Modeling Personalized Item Frequency Information for Next-basket Recommendation

SIGIR '20: The 43rd International ACM SIGIR conference on research and development in Information Retrieval Virtual Event China July, 2020, pp. 1071-1080, 2020.

Cited by: 1|Bibtex|Views131|DOI:https://doi.org/10.1145/3397271.3401066
EI
Other Links: arxiv.org|dl.acm.org|dblp.uni-trier.de|academic.microsoft.com
Weibo:
Next-basket recommendation is a type of recommendation problem that aims to recommend a set of items to a user based on his/her historical purchased baskets, which is prevalent in E-commerce and retail industry

Abstract:

Next-basket recommendation (NBR) is prevalent in e-commerce and retail industry. In this scenario, a user purchases a set of items (a basket) at a time. NBR performs sequential modeling and recommendation based on a sequence of baskets. NBR is in general more complex than the widely studied sequential (session-based) recommendation which ...More

Code:

Data:

0
Introduction
  • Recommendation systems have been applied in many different applications [1]. NBR is a type of recommendation problem that aims to recommend a set of items to a user based on his/her historical purchased baskets [36][45][44][46], which is prevalent in E-commerce and retail industry.
  • Considering the historical records, top-n recommendation and sequential/session-based recommendation can be seen as special cases of NBR when the NBR only has one basket and has a sequence of baskets whose size are all of 1, respectively.
  • Top-n recommendation only recommends new items that are not contained in the user’s historical records, whereas both sequential/session-based recommendation and NBR recommend new and old items.
  • Even though sequential/session-based recommendation is similar to NBR, the authors cannot directly apply sequential/session-based recommendation method to do NBR without messing up the information existing in the sequential sets1
Highlights
  • Recommendation systems have been applied in many different applications [1]
  • Next-basket recommendation (NBR) is a type of recommendation problem that aims to recommend a set of items to a user based on his/her historical purchased baskets [36][45][44][46], which is prevalent in E-commerce and retail industry
  • Unlike top-n recommendation [33] and sequential recommendation [31], the historical record of the next-basket recommendation is a sequence of sets or sequential sets
  • We propose a simple k-nearest neighbors based method which directly captures the two useful patterns associated with personalized item frequency} (PIF)
  • We propose a simple and effective k-nearest neighbors (kNN) based method to directly capture the two useful patterns associated with PIF
  • Our method frequently outperforms the state-of-the-art NBR methods -- including deep learning based methods using Recurrent neural network (RNN) -- when patterns associated with PIF play an important role in the data
  • We introduce a simple kNN-based method8
Methods
  • Simple baselines:

    Top-n frequent (TopFreq): It uses the most frequent s items that appear in all the baskets of the training data as the predicted baskets for all persons.

    Personalized Top-n frequent (PersonTopFreq): It uses the most frequent s items that appear in the past baskets of a given person as the prediction for the basket.
  • This baseline can show the difference between the proposed method and the existing user-based kNN method
Results
  • The authors use recall and NDCG to evaluate the methods.
  • NDCG is a ranking based measure which takes into account the order of elements in a list [15].
  • The authors calculate the NDCG for each basket based on the top s sorted elements list.
  • All the data sets are partitioned across users.
  • The authors reserve the data of 10% users in the training set as the validation set for hyper parameters searching in all the methods
Conclusion
  • The authors introduce a simple kNN-based method8.
  • The proposed method generally outperforms the stateof-the art deep learning based methods.
  • The authors study the reason why RNNs cannot approximate vector addition well, which provides the insight why the proposed method can outperform existing methods.
  • More theoretical analysis is needed.
  • A new optimizer that has the theoretical guarantee to find the global optimal like [43] is needed for RNNs
Summary
  • Introduction:

