The invention discloses an electronic commerce recommending method based on a support vector machine (SVM). The method includes the steps that a user evaluation predication model based on the SVM is established, textual characteristics including semantics are extracted based on machine learning, evaluations are expressed as multi-dimensional characteristic vectors, the evaluations are classified, an SVM classifier divides the commodity evaluations into useful and useless categories, and the evaluations are automatically recognized; according to useful scores of the evaluations in the classifier, a user item matrix is filled by the adoption of a method of scoring predication through training samples according to the scores; according to importance of all items, corresponding vectors of a kernel function are endowed with corresponding weights, and meanwhile the weights of the corresponding vectors are corrected according to user process behaviors so that the purposes of improving predication precision and achieving an ideal recommending effect can be achieved. Statistics, machine learning, intelligent mode recognition and classification and other technologies are used for analyzing customer electronic commerce access behaviors and the commodity evaluations, commodities which a customer is interested in are predicated through the mode, a recommending result is generated and recommended to the customer, and the customer is helped to rapidly and accurately find out commodities really needed in time.