一种改进的缓解推荐系统物品冷启动的方法

摘要

With the information bursting on the internet and the whole world, recommendation system is playing a much more important role on the internet. Recommendation system has solved the overload information, however, it hasn’t solved the cold starting problem yet. In order to alleviate the item cold starting problem, we add an obvious trusted network based on the collaborative filtering and the decision tree. Our experiments have proved that the new item recommendation, which we recommend by the trusting system, is much more accurate compared to the original algorithm. When the rating number is 0, MAE is 16.7% lower than original algorithm. What’s more, the trusted recom-mendation system, not only satisfied most of the users’ interests, but also makes them rely on it more than before.%信息爆炸的时代，推荐系统越来越成为网民的依赖，它有效的解决了信息过载的问题，但是却没有解决推荐系统的冷启动问题。为了缓解新项目的冷启动问题，结合基于物品的协同过滤算法与决策树思想，这篇文章在孙等人的算法上做了改进，把算法的第一步替换为用显示的信任网络对用户做划分的方法。显示的信任网络，可以对用户做了更细致的分类，把信任网络添加到算法中，原算法便被改进为基于信任网络的推荐系统。改进后的算法不仅满足了一大部分用户的偏好与需求，而且使得系统用户更加依赖推荐系统。实验表明，利用显示的信任网络对新项目的推荐，其推荐结果的准确性比原算法高，推荐的结果也更加稳定。在评分个数分别为0，5，10的情况下，平均绝对误差比原算法的低了16.7%，21.6%，31.7%。