In this paper, we attempt to use the rating information to adjust linkage-weight between nodes to improve effectiveness of product recommendation. As we all known, sometimes we purchase goods or watch a movie do not really mean we like or we would purchase them again. Thus, it is necessary to take the relationship of nodes into account, which we use linkage-weight to indicate. However, most of prior work only consider the existence of linkages, ignoring the remarkable meaning of linkage between nodes. To evaluate our algorithm, we compare both the original link prediction local path index and simplified linkage-weight index (-1,0,1) to our index and our results achieve significantly better performance.