Several directions exist for future research. The rich information recorded in advertising platforms could be harnessed to investigate more sophisticated lookalike and CTR models. Meanwhile It is also challenging and meaningful to automatically incorporate the capability of invalid traffic identification into deep learning models. For those users having not enough behavioural data, there can be a possibility for us to exploit short-term sequence features in a time-series perspective.
Another direction is to extend existing works by introducing temporal considerations in mining and managing the user behaviour data. Both user- generated data on advertising and social media platforms have time-stamp attribute. For example, we can introduce recency weight in model learning and query processing for various applications. Furthermore, how to build
References
scalable learning model and index construction in real-time system setting could also be an interesting problem.
References
[1] F. Abel, Q. Gao, G.-J. Houben, and K. Tao, “Analyzing user modeling on twitter for personalized news recommendations,” in UMAP, 2011, pp. 1–12. [Online]. Available: http://dl.acm.org/citation.cfm?id=2021855.2021857
[2] P. P. K. Chan, X. Hu, L. Zhao, D. S. Yeung, D. Liu, and L. Xiao, “Convolutional neural networks based click-through rate prediction with multiple feature se- quences,” in Proc. of IJCAI, 2018, pp. 2007–2013.
[3] H.-T. Cheng, L. Koc, J. Harmsen, T. Shaked, T. Chandra, H. Aradhye, G. Ander- son, G. Corrado, W. Chai, M. Ispir, R. Anil, Z. Haque, L. Hong, V. Jain, X. Liu, and H. Shah, “Wide & deep learning for recommender systems,” in Proc. of the 1st Workshop on Deep Learning for Recommender Systems, 2016, pp. 7–10.
[4] E. Cho, S. A. Myers, and J. Leskovec, “Friendship and mobility: user movement in location-based social networks,” in KDD, 2011, pp. 1082–1090.
[5] L. Derczynski, B. Yang, and C. S. Jensen, “Towards context-aware search and analysis on social media data,” in EDBT, 2013, pp. 137–142.
[6] Y. Doytsher, B. Galon, and Y. Kanza, “Querying geo-social data by bridging spatial networks and social networks.” in GIS-LBSN, 2010, pp. 39–46. [Online]. Available: http://dblp.uni-trier.de/db/conf/gis/lbsn2010.html#DoytsherGK10 [7] ——, “Querying socio-spatial networks on the world-wide web,” in WWW (Com-
panion Volume), 2012, pp. 329–332.
[8] I. D. Felipe, V. Hristidis, and N. Rishe, “Keyword search on spatial databases,” in ICDE, 2008, pp. 656–665.
[9] H. Guo, R. Tang, Y. Ye, Z. Li, and X. He, “Deepfm: A factorization-machine based neural network for ctr prediction,” in Proc. of IJCAI, 2017, pp. 1725–1731. [10] D. Horowitz and S. D. Kamvar, “The anatomy of a large-scale social search en-
gine,” in WWW, 2010, pp. 431–440.
[11] Q. Huang and Y. Liu, “On geo-social network services,” in Geoinformatics, 2009, pp. 1–6.
[12] J. Jiang, X. Lin, J. Yao, R. Ding, H. Lu, and X. Wu, “Deep monitor network for privacy-preserving click-through rate prediction (submitted),” in The World Wide Web Conference, WWW 2020, Taipei, Apr 20-24, 2020, 2020.
[13] J. Jiang, X. Lin, J. Yao, and H. Lu, “Comprehensive audience expansion based on end-to-end neural prediction,” in Proceedings of the SIGIR 2019 Workshop on eCommerce, co-located with the 42st International ACM SIGIR Conference on Research and Development in Information Retrieval, eCom@SIGIR 2019, Paris, France, July 25, 2019., 2019.
References
[14] J. Jiang, H. Lu, P. Li, G. Pan, and X. Xie, “Finding influential local users with similar interest from geo-tagged social media data,” in 18th IEEE International Conference on Mobile Data Management, MDM 2017, Daejeon, South Korea, May 29 - June 1, 2017, 2017, pp. 82–91.
[15] J. Jiang, H. Lu, B. Yang, and B. Cui, “Finding top-k local users in geo-tagged social media data,” in 31st IEEE International Conference on Data Engineering, ICDE 2015, Seoul, South Korea, April 13-17, 2015, 2015, pp. 267–278.
[16] Y. Juan, D. Lefortier, and O. Chapelle, “Field-aware factorization machines in a real-world online advertising system,” in Proc. of WWW Companion, 2017, pp. 680–688.
