推荐系统综述recommder（精华版）.pptx

Recommender Systems IntroductionCheng Shi20.Dec.2016OutlineBackgrounds & historyAlgorithmsContent-based RecommendationCollaborative Filtering-based Recommendation • User-based Recommendation • Item-based Recommendation • Model-based RecommendationProblemBackgroundsInformation overload is one of the most critical problems, and personalized recommendation system is a powerful tool to solve this problem.historyContent Filtering [Before 1992] Grouplens [1994]—Frist recommender system using rating dataMovielens[1997]—Frist movie recommender system—Provide well-known dataset for researchersAmazon—proposed item-based collaborative filtering(Patent is filed in 1998 and issued in 2001)Netflix Prize[2006]—Latent Factor Model(SVD,RSVD,NSVD,SVD++) —Yehuda Koren’s team get prizehistoryAfter 2006—Interpretability of the recommendation results—Join the sentiment analysis to the Matrix Factorization—Recommender Systems with Deep LearningContent-basedItem Profiles- its genre- the participatingactors - Its box officepopularity- so forth User Profiles- movies andscore listexample (movie)collaborative filtering The Long Tail User-based CFKonstan, Joseph A., et al. “GroupLens: applying collaborative filtering to Usenet news.“ Communications of the ACM 40.3 (1997): 77-87User-based CFØGrouplens—The rating servers predict scores based on the heuristic that people who agreed in the past will probably agree again.—Basic idea: recommend items similar to users favorite items.—GroupLens Architecture Overview—A Dynamic and Fast-Paced Information System—Ratings Sparsity—Performance ChallengesUser-based CFEstablishment of user modelSimilarity ComputingFind similar users setp Euclidean Distance SimilaritySimilarity Computingp Cosine Similarityp Jaccard Coefficientp Pearson Correlation SimilarityNeighborhoodsp K-neighborhoods or Threshold-based neighborhoodsSimilarity ComputingSimilarity ComputingItem-based CFLinden, Greg, Brent Smith, and Jeremy York. “Amazon. com recommendations: Item-to-item collaborative filtering.“ IEEE Internet computing 7.1 (2003): 76-80.Item-based CFA-Few details -basic idea-ScalabilityItem-based CFM users, N itemsItem-based CF sampleScalabilityA has more than 29 million customers and several million catalog items. For large retailers like A, a good recommendation algorithm is scalable over very large customer bases and product catalogs, requires only subsecond processing time to generate online recommendations. A ComparisonScalabilityDiversity& PrecisionNetflix PrizeNetflix是一家美国视频网站，公司一 开始的主要业务是提供DVD和Blu-ray光盘的出 租服务。

现在的主要业务是原创内容的网络流 媒体服务2013年凭借高端自制美剧《纸牌屋 》和随后的多部剧集的超高质量引起全球瞩 目2006年，NETFLIX宣布，设立一项大赛，公 开征集电影推荐系统的最佳电脑算法，第一 个能把现有推荐系统的准确率提高10%的参赛 者将获得一百万美元的奖金2009 年 9 月 21 日，来自全世界 186 个国家的四万多个 参赛团队经过近三年的较量，终于有了结 果一个由工程师和统计学家组成的七人团 队夺得了大奖，拿到了那张百万美元的超大 支票Matrix factorizationYehuda, Robert Bell, and Chris Volinsky. “Matrix factorization techniques for recommender systems.“ Computer 42.8 (2009): 30-37.Matrix factorizationA basic matrix factorization modelMatrix factorizationMatrix factorizationMatrix factorizationMatrix factorizationAdditional input sources：Biases：Temporal dynamics：Confidence levels：Browsing history,gender,age group Zip code,income levelMatrix factorizationLFM(Latent factor model)Matrix factorizationProbabilistic MFSalakhutdinov, Ruslan, and Andriy Mnih. “Probabilistic matrix factorization.“NIPS. Vol. 20. 2011.Probabilistic MFProbabilistic MFAutomatic Complexity ControlConstrained PMFExperimental ResultsExperimental Results Problemp 评价一个推荐系统标准p 算法效率、可解释性简单的算法+海量数据应该是能符合实际生产环境Thanks !。