5.1. Main findings While the booming e-commerce is bringing huge transaction volumes, the problem of “information overload” also causes problems for customers and businesses. First, while customers enjoy the convenience of e-commerce virtual shopping, they are often surrounded by a large amount of product information. Second, merchants are concerned that their product information cannot be fully and systematically presented to target customers in massive amounts of information. The e-commerce recommendation system provides a good idea to solve the "information overload" problem. The recommendation system can recommend the product information that meets the user's consumption preference to the target customer and help the customer to complete the conversion from "buy what" to the actual purchase behavior. Through the understanding of personalized recommendation technology, this paper combines the actual needs of the Internet of Things e-commerce project and carries out algorithm research and improvement from solving the cold start of users and improving the accuracy of recommendation results. At the same time, this paper proposes to construct user interest classification model by using user context information to solve the problem of user cold start. In addition, this paper combines the user trust model with the collaborative filtering recommendation algorithm to improve the accuracy of recommendation results. 5.2. Theoretical implications The theoretical significance of this research mainly has the following three points. First of all, this study deeply studies the cold start problem of users in the mobile client recommendation algorithm and forms a systematic understanding of the cold start problem. On this basis, this paper proposes a method of constructing user interest classification model by using user context information to solve the user's cold start problem. By means of an existing user clicking record, the method extracts context information in different environments and constructs a user interest classification model. Meanwhile, when the new user arrives, the context information of the new user in the current environment is extracted and matched with the user interest classification model, thereby predicting the browsing interest of the new user, completing the recommendation of the product information, and solving the cold start problem of the new user. Secondly, based on the definition of trust and the existing trust model learning, the user trust model suitable for mobile e-commerce platform is constructed. At the same time, this paper divides the generation of user trust into two parts: the trust generated by social reputation and the trust generated by social similarity. By extracting the context characteristics that characterize user trust, this paper uses AHP to complete the model construction process. In order to solve the problem that the user-scoring matrix sparseness of the user-item scoring matrix is small and the accuracy of the recommendation result is low, a collaborative filtering recommendation algorithm based on the user trust model is proposed. The algorithm improvement is divided into two parts: On the one hand, through the threshold filtering method, the user's trusted nearest neighbor is merged with the user's similarity nearest neighbor, thereby achieving the purpose of matching more neighbors for the user. On the other hand, the existing score prediction formula is improved, so that the score prediction process can fully consider the influence of the user trust relationship on the recommendation result.