ABSTRACT
E-commerce has seen tremendous growth over the past few years, so much so that only those companies which analyze browsing behaviour of users, can hope to survive the stiff competition in market. Analyzing customer behaviour helps in modeling and recognizing purchase intent which is vital to e-commerce for providing improved personalization and better ranking of search results. In this work, we make use of user clickstreams to model browsing behaviour of users. But clickstreams are known to be noisy and hence generating features from clickstreams and using them in one go for building a predictive model may not always capture the purchase/intent characteristics. There are multiple aspects within clickstreams which are to be considered such as the sequence (path) and temporal behaviour. Hence we model clickstreams as having multiple views, each view, concentrating on an aspect or a component of clickstream. In this work, we develop a Multi-View learning (MVL) framework that predicts whether users would make a purchase or not by analyzing their clickstreams. Recent advances in deep learning allow us to build neural networks that are able to extract complex latent features from the data with minimal human intervention. Separate models known as experts are trained on each view. The experts are then combined using an Expert-Attention (EA) network, where the attention part of the network tries to learn when to attend to which view of the data. Multiple variants have been proposed based on how EA network is trained. Yet another challenge is the extreme class imbalance present in the data since only a small fraction of clickstreams correspond to buyers. We propose a well informed undersampling strategy using autoencoders. This simple undersampling technique ensured that the model trained was not biased to non-buyers and resulted in much improved f-scores. Experimental results show that using EA networks, there is an improvement of 13% over single view methods. Moreover, it was also noticed that MVL using EA network performs better than conventional MVL methods such as Multiple Kernel Learning.
- David Ben-Shimon, Alexander Tsikinovsky, Michael Friedmann, Bracha Shapira, Lior Rokach, and Johannes Hoerle. 2015. Recsys challenge 2015 and the yoochoose dataset. In Proceedings of the 9th ACM Conference on Recommender Systems. ACM, 357--358. Google ScholarDigital Library
- Donald J Berndt and James Clifford. 1994. Using Dynamic Time Warping to Find Patterns in Time Series.. In KDD workshop, Vol. 10. Seattle, WA, 359--370. Google ScholarDigital Library
- Veronika Bogina, Tsvi Kuflik, and Osnat Mokryn. 2016. Learning Item Temporal Dynamics for Predicting Buying Sessions. In Proceedings of the 21st International Conference on Intelligent User Interfaces. ACM, 251--255. Google ScholarDigital Library
- Randolph E Bucklin and Catarina Sismeiro. 2009. Click here for Internet insight: Advances in clickstream data analysis in marketing. Journal of Interactive Marketing 23, 1 (2009), 35--48.Google ScholarCross Ref
- Chih-Chung Chang and Chih-Jen Lin. 2011. LIBSVM: a library for support vector machines. ACM Transactions on Intelligent Systems and Technology 2 (2011), 27. Google ScholarDigital Library
- Yoon Ho Cho, Jae Kyeong Kim, and Soung Hie Kim. 2002. A personalized recommender system based on web usage mining and decision tree induction. Expert systems with Applications 23, 3 (2002), 329--342.Google Scholar
- Chester Curme, Tobias Preis, H Eugene Stanley, and Helen Susannah Moat. 2014. Quantifying the semantics of search behavior before stock market moves. Proceedings of the National Academy of Sciences 111, 32 (2014), 11600--11605.Google ScholarCross Ref
- Krzysztof Dembczynski, Wojciech Kotlowski, and Dawid Weiss. 2008. Predicting ads click-through rate with decision rules. In Workshop on targeting and ranking in online advertising, Vol. 2008.Google Scholar
- Philippe Fournier-Viger, Antonio Gomariz, Manuel Campos, and Rincy Thomas. 2014. Fast vertical mining of sequential patterns using co-occurrence information. In Advances in Knowledge Discovery and Data Mining. Springer, 40--52.Google Scholar
