Curated by: Yehuda Koren
Recommender systems assist users in selecting products or services most suitable to their tastes and needs. With the rapid growth of web content supply and of online item catalogs, the personalized advice offered by recommenders is vital. This, together with the widening availability of user data, has contributed to a vast interest in recommendation technologies.
The nature and quality of a recommender is greatly affected by the kind of signals it takes as an input. Consequently, recommendation technologies are broadly divided into two types: (1) collaborative filtering based on analyzing past user activities like explicit rating of items, or implicit indication of preference such as clicks, purchases etc. (2) Content-based filtering which determines preferences by generalizing predefined item and user attributes like text, tags, genres, demographics, etc. Generally speaking, content-based methods are preferred for combating cold start scenarios when little activity is recorded on an item or a user. Yet, as more activity data is becoming available, collaborative filtering is gaining an edge by being more accurate. Real life situations usually target both new and familiar users and items, which calls for hybrid recommenders that combine collaborative and content filtering.
A recommendation is only as good as the information it holds on the user. Therefore, recent trends in recommendation technology strive for a more complete understanding of the user needs. This involves context aware recommenders that adapt to the particular given context, accounting for current time, location and user need. Some other systems use transfer learning methodologies for extending the user profile by also considering activities in different domains. Another approach for enhancing the user modeling is by following active learning practices and eliciting ratings and preferences from the user.
The design of a recommender involves the optimization of multiple, sometimes conflicting, objectives. Systems can target a point-wise error between each predicted value and the believed ground truth. However, more recent system opt to ranking metrics which emphasize the quality of the few items present at the top of the suggested list. Other tradeoffs that influence the nature of the recommended items are narrow accuracy versus broader diversity of item types, as well as staying with well understood and safe popular items versus admitting riskier long-tail items offering the potential of enhancing perceived serendipity.
In summary, recommendation system is a booming field, merging disciplines like data mining and machine learning, human-computer interaction, system scaling and more. It offers both scientific opportunities and practical engineering challenges. Practitioners who are interested in deeper knowledge are invited to visit the public resources listed below.
- ACM Conference Series on Recommender Systems
- ACM Recsys Wiki
- Coursera: Introduction to Recommender Systems
- Linkedin Group
- Twitter @ACMRecSys
- Wikipedia article
Related KDD2016 Papers
|Title & Authors|
|When Recommendation Goes Wrong - Anomalous Link Discovery in Recommendation Networks|
Author(s): Bryan Perozzi*, Stony Brook University
|Towards Conversational Recommender Systems|
Author(s): Konstantina Christakopoulou*, University of Minnesota; Katja Hofmann, Microsoft; Filip Radlinski, Microsoft
|Collaborative Knowledge Base Embedding for Recommender Systems|
Author(s): Fuzheng Zhang*, Microsoft; Nicholas Jing Yuan, Microsoft Research; Defu Lian, ; Xing Xie, Microsoft Research; Wei-Ying Ma,
|Point-of-Interest Recommendations: Learning Potential Check-ins from Friends|
Author(s): Yong Ge, UNC Charlotte; Huayu Li*, University of North Carolina a; Hengshu Zhu, Baidu Inc.