    Recommendation systems have been applied in many different applications [1]. NBR is a type of recommendation problem that aims to recommend a set of items to a user based on his/her historical purchased baskets [36][45][44][46], which is prevalent in E-commerce and retail industry.
  • Considering the historical records, top-n recommendation and sequential/session-based recommendation can be seen as special cases of NBR when the NBR only has one basket and has a sequence of baskets whose size are all of 1, respectively.
  • Top-n recommendation only recommends new items that are not contained in the user’s historical records, whereas both sequential/session-based recommendation and NBR recommend new and old items.
  • Even though sequential/session-based recommendation is similar to NBR, the authors cannot directly apply sequential/session-based recommendation method to do NBR without messing up the information existing in the sequential sets1
  • Objectives:

    Given the historical purchase records of a user {v1, v2, ..., vi , ..., vt }, where a set of items at the i-th time step is represented as a 0/1 vector vi whose entry cj (j ∈ [0, d]) is set to 1 if the corresponding item appears in the basket, the goal is to predict the set of items vt+1.
  • As the goal is to learn a linear i =1 operation addition, the nonlinear layer is not necessary
  • Methods:

    Simple baselines:

    Top-n frequent (TopFreq): It uses the most frequent s items that appear in all the baskets of the training data as the predicted baskets for all persons.

    Personalized Top-n frequent (PersonTopFreq): It uses the most frequent s items that appear in the past baskets of a given person as the prediction for the basket.
  • This baseline can show the difference between the proposed method and the existing user-based kNN method
  • Results:

    The authors use recall and NDCG to evaluate the methods.
  • NDCG is a ranking based measure which takes into account the order of elements in a list [15].
  • The authors calculate the NDCG for each basket based on the top s sorted elements list.
  • All the data sets are partitioned across users.
  • The authors reserve the data of 10% users in the training set as the validation set for hyper parameters searching in all the methods
  • Conclusion:

    The authors introduce a simple kNN-based method8.
  • The proposed method generally outperforms the stateof-the art deep learning based methods.
  • The authors study the reason why RNNs cannot approximate vector addition well, which provides the insight why the proposed method can outperform existing methods.
  • More theoretical analysis is needed.
  • A new optimizer that has the theoretical guarantee to find the global optimal like [43] is needed for RNNs
Tables
  • Table1: The importance of two patterns
  • Table2: Statistic information after pre-processing
  • Table3: Parameters of our methods in different data sets
  • Table4: Comparison with different methods. The bold is the maximum in (a)-(b). The underline is the maximum in (c)-(h)
  • Table5: The effect of each component in the TIFU-KNN
  • Table6: Sensitivity of hyperparameters: time-decayed ratio rb within each group and time-decayed ratio rд across the groups at Instacart data set
Download tables as Excel
Related work
  • The related works include (1) Traditional collaborative recommendation methods that model user preferences without considering the temporal dynamics and have a set of items as the historical record; (2) Sequential recommendation methods that deal with a sequence of items or actions (each element is an item or action) as the user profile; and (3) NBR methods that deal with a sequence of baskets (each element is a set of items) as the user profile. Traditional collaborative recommendation: Collaborative Filtering (CF) [37] is the classical recommendation method. CF usually learns from user-item ratings matrix and predict only based on this matrix. Existing CF methods can be classified into two categories: neighborhood- and model-based methods. Neighborhood-based methods are widely studied in traditional collaborative recommendation [33]. The neighborhood-based methods contain two ways: user-based or item-based recommendation. User-based method like GroupLens [26] predicts the interest of a target user for an item using the ratings for this item by the most similar users. The itembased method like itemKNN [10] predicts the user-item rating based on the ratings of the target user for similar items. Model-based approaches use these ratings to learn a predictive model [27][4]. Recently, neural networks-based methods are proposed to enhance the CF methods [17][29][16] as more nonlinear relations can be captured by neural networks. Sequential recommendation: The goal of sequential recommendation is to recommend the next item or action based on the past sequence of items [14][24][13][47]. Due to the sequence structure in the historical records, natural language processing methods, like RNNs, attention mechanism, and Markov chain, can be applied to model the data [14][24][18]. Session-based Recommendation also belongs to this type as each session is a short sequence of behaviors or items [30][23]. A kNN-based method shows competitive performance when it is compared to RNN-based method GRU4rec [23]. Our kNN-based method is different from this method in both similarity calculation step and prediction step. Also, their method cannot be directly applied to NBR as discussed in the introduction. Next-basket recommendation: NBR aims at predicting a set of items based on a sequence of past baskets (sets) [36][44][46][45][3][20]. The summary can be found in section 2.3. Unlike traditional collaborative recommendation and sequential recommendation, the study towards kNN-based method on NBR is lacked. There is no clue if this type of methods can provide better performance. Our proposed kNN-based method fills this gap. Difficulty in Training RNNs: The vanishing and the exploding gradient problems are the well-known difficulty in training RNNs [34]. We present another phenomenon that it is difficult for RNNs to learn a simple operation—vector addition. Even though we know training a deep neural network is np-complete in the worst case [6], the phenomenon discovered in this paper is different as we provide a closed-form solution. There is a need for more theoretical analysis to understanding this kind of difficulty in training RNNs.
Reference
  • Charu C Aggarwal et al. 2016. Recommender systems. Springer.
    Google ScholarFindings
  • Ashton Anderson, Ravi Kumar, Andrew Tomkins, and Sergei Vassilvitskii. 2014. The dynamics of repeat consumption. In Proceedings of the 23rd international conference on World wide web. ACM, 419–430.
    Google ScholarLocate open access versionFindings
  • Ting Bai, Jian-Yun Nie, Wayne Xin Zhao, Yutao Zhu, Pan Du, and Ji-Rong Wen.
    Google ScholarFindings
  • 2018. An attribute-aware neural attentive model for next basket recommendation. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 1201–1204.
    Google ScholarLocate open access versionFindings
  • [4] Robert Bell, Yehuda Koren, and Chris Volinsky. 2007. Modeling relationships at multiple scales to improve accuracy of large recommender systems. In Proceedings of the 13th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 95–104.
    Google ScholarLocate open access versionFindings