[17] Y. Juan, Y. Zhuang, W.-S. Chin, and C.-J. Lin, “Field-aware factorization machines for ctr prediction,” in Proc. of RecSys, 2016, pp. 43–50.
[18] A. Kaboutari, S. Branch, J. Bagherzadeh, I. Urmia, and F. Kheradmand, “An evaluation of two-step techniques for positive-unlabeled learning in text classifi- cation,” vol. 3, 2014, pp. 592–594.
[19] C.-H. Lee, H.-C. Yang, T.-F. Chien, and W.-S. Wen, “A novel approach for event detection by mining spatio-temporal information on microblogs,” in ASONAM, 2011, pp. 254–259.
[20] B. Liu, Y. Dai, X. Li, W. S. Lee, and P. S. Yu, “Building text classifiers using positive and unlabeled examples,” in Proc. of ICDM, 2003, pp. 179–186.
[21] B. Liu, R. Tang, Y. Chen, J. Yu, H. Guo, and Y. Zhang, “Feature generation by convolutional neural network for click-through rate prediction,” in The World Wide Web Conference, 2019, pp. 1119–1129.
[22] B. Liu, W. S. Lee, P. S. Yu, and X. Li, “Partially supervised classification of text documents,” in Proc. of ICML, 2002, pp. 387–394.
[23] H. Liu, D. Pardoe, K. Liu, M. Thakur, F. Cao, and C. Li, “Audience expansion for online social network advertising,” in Proc. of KDD, 2016, pp. 165–174.
[24] Q. Liu, F. Yu, S. Wu, and L. Wang, “A convolutional click prediction model,” in Proc. of CIKM, 2015, pp. 1743–1746.
[25] A. Mangalampalli, A. Ratnaparkhi, A. O. Hatch, A. Bagherjeiran, R. Parekh, and V. Pudi, “A feature-pair-based associative classification approach to look-alike modeling for conversion-oriented user-targeting in tail campaigns,” in Proc. of WWW, 2011, pp. 85–86.
[26] M. Michelson and S. A. Macskassy, “Discovering users’ topics of interest on twit- ter: a first look,” in AND, 2010, pp. 73–80.
[27] F. Mordelet and J. P. Vert, “A bagging svm to learn from positive and unlabeled examples,” Pattern Recogn. Lett., vol. 37, pp. 201–209, Feb. 2014.
[28] W. Ouyang, X. Zhang, L. Li, H. Zou, X. Xing, Z. Liu, and Y. Du, “Deep spatio- temporal neural networks for click-through rate prediction,” in Proc. of KDD, 2019, pp. 2078–2086.
[29] J. Pan, J. Xu, A. L. Ruiz, W. Zhao, S. Pan, Y. Sun, and Q. Lu, “Field-weighted factorization machines for click-through rate prediction in display advertising,” in Proc. of WWW, 2018.
References
[30] S. Pandey, M. Aly, A. Bagherjeiran, A. Hatch, P. Ciccolo, A. Ratnaparkhi, and M. Zinkevich, “Learning to target: What works for behavioral targeting,” in Proc. of CIKM, 2011, pp. 1805–1814.
[31] Y. Qu, B. Fang, W. Zhang, R. Tang, M. Niu, H. Guo, Y. Yu, and X. He, “Product- based neural networks for user response prediction over multi-field categorical data,” ACM Trans. Inf. Syst., vol. 37, no. 1, pp. 5:1–5:35, Oct. 2018.
[32] J. Shen, S. C. Geyik, and A. Dasdan, “Effective audience extension in online advertising,” in Proc. of KDD, 2015, pp. 2099–2108.
[33] J. Yan, N. Liu, G. Wang, W. Zhang, Y. Jiang, and Z. Chen, “How much can behavioral targeting help online advertising?” in Proc. of WWW, 2009, pp. 261– 270.
[34] W. Zhang, T. Du, and J. Wang, “Deep learning over multi-field categorical data - - A case study on user response prediction,” in Proc. of ECIR, 2016, pp. 45–57. [35] V. W. Zheng, Y. Zheng, X. Xie, and Q. Yang, “Collaborative location and activity
recommendations with GPS history data,” in WWW, 2010, pp. 1029–1038. [36] Y. Zheng, “Location-based social networks: Users,” in Computing with Spatial
Part II
Paper A
Comprehensive Audience Expansion based on
End-to-End Neural Prediction
Jinling Jiang, Xiaoming Lin, Junjie Yao, Hua Lu
The paper has been published in the
Abstract
In current online advertising applications, lookalike methods are valuable and com- monly used to identify new potential users, tackling the difficulties of audience ex- pansion. However, the demographic information and a variety of user behavior logs are high dimensional,noisy, and increasingly complex, which are challenging to ex- tract suitable user profiles. Usually, rule-based and similarity-based approaches are proposed to profile the users’ interests and expand the audience. However, they are specific and limited in more complex scenarios.