|Contextual Intent Tracking for Personal Assistants|
Author(s): Yu Sun*, University of Melbourne; Nicholas Jing Yuan, Microsoft Research; Yingzi Wang, Microsoft Research; Xing Xie, Microsoft Research; Kieran McDonald, Microsoft Corporation; Rui Zhang, University of Melbourne
|Goal-Directed Inductive Matrix Completion|
Author(s): Si Si*, Ut austin; Kai-Yang Chiang, UT Austin; Cho-Jui Hsieh, UT Austin; Nikhil Rao, Technicolor Research; Inderjit Dhillon, UTexas
|From Online Behaviors to Offline Retailing|
Author(s): Ping Luo*, Chinese Academy of Sciences
|An Empirical Study on Recommendation with Multiple Types of Feedback|
Author(s): Liang Tang*, LinkedIn Corp.; Bo Long, LinkedIn; Bee-Chung Chen, LinkedIn; Deepak Agarwal, LinkedIn
|The Limits of Popularity-Based Recommendations, and the Role of Social Ties|
Author(s): Marco Bressan*, Sapienza University of Rome; Stefano Leucci, Sapienza University of Rome; Alessandro Panconesi, Sapienza University of Rome; Prabhakar Raghavan, Google; Erisa Terolli, Sapienza University of Rome
|The Million Domain Challenge: Broadcast Email Prioritization by Cross-domain Recommendation|
Author(s): BEIDOU WANG*, Simon Fraser University; Martin Ester, Simon Fraser University; Yikang Liao, Zhejiang University; Jiajun Bu, Zhejiang University; Yu Zhu, Zhejiang University; Deng Cai, ; Ziyu Guan,
|Minimizing Legal Exposure for High-Tech Companies through Collaborative Filtering Methods|
Author(s): Bo Jin*, Dalian University of Technology; Chao Che, Dalian University; Kuifei Yu, Zhigu Inc.; Yue Qu, Dalian University of Technology; Li Guo, Dalian University of Technology; Cuili Yao, Dalian University of Technology
|Compute Job Memory Recommender System Using Machine Learning|
Author(s): Taraneh Taghavi*, Qualcomm Inc.; Maria Lupetini, Qualcomm Inc.; Yaron Kretchmer, Qualcomm Inc.
|Online Context-Aware Recommendation with Time Varying Multi-Arm Bandit|
Author(s): Chunqiu Zeng*, Florida International University; Qing Wang, Florida International Univ.; Tao Li, Florida International Univ; Shekoofeh Mokhtari, Florida International University
|Scalable Time-Decaying Adaptive Prediction Algorithm|
Author(s): Yinyan Tan*, Huawei; Zhe Fan, ; Guilin Li, ; Fangshan Wang, ; Zhengbing Li, ; Shikai Liu, ; Qiuling Pan, ; Eric Xing, CMU; Qirong Ho,
|CaSMoS: A Framework for Learning Candidate Selection Models over Structured Queries and Documents|
Author(s): Fedor Borisyuk*, LinkedIn; Krishnaram Kenthapadi, LinkedIn Corporation; David Stein, LinkedIn Corporation; Bo Zhao, LinkedIn Corporation
|Assessing Human Error Against a Benchmark of Perfection|
Author(s): Ashton Anderson*, Stanford University; Jon Kleinberg, Cornell University; Sendhil Mullainathan, Harvard
|Unified Point-of-Interest Recommendation with Temporal Interval Assessment|
Author(s): Yanchi Liu*, Rutgers University; Chuanren Liu, Drexel University; Bin Liu, Rutgers University; Meng Qu, Rutgers University; Hui Xiong, Rutgers
|Continuous Experience-aware Language Model|
Author(s): Subhabrata Mukherjee*, Max Planck Informatics; Stephan Günnemann, Technical University of Munich; Gerhard Weikum, Max Planck Institute for Informatics
|Positive-Unlabeled Learning in Streaming Networks|
Author(s): Shiyu Chang*, UIUC; Yang Zhang, UIUC; Jiliang Tang, Yahoo Labs; Dawei Yin, ; Yi Chang, Yahoo! Labs; Mark Hasegawa-Johnson, UIUC; Thomas Huang, UIUC
|Smart Reply: Automated Response Suggestion for Email|
Author(s): Anjuli Kannan, ; Karol Kurach*, Google; Sujith Ravi, Google; Tobias Kaufmann, Google, Inc.; Andrew Tomkins, ; Balint Miklos, Google, Inc.; Greg Corrado, ; László Lukács, ; Marina Ganea, ; Peter Young, ; Vivek Ramavajjala