With the advancement of information technology, the application of blockchain technology in various fields continues to increase. However, the information isolation between different blockchain networks leads to the phenomenon of information silos, which impedes the exchange of information and the flow of value between networks. In order to solve this problem, cross-chain technology comes into being, among which notary mechanism is widely concerned because of its easy implementation and good scalability. However, the current notary public mechanism is faced with the problem of insufficient reliability of notary public, which limits the popularization of its practical application. This paper proposes a credit score optimization method based on weighted leader ranking algorithm to improve the performance of cross-chain notary mechanism. Firstly, a new trust evaluation model is constructed, which can comprehensively evaluate the behavior of notary public by considering multiple evaluation indexes according to the characteristics of cross-chain transaction of notary public mechanism. Experimental results demonstrate that this model significantly outperforms traditional methods in evaluation performance. It is then applied to the notary mechanism, leading to the proposal of a cross-chain notary mechanism that effectively quantifies notary node behavior, thereby enhancing reliability. The model shows strong performance in selecting trusted notary nodes. Finally, to address the centralization issue in notary elections, a deep Q network-based election algorithm is introduced, ensuring both cross-chain transaction performance and election fairness, while mitigating centralization tendencies.