In this paper, we propose a new end-to-end solution, unifying the feature ex- traction and profile prediction stages. Specifically, we present a neural prediction framework and leverage it with the intuitive audience feature extraction stages. We conduct extensive study on a real and large advertisement dataset. The results demonstrate the advantage of the proposed approach, not only in accuracy but also generality.
c
2019 ACM, with permission, from Jinling Jiang, Xiaoming Lin, Junjie Yao, Hua Lu: "Comprehensive Audience Expansion based on End-to-End Neural Prediction," eCOM@SIGIR 2019, 8 pages.
A.1. Introduction
A.1
Introduction
The stunning growth of online advertisement enables the advertisers to sync up their products according to the fast changing needs of consumer. As the development of e-commerce platforms has introduced SMEs to enter con- sumers’ sight, large enterprise advertisers face the crisis of slowing busi- ness growth and falling revenue. Therefore, brand advertisers have begun to pay more attention to the contribution of advertising to sales conversion, the actual revenue brought by advertising, requiring advertising agencies and third-party suppliers to provide more refined performance data of advertis- ing effects.
Meanwhile, the emergence of big data technology has subverted the op- eration model of entire advertising industry and the traditional way of eval- uating advertising effects. By tracking and obtaining user behaviour data, third-party supplier of advertising monitor can analyse the data according to the advertiser needs, not only understanding the communication effects and sales conversion rate generated by the advertisement in time, but also pre- dicting the user conversion probability to some extent. Through analysis and modeling on massive data of user behaviour, advertisers can accurately reach the target consumer. Therefore, how to better utilize the advertising monitor data in order to optimize ad serving and improve marketing conversion rate has become an important issue.
Fig. A.1:Audience Expansion Dataflow
One of the main challenges in ad serving is how to find the best con- verting prospects. A typical way is to do audience expansion, that is, to identify and reach new audiences with similar interests to the original target audience. Usually, the methodology used in audience expansion problem is called lookalike modeling. Given a seed user set S from a universal set U, lookalike models essentially find groups of audiences fromkU−Skwho look and act like the audience in S.
Paper A.
But both traditional and current lookalike strategies for advertiser to look for target audience are mainly based on user demographics. There are two main problems with demographics-based audience segmentation: user de- mographics (age, gender and geographical location) itself is not precise as it is estimated via various statistical methods or machine learning models based on a small group of surveyed samples (10-100 thousands); the num- ber of users that are specified by demographics is large, more sophisticated screening is required. Accordingly, the details of user behaviour data should be harnessed in machine learning models so as to target accurate audience segment. At the same time, there are two main problems that need to be solved based on user behaviour data modeling: user generated behaviour data through Internet is generally high-dimensional and sparse; advertisers usually can only provide positive samples, while negative samples need to be carefully picked up from a huge unlabeled sample set.
In addition, the ecologically closed Internet tycoons (represented by Face- book, Amazon, Tencent, Alibaba and etc.) provide for advertisers the capa- bility to perform audience expansion within their own platforms. However, ad serving data of these platforms are not connected with the advertiser’s CRM (Customer Relationship Management) system, so it is difficult to di- rectly track the real conversion rate. In order to verify that lookalike mod- els based on the user behaviour work better than traditional demographics- based approach regarding the sales conversation rate, we need to integrate data flow during the whole advertising life cycle.
The data flow of audience expansion service is illustrated in Figure A.1. The data runs between advertisers and our universal advertising monitor system across different media platforms. The original users come from the advertiser’s CRM System selecting the consumers who recently exercise the purchase actions. Then the users who are tracked by the universal adver- tising monitor will be matched and treated as "seed" users. Based on the "seed" users and universal user set from advertising monitor, we build looka- like model to predict the probability to be target audience for all users. Af- terwards, according to advertising budget, lookalike model will yield cor- responding number of expanded users to be reached through ad serving system. Finally, the ad serving performance is evaluated by advertiser’s site monitor system that record sales conversion in the near future.
In this paper, we build up a closed-loop data solution for brand adver- tisers and combines multiple techniques of selecting negative samples and extracting features, as well as machine learning lookalike models to reach targeted audience. which greatly enhances the conversion effect of ad serv- ing.
The rest of the paper is organized as follows. Section A.2 gives out the formal problem statement and specifies the notations used in the paper. In Section A.3, we review the related work on various kinds of lookalike models