  • [5] Austin R Benson, Ravi Kumar, and Andrew Tomkins. 201Modeling user consumption sequences. In Proceedings of the 25th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 519–529.
    Google ScholarLocate open access versionFindings
  • [6] Avrim Blum and Ronald L Rivest. 1989. Training a 3-node neural network is NP-complete. In Advances in neural information processing systems. 494–501.
    Google ScholarLocate open access versionFindings
  • [7] Léon Bottou. 2010. Large-scale machine learning with stochastic gradient descent. In Proceedings of COMPSTAT’2010. Springer, 177–186.
    Google ScholarLocate open access versionFindings
  • [8] Kyunghyun Cho, Bart Van Merriënboer, Dzmitry Bahdanau, and Yoshua Bengio. 2014. On the properties of neural machine translation: Encoder-decoder approaches. arXiv preprint arXiv:1409.1259 (2014).
    Findings
  • [9] John Dawes, Lars Meyer-Waarden, and Carl Driesener. 2015. Has brand loyalty declined? A longitudinal analysis of repeat purchase behavior in the UK and the USA. Journal of Business Research 68, 2 (2015), 425–432.
    Google ScholarLocate open access versionFindings
  • [10] Mukund Deshpande and George Karypis. 2004. Item-based top-n recommendation algorithms. ACM Transactions on Information Systems (TOIS) 22, 1 (2004), 143–177.
    Google ScholarLocate open access versionFindings
  • [11] Ian J Goodfellow, Oriol Vinyals, and Andrew M Saxe. 2014. Qualitatively characterizing neural network optimization problems. arXiv preprint arXiv:1412.6544 (2014).
    Findings
  • [12] Klaus Greff, Rupesh K Srivastava, Jan Koutník, Bas R Steunebrink, and Jürgen Schmidhuber. 2016. LSTM: A search space odyssey. IEEE transactions on neural networks and learning systems 28, 10 (2016), 2222–2232.
    Google ScholarLocate open access versionFindings
  • [13] Ruining He, Wang-Cheng Kang, and Julian McAuley. 2018. Translation-based Recommendation: A Scalable Method for Modeling Sequential Behavior.. In IJCAI. 5264–5268.
    Google ScholarFindings
  • [14] Ruining He and Julian McAuley. 2016. Fusing similarity models with markov chains for sparse sequential recommendation. In 2016 IEEE 16th International Conference on Data Mining. IEEE, 191–200.
    Google ScholarLocate open access versionFindings
  • [15] Xiangnan He, Tao Chen, Min-Yen Kan, and Xiao Chen. 2015. Trirank: Reviewaware explainable recommendation by modeling aspects. In Proceedings of the 24th ACM International on Conference on Information and Knowledge Management. ACM, 1661–1670.
    Google ScholarLocate open access versionFindings
  • [16] Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation. In The 43st International ACM SIGIR Conference on Research & Development in Information Retrieval.
    Google ScholarLocate open access versionFindings
  • [17] Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. In Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 173–182.
    Google ScholarLocate open access versionFindings
  • [18] Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk. 2015. Session-based recommendations with recurrent neural networks. arXiv preprint arXiv:1511.06939 (2015).
    Findings
  • [19] Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation 9, 8 (1997), 1735–1780.
    Google ScholarLocate open access versionFindings
  • [20] Haoji Hu and Xiangnan He. 2019. Sets2Sets: Learning from Sequential Sets with Neural Networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 1491–1499.
    Google ScholarLocate open access versionFindings
  • [21] Wei Hu and Jeffrey Pennington. 2020. Provable Benefit of Orthogonal Initialization in Optimizing Deep Linear Networks. In International Conference on Learning Representations.
    Google ScholarLocate open access versionFindings
  • [22] Jacob Jacoby and David B Kyner. 1973. Brand loyalty vs. repeat purchasing behavior. Journal of Marketing research 10, 1 (1973), 1–9.
    Google ScholarLocate open access versionFindings
  • [23] Dietmar Jannach and Malte Ludewig. 2017. When recurrent neural networks meet the neighborhood for session-based recommendation. In Proceedings of the Eleventh ACM Conference on Recommender Systems. ACM, 306–310.
    Google ScholarLocate open access versionFindings
  • [24] Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recommendation. In 2018 IEEE International Conference on Data Mining (ICDM). IEEE, 197–206.
    Google ScholarLocate open access versionFindings
  • [25] Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014).
    Findings
  • [26] Joseph A Konstan, Bradley N Miller, David Maltz, Jonathan L Herlocker, Lee R Gordon, and John Riedl. 1997. GroupLens: applying collaborative filtering to Usenet news. Commun. ACM 40, 3 (1997), 77–87.
    Google ScholarLocate open access versionFindings
  • [27] Yehuda Koren. 2008. Factorization meets the neighborhood: a multifaceted collaborative filtering model. In Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 426–434.
    Google ScholarLocate open access versionFindings
  • [28] Yehuda Koren. 2009. Collaborative filtering with temporal dynamics. In Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 447–456.
    Google ScholarLocate open access versionFindings
