Privacy protection is one of the important concepts in communities highly tied up by computer networks, especially in social networking services (SNS) since personal information can be found in SNS e.g. which area a user lives, which school a user belongs to, etc. In existing researches, filtering methods for privacy protection are proposed in order to prevent a user from writing information which indicates his/her physical address or affiliation. However, the filter is not secure enough in SNS, since information which is related with a user's privacy can be posted not only by the user but also by other users. Therefore, in this paper, we propose a new filter which mutually prevents users from posting other's personal information. Moreover, by the filter, proper nouns in contents are appropriately replaced with common nouns by considering readers' relation with a targeted user. Furthermore, we apply multistage Bloom filter to the filter in order to reduce the dealing cost of n square times filtering which is brought from mutual reference by n number of users. In addition, we evaluate the filtering performance of our method by comparing with ordinary filtering methods.