Attribute-Based Encryption (ABE) has been widely used for ciphertext retrieval in the cloud environment. However, bi-flexible attribute control and privacy keywords are difficult problems that have yet to be solved. In this paper, we introduce the denial of access policy and the mutual matching algorithm of a dataset used to realize bidirectional control of attributes in the cloud server. To solve the problem of keyword privacy, we construct a security trapdoor by adding random numbers that effectively resist keyword guessing attacks from cloud servers and external attackers. System security is reduced to the Deterministic Bilinear Diffie-Hellman (DBDH) hypothesis problem. We validate our scheme through theoretical security analysis and experimental verification. Experiments are conducted on a real dataset, and results show that the scheme has higher security and retrieval efficiency than previous methods.