摘要
由于现有的用户大数据推荐算法普遍存在数据稀疏与准确度问题,提出了一种结合用户相似度与路径相似度的协同过滤算法。在分析兴趣偏好时,采用TF-IDF策略得到用户与某属性的兴趣度矩阵,考虑到兴趣的时间影响与稀疏特性,引入时间衰减公式和稀疏信息补充规则。在分析评分差异时,考虑到评价数据的杂乱性,采取信息熵进行计算用户相似性,上述过程引入时间间隔权重来处理时间影响。最后构建行为路径,利用距离计算判断用户行为的相似性,并将各相似度得到的推荐结果进行合并,得到最终推荐。通过MovieLens-1M与sample Criteo数据集,对算法的MAE、、和AUC指标进行对比验证,结果表明,所提算法可以通过兴趣与用户行为,获得更好的表达性能,降低对相似用户信息的依赖,具有良好的抗稀疏性能和推荐准确性。
Because the existing user big data recommendation algorithms generally have the problems of data sparsity and accuracy,a collaborative filtering algorithm combining user similarity and path similarity is proposed.When analyzing interest preferences,the TF-IDF strategy was used to obtain the interest matrix of users and an attribute.Considering the time influence and sparse characteristics of interest,the time attenuation formula and sparse information supplement rule were introduced.When analyzing the score difference,considering the clutter of evaluation data,information entropy was used to calculate user similarity.In this process,time interval weight was introduced to deal with the impact of time.Finally,the behavior path was constructed,the similarity of user behavior was judged by distance calculation,and the recommendation results obtained from each similarity were combined to obtain the final recommendation.The MAE,precision,recall and AUC indexes of the algorithm were compared and verified through Movielens-1M and sample Criteo data sets.The results show that the proposed algorithm can obtain better expression performance through interest and user behavior,reduce the dependence on similar user information,and has good anti-sparsity performance and recommendation accuracy.
作者
侯立
王健
HOU Li;WANG Jian(Jilin Normal University,Changchun Jilin 130022,China)
出处
《计算机仿真》
北大核心
2023年第3期476-479,489,共5页
Computer Simulation
基金
吉林省教育科学规划编号:GS2111。
关键词
兴趣偏好
评分差异
行为路径
协同过滤
大数据推荐
Interest preference
Score difference
Behavior path
Collaborative filtering
Big data recommendation
作者简介
侯立(1980.07-),男(汉族),吉林长春人,博士,讲师,研究方向:习近平思想,数据处理;王健(1984.9-),男(汉族),安徽宿州人,博士,讲师,研究方向:凝聚态物理专业。
