A Personalized Interest-Forgetting Markov Model for Recommendations Jun Chen , Chaokun Wang, Jianmin Wang Tsinghua University, China chenjun14@mails.tsinghua.edu.cn, {chaokun, jimwang}@tsinghua.edu.cn AAAI-15 28-Jan-15 1
Review on Forgetting Curve (FC) Memory Forgetting Ebbinghaus FC More FCs An intelligent Starting experience recommender system? Forgetting speeds A Personalized Interest-Forgetting Markov Model for Recommendations 2
Forgetting of User Interests ο§ Interest-Forgetting ο§ User π£ βs interest upon item π¦ loses as time elapses after the consumption. ο§ Importance to influence current user interest. ο§ Some issues ο§ Modeling of interest-forgetting. ο§ Personalization ο§ Forgetting speeds ο§ Starting experience ο§ Re-learning/Reconsumption A Personalized Interest-Forgetting Markov Model for Recommendations 3
Major Contributions 1. We considered the interest-forgetting in recommendations towards a β human-minded β recommender system. 2. A personalized framework for interest-forgetting Markov model with multiple implementations on experience and interest retention. 3. An effective solution to item recommendation problem compared with the state-of-the-art. A Personalized Interest-Forgetting Markov Model for Recommendations 4
Related Works ο§ Markov model & Recommendation ο§ First-order Markov chain model [Rendle et al 2010, Cheng et al 2013] . ο§ High-order Markov model [Raftery 1985] ο§ Variable-order Markov models [Begleiter et al 2004, Dimitrakakis 2010] . ο§ Memory Forgetting & Learning ο§ Forgetting models [Ebbinghaus 1885, Nembhard et al 2001, Averell et al 2011] . ο§ Learning models [Jaber et al 1997, Anzanello et al 2011] ο§ Data filtering & updating [Packer et al 2011, Freedman et al 2011] A Personalized Interest-Forgetting Markov Model for Recommendations 5
Problem Formulation ο§ Variable-Order Markov (VOM) model based Recommendation Given an item trace π΄ π£,π’ = {π¦ 1 π£ , π¦ 2 π£ , β¦ , π¦ π’ π£ } of user π£ , recommend Top-N ο§ unseen items with the largest transition probability: π£ = π¦ 1 π π¦|π΄ π£,π’ = π(X π’+1 π£ = π¦ π’ π£ π£ , β¦ , X 1 π£ ) = π¦|X π’ ο§ Exponential expansion on the number of states ο§ π -VOM ο§ Step-wise weighted first-order Markov model π’ π’ π£,π’ π X π’+1 π£,π’ π π¦ π¦ π’+1βπ π π¦|π΄ π£,π’ = π£ π£ π£ π£ π π = π¦ X π’+1βπ = π¦ π’+1βπ = π π π=1 π=1 A Personalized Interest-Forgetting Markov Model for Recommendations 6
Framework ο§ π -VOM π’ π’ π£,π’ π π¦ π¦ π’+1βπ π π¦|π΄ π£,π’ = π£,π’ π£,π’ π£ π£ π π = Ξ₯ π¦ π’+1βπ Ξ¦ π¦ π’+1βπ π π¦ π¦ π’+1βπ π£ π£ π=1 π=1 π£ π π¦ π¦ π’+1βπ : one-step transition probability. π£,π’ : personalized interest-forgetting component. π π π£,π’ = Ξ₯ π¦ π’+1βπ π£,π’ π£,π’ π π Ξ¦ π¦ π’+1βπ π£ π£ π£,π’ β π π¦, π£, π’ , monotonically increasing with frequency. οΆ Starting Experience: Ξ₯ π¦ π£,π’ β 1/π , monotonically decreasing with elapsed time steps. οΆ Interest Retention: Ξ¦ π¦ A Personalized Interest-Forgetting Markov Model for Recommendations 7
IFMM Framework ο§ Objective ο§ Minimize the negative log-likelihood of the probabilities to recommend the last item in each training trace. Parameters Ξ β could be learned via stochastic gradient descent method. ο§ ο§ One-step transition probability can be directly computed. A Personalized Interest-Forgetting Markov Model for Recommendations 8
Framework Specifications ο§ One-Step Transition Probability ο§ Conditional probability of observing π¦ π after π¦ π . A Personalized Interest-Forgetting Markov Model for Recommendations 9
