节点文献
文本信息增强的矩阵分解模型及其在推荐系统中的应用研究
Text Content Enhanced Matrix Factorization and Application in Recommender Systems
【作者】 陈晓宇；
【导师】 徐锋；
【作者基本信息】 南京大学 ， 计算机技术（专业学位）， 2015， 硕士
【摘要】 在过去的十多年中,推荐系统被学者们广泛研究,一些实用的推荐方法也被运用到了现实的工业系统之中,如移动软件市场、电商网站、电影网站等等。传统的推荐系统的方法主要集中研究用户对物品的评分,但是在现实的系统中,用户在对物品打分时,通常会留下一段文本描述自己的感受,表明自己评分的原因和立场。这些文本信息中通常包含丰富且重要的信息,如用户的偏好,物品的特性。但是,大多数已有的推荐系统模型常常会忽略这些文本信息,原因在于文本信息处理困难并常带有大量噪音、数据维度不一致、难以与传统模型相整合等等。本文提出了一种文本信息增强的矩阵分解模型,尝试同时利用评分和文本信息,并研究其在推荐系统中的各项应用。本文的贡献如下：一、提出了一个利用文本中名词的主题特征建模来增强用户特征矩阵的矩阵分解模型。二、进一步提出了一个同时利用文本中名词和修饰词来增强物品特征矩阵的矩阵分解模型。三、实现了一个基于本文提出的模型的原型系统,展示了本文的模型能在现实中被合理地利用。同时,依据提出的模型和方法,本文在现实的数据集上进行了大量的相关实验,证明了模型对文本信息的有效利用,并研究了模型在推荐系统上的各项应用,如提高推荐准确度、改善冷启动问题等等。
【Abstract】 Recommender systems have been widely studied in the last decades, some have been applied in many real applications such as mobile App stores, e-commerce sites and movie sites. Most conventional recommender systems focus on users’historic rat-ings over items. However, in real-world systems besides rating, users usually provide their feedback towards the items with a few words (i.e., review content) to justify their ratings. Such review content may contain rich information about user tastes and item characteristics.However, due to the difficulty of text processing (too much noise, divergence of dimension), existing recommendation methods mainly make use of the historical ratings while ignore the content information. In this paper, we propose to take use of the review content alone with ratings for better recommendation via matrix factorization model. In particular, this paper has the following contributions. First, on the base of topic model, we propose the GTRT model using guidance term and regularization term to leverage the nouns in review content and enhance the learning process of user-side matrix. Second, we further extract item emotion features from nouns and modifiers to enhance the learning process of item-side matrix. Finally, we make a prototype system based on the proposed model. Also, experimental evaluations on two real data sets demonstrate the usefulness of review content and the effectiveness of the proposed method for recommendation.