In this paper, we propose a new recommender algorithm based on Slope One algorithm and new similarity measurements. We incorporate additional sources of information about the users to relieve the cold start problem. Users generate a large number of interactions while browsing a website. These users' interactions are considered accurate enough to make recommendation. Then, we propose to take into account all the users interactions, to create a new method based on several communities in order to predict recommendation. We evaluated our improved algorithm on tourism datasets and we have shown positive results. We compared as well our algorithm to SVD, Slope One, Weight Slope One and baseline algorithms (Item-Item and User-User). We have obtained an improvement of 6% in precision and recall as well an improvement of 16% in RMSE and nDCG.