Framework Specifications ο§ Starting Experience π£,π’ β π π¦, π£, π’ , Starting Experience: Ξ₯ π¦ ο§ monotonically increasing with frequency. Logistic function π£,π’ Ξ₯ π¦ ο§ Rational function (normalized frequencies) π π£,π’ (π¦) Starting experience measures the personalized accumulative interest a user has upon a certain item before forgetting. Experience Curves A Personalized Interest-Forgetting Markov Model for Recommendations 10
Framework Specifications ο§ Interest Retention π£,π’ β 1/π , Interest Retention: Ξ¦ π¦ ο§ Log-Linear function [Wright 1936] monotonically decreasing with elapsed time. ο§ Exponential function [Knecht 1974] π£,π’ Ξ¦ π¦ ο§ Hypobolic function [Mazur and Hastie 1978] π Interest retention measures the personalized residual Interest Retention Curves interest of a user upon a certain item after forgetting. A Personalized Interest-Forgetting Markov Model for Recommendations 11
Personalized Recommendation ο§ IFMM Framework π’ π π¦|π΄ π£,π’ = π£,π’ π£,π’ π£ Ξ₯ π¦ π’+1βπ Ξ¦ π¦ π’+1βπ π π¦ π¦ π’+1βπ π£ π£ π=1 ο§ Forgetting speeds ο§ Starting experience ο§ Re-learning/Reconsumption Top-N item recommendation with the largest values of π π¦|π΄ π£,π’ . A Personalized Interest-Forgetting Markov Model for Recommendations 12
Experiments ο§ Data Set ο§ Last.fm music listening data set. ο§ 992 users, 964,464 songs, 16,986,614 listening records. ο§ Partition each userβs listening history with a time shreshold, e.g. 1 hour. ο§ Remove listening records whose duration is less than 30 secs. ο§ 80% traces for training, 20% traces for test. ο§ Comparative Methods ο§ Markov model based ο§ Factorizing Personalized Markov Chain (FPMC) [Rendle 2010, Cheng 2013] ο§ Topic Sensitive PageRank (TSPR) [Haveliwala 2002] ο§ Graph-based preference fusion (STG) [Xiang 2010] ο§ Sequential pattern based (SEQ) [Hariri et al 2012] A Personalized Interest-Forgetting Markov Model for Recommendations 13
Experiments ο§ Accuracy of the proposed methods ο§ Starting Experience ο§ NM: rational function ο§ NO: logistic function ο§ Interest Retention ο§ LL: log-linear ο§ EX: exponential ο§ HY: hypobolic NO+HY performs the best, and is selected as the representative. A Personalized Interest-Forgetting Markov Model for Recommendations 14
Experiments ο§ Accuracy Comparisons ο§ NO+HY ο§ SEQ ο§ - s5w4 : sup 5 , winsize 4 ο§ - s7w3 : sup 7 , winsize 3 ο§ - s6w2 : sup 6 , winsize 2 ο§ FPMC ο§ STG ο§ TSPR NO+HY improves 10%-20% in recommendation accuracy compared with the best of the reference methods. A Personalized Interest-Forgetting Markov Model for Recommendations 15
Experiments ο§ Personalized parameters distribution ο§ NO+HY π π£ π· π£ π½ π£ A Personalized Interest-Forgetting Markov Model for Recommendations 16
Conclusions ο§ Forgetting is an intrinsic feature of human beings, and should be taken into account in recommender systems. ο§ We proposed π -VOM to simplify the computation of variable-order Markov model. ο§ We brought forward a personalized framework which integrates interest-forgetting and Markov model. ο§ Multiple forgetting curve models and experience models have been evaluated under our framework to find an optimal solution. ο§ IFMM provides various strategies for personalization . ο§ The experimental results proved the effectiveness of our method in recommendation tasks. A Personalized Interest-Forgetting Markov Model for Recommendations 17
A Personalized Interest-Forgetting Markov Model for Recommendations Thank You ~ Any Question? Jun Chen , Chaokun Wang, Jianmin Wang Tsinghua University, China chenjun14@mails.tsinghua.edu.cn, {chaokun, jimwang}@tsinghua.edu.cn AAAI-15 28-Jan-15 18
Experiments ο§ Timeout Threshold ο§ Influence general length of traces. ο§ Larger value, longer traces. Very slight impact upon the recommendation accuracy A Personalized Interest-Forgetting Markov Model for Recommendations 19
Recommend
More recommend