基于人工蜂群的移动终端大数据半监督推荐  

Semi Supervised Recommendation of Mobile Terminal Big Data Based on Artificial Bee Colony

作　　者：郑捷 杨兴耀[1] 于炯[1] 李想 ZHENG Jie;YANG Xing-yao;YU Jiong;LI Xiang(College of Software,Xinjiang Uygur Autonomous Region,Xinjiang University,Xinjiang Urumqi 830046,China)

机构地区：[1]新疆大学软件学院,新疆乌鲁木齐830046

出　　处：《计算机仿真》2022年第7期497-501,共5页Computer Simulation

基　　金：国家自然科学基金项目(61862060,61966035,61562086);新疆维吾尔自治区教育厅项目(XJEDU2016S035);疆大学博士科研启动基金项目(BS150257)。

摘　　要：传统移动终端大数据推荐算法无法获取数据项目类间与类内分布信息熵,样本标记出现偏差,导致推荐精度偏低,用户满意度偏低。提出移动终端大数据半监督推荐算法。结合项目评分矩阵和属性矩阵,运算用户偏好权值,引入信息熵概念,获取项目属性类间与类内分布信息熵。构建用户偏好模型,确定聚类目标函数。利用人工蜂群半监督算法,通过标记与未标记样本,优化目标函数;选取最佳聚类中心,使用余弦法计算具有相同偏好用户间的相似度,确定近邻数量。采用加权平均值实现用户对所有项目评分,并按从高到低的顺序对评分结果排序,将得分靠前的项目推荐给用户。仿真结果证明,上述方法下大数据聚类结果精确,可避免陷入局部最优,且大数据推荐误差低,提高了用户满意度。Traditionally,the algorithm of recommendation for big data in mobile terminal was unable to obtain the inter-class and intra-class information entropies of data items,leading to low recommendation accuracy and low user satisfaction.Therefore,a semi-supervised recommendation algorithm for big data in mobile terminal was put forward.Combining project scoring matrix with attribute matrix,we calculated the user preference weight at first,and then introduced the concept of information entropy to obtain the inter-class and intra-class information entropies of project attributes.Moreover,we built a model of user preference to determine the clustering objective function.According to the labeled and unlabeled samples,we used the semi-supervised algorithm based on artificial bee colony to optimize objective functions.Furthermore,we selected the best cluster center and used the cosine method to calculate the similarity between users with the same preference,thus determining the number of nearest neighbors.Meanwhile,we used the weighted mean to grade all items and sorted the results from high to low.Finally,we recommend the items with high score to users.Simulation experiments prove that this method has the advantages of accurate big data clustering results and can avoid falling into local optimization.In addition,the big data recommendation error is low,which improves user satisfaction.

关 键 词：移动终端大数据 半监督学习算法 人工蜂群算法 目标函数 加权平均值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]