The invention discloses a conversation method and a system for the intelligent customer service. According to the technical scheme of the invention, problems are classified, and the occurrence frequencies of all problems in each category are sorted. According to the sequential order of the sorted result, the calculation is conducted to obtain the attribute values of the problems and create a scoring function model. After that, based on the scoring function model, the problems are scored, so that the scores of the problems are obtained. During conversation, the problem of a visitor is obtained and scored based on the scoring function model, and then the score of the problem of the visitor is obtained. Meanwhile, the score of the problem of the visitor is compared and analyzed with the scores of problems in a database, and then a problem with the score thereof most closet to the score of the problem of the visitor and the corresponding answer of the problem are obtained. The obtained corresponding answer is adopted as a recommended answer for the problem of the visitor. Therefore, a similar problem with the score closet to the score of the problem of the visitor can be quickly found out in the database. Meanwhile, a corresponding recommended answer is provided for the visitor according to the similar problem. as a result, the communication efficiency and the question answering accuracy are improved.