基于时间加权ALS模型协同过滤推荐算法

Collaborative filtering recommendation algorithm based on time weighted ALS model

大数据时代背景下,网络产生的数据爆炸式增长,用户想获取符合个性化需求的数据信息变得十分困难。推荐系统的目的就是根据用户的个性化兴趣从而从大量的数据中寻找用户所关心的信息,并推荐给用户。正因如此,推荐系统一直是学术界讨论的热点,但是传统的推荐算法忽略了时间的推移会导致用户兴趣的偏移,物品特征对推荐的影响,从而导致的准确率不高。针对以上问题,对矩阵分解模型进行了优化,将时间因子和物品特征因子加入ALS算法中进行融合加权,并在Spark平台并行化实现。实验表明,优化后的ALS算法的RSME值下降了6.2%,推荐结果的准确度有所提升。

With the era of big data,the explosive growth of data generated by the network makes it very difficult for users to obtain data information that meets their personalized needs.The purpose of the recommendation system is to find out what the user cares about from a large amount of data based on the user’s personalized interests and to provide recommendations.This is why the recommendation system has been a hot topic in academic circles.However,the traditional recommendation algorithm ignores the effect of time on user’s interest,the influence of item characteristics on recommendation,and the low accuracy.To solve the above problems,the matrix decomposition model is optimized.Adding time factor and item feature factor to ALS algorithm for fusion weighting,and implements it in parallel on Spark platform.The experimental results have shown that the RSME value of the optimized ALS algorithm has been reduced by 6.2%,and the accuracy of the recommended results have been improved.