摘要
协同过滤推荐算法通常是基于兴趣相似的用户行为来实现个性化推荐,其核心是定义用户之间的兴趣相似度.本文在传统的协同过滤推荐算法基础上,考虑在线评论对用户相似度识别的影响.在混合商品推荐中,粗粒度识别评论情感极性;而在同类商品推荐中,细粒度识别每个商品特征的情感极性.如果用户对产品的某个特征评价次数大于平均次数,表明用户对该特征较关注;如果对产品的某个特征评价低于平均评价,表明用户对该特征较挑剔.进而根据用户评论来建立用户偏好模型,用户在评论中反映出来的相似度越高,表明用户之间的偏好越一致.实验表明,同传统的协同过滤算法相比,基于在线评论情感分析的用户协同过滤算法在准确率和召回率指标上有显著提升.
Collaborative filtering recommendation algorithm bases on user behavior with similar interests to produce personalized recommendation. The core of the algorithm is to define the distance between the user＇s interest similarities. The paper considers the online review sentiment impact on user similarity recognition. In mixed products recommendation, coarse-grained sentimental polarity is identified; while in same category products recommendation, fine-grained sentimental analysis is employed for each feature. If the user＇s evaluation frequency is greater than the average on a special feature, it indicates that the user pays close attention to the feature; while if the user＇s rating is smaller than the average rating on a special feature, it means the user has a strict requirement on this feature. And then the user＇s preference model is created according to reviews, the higher the similarity between users in the reviews, the more consistent preferences between users. Experiment results show that the proposed collaborative filtering algorithm based on sentiment analysis of online reviews improves the traditional algorithm significantly on accuracy and recall.
出处
《系统工程理论与实践》
EI
CSSCI
CSCD
北大核心
2014年第12期3238-3249,共12页
Systems Engineering-Theory & Practice
基金
国家自然科学基金(70971099
71371144)
上海市哲学社会科学规划课题一般项目(2013BGL004)
中央高校基本科研业务费专项资金(1200219198)
关键词
推荐系统
推荐算法
协同过滤
在线评论
情感分析
recommendation system
recommendation algorithm
collaborative filtering
online review
sentiment analysis
