In this paper, we proposed a hybrid recommendation model to tackle two challenges in the recommendation system. First, many of the products have been browsed frequently but may not consequentially be ordered. As a result, the users' actions may not directly be considered as the preference on a specific item. Second, the popularity of sold products has a highly skewed distribution which results in the cold start problem in the recommendation. In order to extract knowledge from users' implicit feedback, we develop the neighborhood structure of users and behaviors on products in the multi-behavior interaction network (MBIN) that incorporates the multiple behaviors simultaneously. To deal with the cold product issue and the skewed distribution problem, we take the product information into consideration by using the metadata of products and extracting more features from the textual contents to form a knowledge graph. By applying embedding algorithms to the multi-behavior interaction network and the knowledge graph, we are able to catch the user's preference from the collaborative implicit feedback aspect and the product information aspect. To evaluate the performance of our model, we conduct extensive experiments on the real-world dataset. The result of our approaches outperforms several widely used methods for the recommendation systems.