Location-based recommendation services, such as Foursquare, enhance the convenience in the life of consumers. Users are usually sensitive to disclose their personal information. Unavoidable security concerns arise because malicious third parties could misuse confidential information, such as the users' preferences. The mainstream research to this problem is employing the privacy-preserving k-NN search algorithm. However, two major bottlenecks exist. One is that it only provides the nearest points of interest (POI) to the users without any recommendations based on the users' behavior history. This limited service eventually results in a situation in which no user would prefer to continue using it. The other is that only a single user holds the private key; thus, the service providers cannot obtain any user's information to analyze to make a profit. To solve the first problem, our proposed protocol provides recommendation services by adopting collaborative filtering techniques with an encrypted database based on fully homomorphic encryption aside from encrypting both the user's location and preferences. For the second problem, a privacy service provider (PSP) is designed to generate and hold the private key. Thus, service providers can homomorphically compute aggregate information concerning user behavior patterns and send the encrypted results to PSP to ensure decryption while maintaining the privacy of individual users. Compared with the previous studies, the novelty of the proposed protocol is the design of a commercially valuable privacy recommendation mechanism that could benefit both consumers and service providers on LBS.