The referral system is to recommend the experts when the user has a question. Most pieces of the referral system consider the judgement of relevant expertise. This thesis concentrates on the will of helping. The personalized referral agent recommends the expert that can answer according different users. The acquaintances of the user can split into two groups. The acquaintances that the user is willing to help belong to one group, and others belong to another group. The referral agent uses the analysis of message history to help the user classify his acquaintances more easily. If all the users classify their acquaintances, the referral agent can recommend the experts that are willing to help the user with the cooperation between agents.