  • [29] Dawen Liang, Rahul G Krishnan, Matthew D Hoffman, and Tony Jebara. 2018. Variational autoencoders for collaborative filtering. In Proceedings of the 2018 World Wide Web Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 689–698.
    Google ScholarLocate open access versionFindings
  • [30] Malte Ludewig and Dietmar Jannach. 2018. Evaluation of session-based recommendation algorithms. User Modeling and User-Adapted Interaction 28, 4-5 (2018), 331–390.
    Google ScholarLocate open access versionFindings
  • [31] Malte Ludewig, Noemi Mauro, Sara Latifi, and Dietmar Jannach. 2019. Performance comparison of neural and non-neural approaches to session-based recommendation. In Proceedings of the 13th ACM Conference on Recommender Systems. ACM, 462–466.
    Google ScholarLocate open access versionFindings
  • [32] Danilo P Mandic and Jonathon Chambers. 2001. Recurrent neural networks for prediction: learning algorithms, architectures and stability. John Wiley & Sons, Inc.
    Google ScholarFindings
  • [33] Xia Ning, Christian Desrosiers, and George Karypis. 2015. A comprehensive survey of neighborhood-based recommendation methods. In Recommender systems handbook. Springer, 37–76.
    Google ScholarFindings
  • [34] Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. 2013. On the difficulty of training recurrent neural networks. In International conference on machine learning. 1310–1318.
    Google ScholarLocate open access versionFindings
  • [35] Pengjie Ren, Zhumin Chen, Jing Li, Zhaochun Ren, Jun Ma, and Maarten de Rijke. 2019. RepeatNet: A repeat aware neural recommendation machine for session-based recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 4806–4813.
    Google ScholarLocate open access versionFindings
  • [36] Steffen Rendle, Christoph Freudenthaler, and Lars Schmidt-Thieme. 2010. Factorizing personalized markov chains for next-basket recommendation. In Proceedings of the 19th international conference on World wide web. ACM, 811–820.
    Google ScholarLocate open access versionFindings
  • [37] Francesco Ricci, Lior Rokach, and Bracha Shapira. 2011. Introduction to recommender systems handbook. In Recommender systems handbook. Springer, 1–35.
    Google ScholarLocate open access versionFindings
  • [38] Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl. 2001. Item-based collaborative filtering recommendation algorithms. In Proceedings of the 10th international conference on World Wide Web. 285–295.
    Google ScholarLocate open access versionFindings
  • [39] Mohit Sharma and George Karypis. 2019. Adaptive matrix completion for the users and the items in tail. In The World Wide Web Conference. ACM, 3223–3229.
    Google ScholarLocate open access versionFindings
  • [40] Hava T Siegelmann and Eduardo D Sontag. 1992. On the computational power of neural nets. In Proceedings of the fifth annual workshop on Computational learning theory. ACM, 440–449.
    Google ScholarLocate open access versionFindings
  • [41] Mengting Wan, Di Wang, Jie Liu, Paul Bennett, and Julian McAuley. 2018. Representing and Recommending Shopping Baskets with Complementarity, Compatibility and Loyalty. In Proceedings of the 27th ACM International Conference on Information and Knowledge Management. ACM, 1133–1142.
    Google ScholarLocate open access versionFindings
  • [42] Chenyang Wang, Min Zhang, Weizhi Ma, Yiqun Liu, and Shaoping Ma. 2019. Modeling Item-Specific Temporal Dynamics of Repeat Consumption for Recommender Systems. In The World Wide Web Conference. 1977–1987.
    Google ScholarFindings
  • [43] Gang Wang, Georgios B Giannakis, and Jie Chen. 2019. Learning ReLU networks on linearly separable data: Algorithm, optimality, and generalization. IEEE Transactions on Signal Processing 67, 9 (2019), 2357–2370.
    Google ScholarLocate open access versionFindings
  • [44] Pengfei Wang, Jiafeng Guo, Yanyan Lan, Jun Xu, Shengxian Wan, and Xueqi Cheng. 2015. Learning hierarchical representation model for next-basket recommendation. In Proceedings of the 38th International ACM SIGIR conference on Research and Development in Information Retrieval. ACM, 403–412.
    Google ScholarLocate open access versionFindings
  • [45] Haochao Ying, Fuzhen Zhuang, Fuzheng Zhang, Yanchi Liu, Guandong Xu, Xing Xie, Hui Xiong, and Jian Wu. 2018. Sequential recommender system based on hierarchical attention networks. In the 27th International Joint Conference on Artificial Intelligence.
    Google ScholarLocate open access versionFindings
  • [46] Feng Yu, Qiang Liu, Shu Wu, Liang Wang, and Tieniu Tan. 2016. A dynamic recurrent model for next basket recommendation. In Proceedings of the 39th International ACM SIGIR conference on Research and Development in Information Retrieval. ACM, 729–732.
    Google ScholarLocate open access versionFindings
  • [47] Fajie Yuan, Xiangnan He, Haochuan Jiang, Guibing Guo, Jian Xiong, Zhezhao Xu, and Yilin Xiong. 2020. Future Data Helps Training: Modeling Future Contexts for Session-based Recommendation. In Proceedings of The Web Conference 2020. 303–313.
    Google ScholarLocate open access versionFindings
Full Text
Your rating :
0

 

Tags
Comments