- Geoffrey E Hinton and Ruslan R Salakhutdinov. 2006. Reducing the dimensionality of data with neural networks. Science 313, 5786 (2006), 504--507.Google Scholar
- Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation 9, 8 (1997), 1735--1780. Google ScholarDigital Library
- Rajan Lukose, Jiye Li, Jing Zhou, and Satyanarayana Raju Penmetsa. 2008. Learning user purchase intent from user-centric data. In PAKDD. Springer, 673--680. Google ScholarDigital Library
- Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781 (2013).Google Scholar
- Wendy W Moe. 2003. Buying, searching, or browsing: Differentiating between online shoppers using in-store navigational clickstream. Journal of consumer psychology 13, 1 (2003), 29--39.Google ScholarCross Ref
- Wendy W Moe and Peter S Fader. 2004. Dynamic conversion behavior at e-commerce sites. Management Science 50, 3 (2004), 326--335. Google ScholarDigital Library
- Alan L Montgomery, Shibo Li, Kannan Srinivasan, and John C Liechty. 2004. Modeling online browsing and path analysis using clickstream data. Marketing Science 23, 4 (2004), 579--595.Google ScholarDigital Library
- Jooyoung Park and Irwin W Sandberg. 1991. Universal approximation using radial-basis-function networks. Neural computation 3, 2 (1991), 246--257.Google Scholar
- Barak A Pearlmutter. 1989. Learning state space trajectories in recurrent neural networks. Neural Computation 1, 2 (1989), 263--269. Google ScholarDigital Library
- Manoj Kumar Priya, Siddhartha Ghosh. 2015. Visualizing Website Clickstream Data. (2015). http://www.ijraset.com/fileserve.php?FID=2593Google Scholar
- Eric Sven Ristad and Peter N Yianilos. 1998. Learning string-edit distance. IEEE Transactions on Pattern Analysis and Machine Intelligence 20, 5 (1998), 522--532. Google ScholarDigital Library
- Peter Romov and Evgeny Sokolov. 2015. RecSys Challenge 2015: ensemble learning with categorical features. In Proceedings of the 2015 International ACM Recommender Systems Challenge. ACM, 1. Google ScholarDigital Library
- Dirk Van den Poel and Wouter Buckinx. 2005. Predicting online-purchasing behaviour. European Journal of Operational Research 166, 2 (2005), 557--575.Google ScholarCross Ref
- Zhenzhou Wu, Bao Hong Tan, Rubing Duan, Yong Liu, and Rick Siow Mong Goh. 2015. Neural Modeling of Buying Behaviour for E-Commerce from Clicking Patterns. In Proceedings of the 2015 International ACM Recommender Systems Challenge. ACM, 12. Google ScholarDigital Library
- Peng Yan, Xiaocong Zhou, and Yitao Duan. 2015. E-Commerce Item Recommendation Based on Field-aware Factorization Machine. In Proceedings of the 2015 International ACM Recommender Systems Challenge. ACM, 2. Google ScholarDigital Library
- Mingyue Zhang, Guoqing Chen, and Qiang Wei. 2016. Discovering ConsumersâĂŹ Purchase Intentions Based on Mobile Search Behaviors. In Flexible Query Answering Systems 2015. Springer, 15--28.Google Scholar
Index Terms
- A neural attention based approach for clickstream mining
Recommendations
ClickGraph: Web Page Embedding using Clickstream Data for Multitask Learning
WWW '19: Companion Proceedings of The 2019 World Wide Web ConferenceThe rise of big data frameworks has given website administrators the ability to track user clickstream data with more detail than ever before. These clickstreams can represent the user’s intent and purpose in visiting the site. While existing work has ...
Combining Oversampling with Recurrent Neural Networks for Intrusion Detection
Database Systems for Advanced Applications. DASFAA 2021 International WorkshopsAbstractPrevious studies on intrusion detection focus on analyzing features from existing datasets. With various types of fast-changing attacks, we need to adapt to new features for effective protection. Since the real network traffic is very imbalanced, ...
Attention Based Recurrent Neural Networks for Online Advertising
WWW '16 Companion: Proceedings of the 25th International Conference Companion on World Wide WebWe investigate the use of recurrent neural networks (RNNs) in the context of online advertising, where we use RNNs to map both query and ads to real valued vectors. In addition, we propose an attention network that assigns scores to different word ...
Comments