摘要
在推荐系统中,通过收集和分析用户在系统中的所有行为信息,创建用户独有的偏好模型,从而根据该模型推断出用户可能感兴趣的物品。传统协同推荐算法一般都是利用收集的用户行为信息,根据偏好模型分析用户的行为特点,筛选出向用户推荐的物品列表,但推荐列表大同小异。为了提高推荐的准确性和精确性,让用户在不同的时间可以看到不同的推荐结果,提出了以传统协同过滤算法为基础的改进算法。在分析用户行为信息,建立用户行为特征时,考虑不同时间下用户历史信息也不同,时间越近越能反映当前用户行为特征。用户在较短时间间隔内感兴趣的物品,具有更高的相似度。故以时间作为权重因子引入到算法中,加重近期行为在算法中所占的比重,优先向用户推荐与他浏览过的物品类似的物品,从而提高推荐物品的多样性。在典型数据集上的实验表明,在保证推荐准确度的前提下,融合时间的推荐算法准确率和召回率明显提高,验证了该算法的有效性。
In recommending systems, by collecting and analyzing all behavior information of system for users, the unique user preference model is created and according to it,interested goods of users are inferred. The traditional collaborative recommendation algorithm recom- mends the goods by analyzing the characteristics of the user behavior according to the collected user behavior information. However,the items are similar. In order to improve accuracy and precision that users often see the different recommendation results, an improved method is proposed based on the traditional collaborative filtering one. While users' behavior characteristic is extracted, they have different effects at different time points. It' s getting closer time to reflect user features and higher similarity, so the weight factor is introduced into the al- gorithm to aggravate proportion about recent behavior. Giving priority to recommend similar items the users' likes recently can improve recommendation diversity. On the premise of accuracy, the effectiveness has been verified in typical recommendation data set. The results show that it improves the accuracy and recall rate by experimental analysis.
出处
《计算机技术与发展》
2017年第7期79-82,共4页
Computer Technology and Development
基金
国家"863"高技术发展计划项目(2012AA121005-3)
关键词
协同过滤
商品推荐
时间影响
权重因子
时间衰减
collaborative filter
commodity recommendation
time effect
weight factor
time decay