, and real-time optimization. The ISAC algorithm enhances standard SAC by introducing an ERRC mechanism and a DQV scheme. Additionally, the integration of BERT allows the framework to leverage rich semantic information from enterprise consultation texts, improving decision-making accuracy and adaptability. We validate the system through empirical experiments using real enterprise management data. Results show that ISAC-based decision policies outperform DDPG, PPO, and traditional SAC in terms of convergence speed, reward stability, and adaptability, with a 21.4% improvement in average reward variance reduction.