基于用户偏好和可疑度的推荐方法研究

Research on recommendation method based on user preference and suspicious degree

针对传统推荐算法在推荐过程中存在忽略用户偏好、用户恶意虚假信息和时间序列等问题,引入用户兴趣模型,结合用户可疑度与时间效应计算更新用户相似度,经过深度学习网络得到最佳推荐目标。为避免出现数据过拟合情况,在利用贪心思想训练用户数据时,给隐含层和可见层均加上了用户偏好,一定程度上提高了深度学习网络的自学习能力。将改进的算法与传统协同过滤推荐算法在MovieLens数据集上作推荐对比实验,实验结果证明,相对于传统的推荐算法,改进的推荐算法可以大大提高项目推荐的精确度。

In the recommended process,the traditional algorithm has several problems. For examples,the algorithm ignored the customer preference and the customer＇s malevolence,also,it provided a variety of false information,even for time series problems. This paper introduced users＇ interests modeling while it built the model. Then,it toke about users＇ suspicious degree and time effects used to calculate and updated users＇ similarity. In the end,it would highlight the optimum recommended target. In order to avoid data overfitting,it used the greedy algorithm to train the user＇s data,which added a preference to the hidden layer and the visible layer. Finally,this method found the optimum solution and improved study ability with deep learning network. It compared the experiments about improvement recommended algorithms and traditional recommended in the Movie-Lens of data collections. The result displays that compare to the old fashioned recommended algorithms,the improved one can greatly improve the accuracy of project recommendation.