CASIA OpenIR  > 毕业生  > 硕士学位论文
隐式反馈中推荐系统关键技术研究
Alternative TitlePersonalized Recommendation Based on Implicit feedback
刘旭东
Subtype工学硕士
Thesis Advisor杨青
2015-05-30
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword推荐系统 隐式反馈 协同过滤 协同聚类 Recommender System Implicit Feedback Collaborative Filtering Coclustering
Abstract推荐系统通过对内容和用户行为的分析,建立适当的模型,帮助用户从海量的数据中找到自己感兴趣的内容。推荐系统中用户的行为反馈包括显式反馈和隐式反馈,隐式反馈信息在推荐系统算法中被广泛应用。隐式反馈体现着用户的兴趣爱好,对隐式反馈信息的挖掘有助于提高推荐系统的效果,以更好地设计推荐系统。 本文对隐式反馈中个性化推荐的关键技术进行了深入的研究,包括在隐式反馈中的评分预测问题和TopN推荐问题。本文的主要工作和贡献如下: 1. 提出基于隐式反馈的隐特征FM模型,并把模型应用于评分预测中。用户的浏览记录对用户当前的评分是有影响的,但是用户的隐式反馈是个高维稀疏的向量。我们利用Latent Dirichlet Allocation(LDA)模型对隐式反馈进行降维,并把降维后的隐式反馈表示应用于Factorization Machine(FM)模型中,使隐式反馈信息在评分预测中发挥作用;进一步我们运用word2vec模型学习出用户的具有浏览顺序特性的隐式反馈表达,同样应用于FM模型中。实验表明,以上两种方法有效地提高了评分预测的精度,对隐式反馈进行了降维,并把浏览顺序这个信息应用于隐特征的学习上,提高了方法的效果。 2.提出了基于隐式反馈的协同聚类(co-cluster)模型,CCCF模型,并把模型应用到TopN推荐中。传统的协同过滤算法是在整个物品(item)空间上衡量用户(user)的兴趣,然而实际的互联网中用户可能在多个领域中有不同的兴趣。本文利用隐式反馈,提出了一种基于“用户-物品”的协同聚类算法,并在每个子类中分别独立实现协同过滤算法,之后把每个类别中的推荐结果融合以提升TopN推荐的精度。在我们的模型中,类别与类别之间可以交叉,同时我们的模型可以描述user(item)与类别之间的连接强度。我们的模型复杂度低,具有可解释性、灵活性和可拓展性等优点。大量的实验结果表明,CCCF模型可以提高各种协同过滤算法的效 果,并且推荐结果具有可解释性。
Other AbstractRecommender System could model users’ interests based on their historical behaviors and contents. The users’ feedback in a Recommender System includes explicit feedback and implicit feedback. Implicit feedback data is widely used in a Recommender System. The implicit feedback information can express the user’s interests. Mining implicit feedback information can help to improve the recommendation performance so that we can design a better system. This work investigate the personalized recommendation technique based on implicit feedback, including rating prediction task and Top-N recommendation task. Following are the main contributions of this paper: 1. Latent feature based FM model in rating prediction: A user’s historical data can influence his/her behavior in rating a new item. However, the user’s implicit feedback is high dimensional and sparse. We use LDA model to train the user’s latent topic based on the implicit feedback and use these topics to train FM model. Further more, we use word2vec model to train another kind of latent features acoording to the history order information so that the latent features can express the order info. Experiment results show that our method can help to reduce the dimension of implicit feedback and improve the quality of a recommender system. 2.CCCF: Improving collaborative filtering via scalable user-item co-clustering: Most of the existing CF methods measure users’preferences by their behaviors over all the items. However, users might have multiple interests. We propose a novel and scalable method CCCF which improves the performance of CF methods via user-item co-clustering. CCCF first clusters users and items into several subgroups. Then based on the coclustering results, traditional CF methods can be easily applied to each subgroup and the recommendation results from all the subgroups can be easily aggregated. The subgroups are overlapped. We also leverage the cluster affiliation strengths to further improve the final recommendations. Our model has the following characters: scalability, flexibility, interpretability and extensibility. These characters have been proved by experiments.
Other Identifier201228014628046
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7769
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
刘旭东. 隐式反馈中推荐系统关键技术研究[D]. 中国科学院自动化研究所. 中国科学院大学,2015.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20122801462804(1552KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘旭东]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘旭东]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘旭东]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.