Rating Prediction with Informative Ensemble of Multi-Resolution Dynamic Models

Zhao Zheng, Tianqi Chen, Nathan Liu, Qiang Yang, Yong Yu
Proceedings of KDD Cup 2011, PMLR 18:75-97, 2012.

Abstract

The Yahoo! music rating data set in KDD Cup 2011 raises several interesting challenges: (1) The data covers a lengthy time period of more than eight years. (2) Not only are training ratings associated date and time information, so are the test ratings. (3) The items form a hierarchy consisting of four types of items: genres, artists, albums and tracks. To capture the rich temporal dynamics within the data set, we design a class of time-aware matrix/tensor factorization models, which adopts time series based parameterizations and models user/item drifting behaviors at multiple temporal resolutions. We also incorporate the taxonomical structure into the item parameters by introducing sharing parameters between ancestors and descendants in the taxonomy. Finally, we have identified some conditions that systematically affect the effectiveness of different types of models and parameter settings. Based on these findings, we designed an informative ensemble framework, which considers additional meta features when making predictions for a particular pair of user and item. Using these techniques, we built the best single model reported officially, and our final ensemble model got third place in KDD Cup 2011.

Cite this Paper


BibTeX
@InProceedings{pmlr-v18-zheng12a, title = {Rating Prediction with Informative Ensemble of Multi-Resolution Dynamic Models}, author = {Zheng, Zhao and Chen, Tianqi and Liu, Nathan and Yang, Qiang and Yu, Yong}, booktitle = {Proceedings of KDD Cup 2011}, pages = {75--97}, year = {2012}, editor = {Dror, Gideon and Koren, Yehuda and Weimer, Markus}, volume = {18}, series = {Proceedings of Machine Learning Research}, month = {21 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v18/zheng12a/zheng12a.pdf}, url = {https://proceedings.mlr.press/v18/zheng12a.html}, abstract = {The Yahoo! music rating data set in KDD Cup 2011 raises several interesting challenges: (1) The data covers a lengthy time period of more than eight years. (2) Not only are training ratings associated date and time information, so are the test ratings. (3) The items form a hierarchy consisting of four types of items: genres, artists, albums and tracks. To capture the rich temporal dynamics within the data set, we design a class of time-aware matrix/tensor factorization models, which adopts time series based parameterizations and models user/item drifting behaviors at multiple temporal resolutions. We also incorporate the taxonomical structure into the item parameters by introducing sharing parameters between ancestors and descendants in the taxonomy. Finally, we have identified some conditions that systematically affect the effectiveness of different types of models and parameter settings. Based on these findings, we designed an informative ensemble framework, which considers additional meta features when making predictions for a particular pair of user and item. Using these techniques, we built the best single model reported officially, and our final ensemble model got third place in KDD Cup 2011.} }
Endnote
%0 Conference Paper %T Rating Prediction with Informative Ensemble of Multi-Resolution Dynamic Models %A Zhao Zheng %A Tianqi Chen %A Nathan Liu %A Qiang Yang %A Yong Yu %B Proceedings of KDD Cup 2011 %C Proceedings of Machine Learning Research %D 2012 %E Gideon Dror %E Yehuda Koren %E Markus Weimer %F pmlr-v18-zheng12a %I PMLR %P 75--97 %U https://proceedings.mlr.press/v18/zheng12a.html %V 18 %X The Yahoo! music rating data set in KDD Cup 2011 raises several interesting challenges: (1) The data covers a lengthy time period of more than eight years. (2) Not only are training ratings associated date and time information, so are the test ratings. (3) The items form a hierarchy consisting of four types of items: genres, artists, albums and tracks. To capture the rich temporal dynamics within the data set, we design a class of time-aware matrix/tensor factorization models, which adopts time series based parameterizations and models user/item drifting behaviors at multiple temporal resolutions. We also incorporate the taxonomical structure into the item parameters by introducing sharing parameters between ancestors and descendants in the taxonomy. Finally, we have identified some conditions that systematically affect the effectiveness of different types of models and parameter settings. Based on these findings, we designed an informative ensemble framework, which considers additional meta features when making predictions for a particular pair of user and item. Using these techniques, we built the best single model reported officially, and our final ensemble model got third place in KDD Cup 2011.
RIS
TY - CPAPER TI - Rating Prediction with Informative Ensemble of Multi-Resolution Dynamic Models AU - Zhao Zheng AU - Tianqi Chen AU - Nathan Liu AU - Qiang Yang AU - Yong Yu BT - Proceedings of KDD Cup 2011 DA - 2012/06/01 ED - Gideon Dror ED - Yehuda Koren ED - Markus Weimer ID - pmlr-v18-zheng12a PB - PMLR DP - Proceedings of Machine Learning Research VL - 18 SP - 75 EP - 97 L1 - http://proceedings.mlr.press/v18/zheng12a/zheng12a.pdf UR - https://proceedings.mlr.press/v18/zheng12a.html AB - The Yahoo! music rating data set in KDD Cup 2011 raises several interesting challenges: (1) The data covers a lengthy time period of more than eight years. (2) Not only are training ratings associated date and time information, so are the test ratings. (3) The items form a hierarchy consisting of four types of items: genres, artists, albums and tracks. To capture the rich temporal dynamics within the data set, we design a class of time-aware matrix/tensor factorization models, which adopts time series based parameterizations and models user/item drifting behaviors at multiple temporal resolutions. We also incorporate the taxonomical structure into the item parameters by introducing sharing parameters between ancestors and descendants in the taxonomy. Finally, we have identified some conditions that systematically affect the effectiveness of different types of models and parameter settings. Based on these findings, we designed an informative ensemble framework, which considers additional meta features when making predictions for a particular pair of user and item. Using these techniques, we built the best single model reported officially, and our final ensemble model got third place in KDD Cup 2011. ER -
APA
Zheng, Z., Chen, T., Liu, N., Yang, Q. & Yu, Y.. (2012). Rating Prediction with Informative Ensemble of Multi-Resolution Dynamic Models. Proceedings of KDD Cup 2011, in Proceedings of Machine Learning Research 18:75-97 Available from https://proceedings.mlr.press/v18/zheng12a.html.

Related Material