一种融合多因素社交活动个性化推荐模型

针对当前网络社交活动个性化推荐精度较低的问题,融合用户对活动兴趣度、召集者影响力以及地理位置偏好等三方面因素,提出一种融合多因素社交活动个性推荐模型。采用LDA文件主题模型求取用户与其参加过的所有社交活动的主题分布,利用隐含主题概率分布来表征用户的兴趣度,并构建用户与召集者间的影响力矩阵。根据活动举办地与用户常住地,建立距离幂律分布,并结合用户参加活动的频数,建立用户地理位置偏好概率模型。采用不同权值配比,综合三方面的因素形成最终的社交活动个性推荐。对比实验表明,该算法与三个因素个性推荐算法相比,准确率至少提高了36.7%,召回率至少提高了35.9%;与其他两个同类网络社交活动推荐算法相比准确率至少提高了8.77%,召回率至少提高了8.57%。

A MULTI-FACTOR SOCIAL ACTIVITY PERSONALIZED RECOMMENDATION MODEL

In view of the low accuracy of personalized recommendation in current online social activities,a multi-factor personalized recommendation model for social activities is proposed by integrating user interest in activities,convener influence and geographical location preference.The topic distribution of all social activities that the user participated in was extracted by the model using the LDA file theme model,and the user s interest degree was characterized by the implicit topic probability distribution.And the influence matrix between the users and the conveners was constructed.The power law distribution of distance was established according to the host site and the users habitat.Combined with the frequency of users participation in activities,the probability model of user s geographic location preference was established.Different weight ratios were used to integrate three factors to form the final social activity personalized recommendation.Compared with the three-factor personality recommendation algorithm,the proposed algorithm improves the accuracy by at least 36.7%and recall rate by at least 35.9%.Compared with the other two recommendation algorithms,the proposed algorithm improves the accuracy by at least 8.77%and recall rate by at least 8.57%.