多层次场景感知评分预测研究

近年来,评论在电商等网络平台中起着越来越重要的作用。充分利用评论信息,可以更好地理解用户兴趣和物品性质,提升推荐系统的性能。但是,现有的基于评论的推荐模型都只在“单词”层面或“评论”层面之一建模,且没有考虑交互场景对用户兴趣和物品性质的影响。因此提出一个新模型SCRM(Scene Context-aware Rating Prediction at Muti-level),同时在两个层面层次化、细粒度地抽取相关特征;在“评论”层面加入了场景上下文信息,突出当前场景中起主要影响的因素。在来自Amazon的不同领域上的四个公开数据集上进行了实验,结果显示基于均方误差SCRM整体上显著地超过了最先进的方法,包括MF、DeepCoNN、D-ATT和NARRE。

SCENE CONTEXT-AWARE RATING PREDICTION AT MULTI-LEVEL

Recently,reviews have played an increasingly important role in online platforms such as e-commerce.Review information is useful to better understand the user s interests and the nature of the item and improve the performance of the recommendation system.However,the existing review-based recommendation models are modeled only at one of the “word” level or the “review” level,and do not consider the impact of the interaction scene on user interests and the nature of the item.Therefore,we proposed a new model SCRM(Scene Context-aware Rating Prediction at Multi-level).It extracted relevant features hierarchically at two levels,and added scene context information at the “review” level to highlight the main influence factors in the current scene.Experiments on four public datasets from different areas of Amazon are conducted,and our results show that SCRM constantly and significantly outperform existing state-of-the-art models,including MF,DeepCoNN,D-ATT,and NARRE.