基于局部模型融合的交互式电影推荐系统

Interactive Movie Recommendation System Based on Local Model Fusion

【目的】设计并实现一个交互式可视推荐系统,帮助用户理解推荐结果的产生原因,提高使用体验以及对推荐系统的信任。【方法】从用户历史观影标签集合中提取用户偏好特征,通过LDA模型基于此特征对用户进行聚类,并利用SLIM模型对不同用户子群分别训练局部模型,最后利用训练过程的上下文语义信息设计和实现最终的交互式电影推荐系统。【结果】设计了一个交互式的电影推荐系统RecVis,能够可视化推荐原因和用户画像,向用户提供推荐解释和交互反馈功能,以及实时获得根据其交互反馈的感兴趣的最新推荐结果。【结论】通过豆瓣电影数据集的测试,证明了该系统在推荐方面的有效性,并通过一系列案例分析验证了RecVis能够帮助用户理解推荐结果,增加对推荐系统的信任。

[Objective]To improve user experience and confidence in the recommendation system,and help users understand the cause of recommendation,we design and implement an interactive visual recommendation system.[Methods]Firstly,we extract the preference features of each user from the user’s history viewing tag set,and allow all users to be clustered into different subgroups based on these preference features using the LDA topic model.After that,we apply the SLIM sparse linear model to train the local recommendation model separately for each user subgroup.Finally,we use the contextual semantic information of the training process to design and implement the interactive movie recommendation system.[Results]This system,RecVis,can visualize the semantic information of recommendation and user portraits,provide recommendation explanation and interactive feedback function,and obtain the latest recommendation according to the user’s interactive feedback in realtime.[Conclusions]The test on the Douban movie dataset proves the effectiveness of the system in terms of recommendation.In addition,a series of case studies verify that RecVis can help users understand the recommendation results and increase trust in the recommendation system.