The invention discloses a similarity-based article recommendation method combining multiple user feedbacks. The method comprises the following steps: modeling implicit semantic information in complexauxiliary information, calculating the similarity between different types of objects by adopting different asynchronous bidirectional walking modes for different types of user feedback types, combining the similarity calculation results by using a Bayesian sorting optimization technology, and carrying out optimization on the similarity calculation results according to the combined user-Bayesian sorting optimization technology; generating a Top-N recommendation result for each user based on a combined user-item similarity matrix. The influence of any node in the path on the final path calculation result can be modeled, so that the implicit semantics can be fully mined, and the semantics can be quantized to obtain the similarity between different types of objects.