In the next-generation heterogeneous networks, 5G New Radio (NR) base stations will contend with Wi-Fi access points for the use of unlicensed frequency bands to increase the transmission rate. In 3GPP Licensed Assisted Access (LAA) standards, the technology called Listen Before Talk (LBT) is introduced for the coexistence of NR base stations and Wi-Fi access points. However, the contention for unlicensed bands between LAA and Wi-Fi is rather unfair; throughput with LBT used in LAA is much higher than that with Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) used in Wi-Fi. Since the performance of Wi-Fi when coexisting with LAA mainly relies on how the LBT parameters are configured by the LAA, we propose in this paper a radio resources allocation scheme which adjusts the Transmission Opportunity (TXOP) duration of LAA based on reinforcement learning to improve the fairness between LAA and Wi-Fi. The simulation results illustrate that the proposed scheme effectively improves the fairness between LAA and Wi-Fi in terms of throughput.