计算机科学 ›› 2023, Vol. 50 ›› Issue (7): 66-71.doi: 10.11896/jsjkx.220900125
朱文韬1,3, 刘威1,3, 梁上松1,3, 朱怀杰2,3, 印鉴2,3
ZHU Wentao1,3, LIU Wei1,3, LIANG Shangsong1,3, ZHU Huaijie2,3, YIN Jian2,3
摘要: 元学习方法近年被引入推荐系统以缓解冷启动问题。现有元学习算法只能提高算法处理一组静态分布的数据集(任务)的能力。当面对多个服从非平稳分布的数据集时,现有模型往往会出现负知识转移以及灾难性遗忘问题,导致算法推荐性能大幅下降。探索了基于变分持续贝叶斯元学习(Variational Continuous Bayesian Meta-Learning,VC-BML)的推荐算法。首先,算法假设元参数服从动态混合高斯模型,使其具有更大的参数空间,提高了模型适应不同任务的能力,缓解了负知识转移问题。然后,VC-BML的任务集群数量由中国餐馆过程(Chinese Restaurant Process,CRP)来灵活确定,使得模型在不同的混合分量中存储不同任务分布的知识,并在类似任务出现时调用这些知识,有助于缓解传统算法中的灾难性遗忘问题。为了估计模型参数的后验概率,算法采用了一种更稳健的结构化变分推理方法来近似后验值,以避免遗忘知识。最后,VC-BML在4个非平稳分布的数据集上的表现均优于基准算法。与基于点估计的基准算法相比,VC-BML提高了模型的稳健型,有助于缓解灾难性遗忘问题。
中图分类号:
[1]MCNEE S M,RIEDL J,KONSTAN J A.Being accurate is not enough:how accuracy metrics have hurt recommender systems[C]//Extended Abstracts(CHI’06).2006:1097-1101. [2]LI X,SHE J.Collaborative variational autoencoder for recommender systems[C]//KDD.2017:305-314. [3]OORD A V D,DIELEMAN S,SCHRAUWEN B.Deep content-based music recommendation[C]//NIPS.2013:2643-2651. [4]WEI J,HE J H,CHEN K,et al.Collaborative filtering and deeplearning based recommendation system for cold start items[J].Expert Systems with Applications,2017,69:29-39. [5]LEE H,IM J,JANG S,et al.Melu:Meta-learned user preferenceestimator for cold-start recommendation[C]//KDD.2019:1073-1082. [6]DENEVI G,STAMOS D,CILIBERTO C,et al.Online-within-online meta-learning[C]//NIPS.2019:1-11. [7]VILALTA R,DRISSI Y.A perspective view and survey ofmeta-learning[J].Artificial Intelligence Review,2002,18(2):77-95. [8]FINN C,ABBEEL P,LEVINE S.Model-agnostic meta-learning for fast adaptation of deep networks[C]//ICML.2017:1126-1135. [9]LIANG S S,YILMAZ E,KANOULAS E.Collaboratively tra-cking interests for user clustering in streams of short texts[J].IEEE Transactions on Knowledge Data Engineering,2018,31(2):257-272. [10]ZHANG Q,FANG J Y,MENG Z Q,et al.Varisational Conti-nual Bayesian Meta-Learning[C]//NIPS.2021:24556-24568. [11]CHENG H T,KOC L,HARMSEN J,et al.Wide deep learning for recommender systems[C]//DLRS.2016:7-10. [12]HE X N,LIAO L Z,ZHANG H W,et al.Neural collaborative filtering[C]//WWW.2017:173-182. [13]BARKAN O,KOENIGSTEIN N.Item2vec:Neural item embedding for collaborative filtering[C]//IEEE MLSP.2016:1-6. [14]ZHANG F Z,YUAN N J,LIAN D F,et al.Collaborative know-ledge base embedding for recommender systems[C]//KDD.2016:353-362. [15]NAIR V,EHINTON G.Rectified linear units improve restricted boltzmann machines vinodnair[C]//ICML-10.2010:807-814. [16]FINN C,RAJESWARAN A,KAKADE S,et al.Online meta-learning[EB/OL].(2019-02-22) [2019-07-03].https://doi.org/10.48550/arXiv.1902.08438. [17]KALAI A,VEMPALA S.Efficient algorithms for online decision problems[J].Journal of Computer and System Sciences,2005,71(3):291-307. [18]JERFEL G,GRANT E,GRIFFITHS T,et al.Reconciling meta-learning and continual learning with online mixtures of tasks[EB/OL].(2018-12-14) [2019-06-19].https://doi.org/10.48550/arXiv.1812.06080. [19]YAP P,RITTER H,BARBER D.Bayesian online meta-lear-ning with Online Laplace Approximation[J].arXiv:2005.00146,2020. |
|