Practical Lessons from Predicting Clicks on Ads at Facebook

By Xinran He, Junfeng Pan, Ou Jin, Tianbing Xu, Boyang Liu, Tao Xu, Tao Xu, Yanxin Shi, A. Atallah, Ralf Herbrich and 2 more. 2014

977 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Management Science and Operations Research, Marketing, Recommender Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Management Science and Operations Research

In Marketing

In Recommender Systems

Citations in the corpus

Deep Interest Network for Click-Through Rate Prediction (2017, methodology): “dd fully connected layers (in other words, multilayer perceptrons, MLPs) to automatically learn the nonlinear relations among features. Compared to traditional commonly used logistic regression model [5, 6]. MLPs can reduce a lot of feature engineering jobs, which is time and manpower consuming in i”
SecureBoost: A Lossless Federated Learning Framework (2019, methodology): “As one of the most widely-used machine-learning algorithms, the gradient-tree boosting model (Friedman et al. 2000) excels in many machine learning tasks, such as fraud detection (Oentaryo et al. 2014), feature selection (Li et al. 2017) and product recommendation (He et al. 2014).”
Machine Unlearning (2019, background): “Morever, data pipelines are often not static [3]: new data is collected regularly and incrementally used to further reﬁne existing models following the online learning paradigm [4].”
FiBiNET: combining feature importance and bilinear feature interaction for click-through rate prediction (2019, methodology): “Many models have been proposed in this field such as logistic regression (LR)[17], polynomial-2 (Poly2)[10], tree based models[7], tensor-based models[13], Bayesian models[3], and factorization machines based models[9, 10, 19, 20].”
Practical Evaluation and Optimization of Contextual Bandit Algorithms (2018, background): “Online learning is an important tool for having machine learning systems that quickly and efficiently adapt to observed data (Agarwal et al., 2016; He et al., 2014; McMahan et al., 2013).”
XGBoost : Reliable Large-scale Tree Boosting System (2015, background): “They are also deployed into real world production systems, such as online advertising [7].”
High-performance, Distributed Training of Large-scale Deep Learning Recommendation Models (2021, background): “The model quality is evaluated in normalized entropy [16] and the performance is measured in queries per second (QPS).”
A Meta-Learning Perspective on Cold-Start Recommendations for Items (2017, methodology): “PROD-BEST uses features not only for the current Tweet but historical features as described in [8].”
ActiVis: Visual Exploration of Industry-Scale Deep Neural Network Models (2017, background): “The ranking models can be used to recommend relevant content to users by analyzing a large number of numerical features extracted from databases [5, 15].”
Applied Machine Learning at Facebook: A Datacenter Infrastructure Perspective (2018, background): “In some cases, particularly for recommendation systems, additional training is also performed online in a continuous manner [5].”
A Contextual Bandit Bake-off (2018, positive): “Online learning in VW. Online learning is an important tool for having machine learning systems that quickly and eﬃciently adapt to observed data (Agarwal et al., 2016; He et al., 2014; McMahan et al., 2013).”
Update or wait: How to keep your data fresh (2016, background): “resh data is critical in online learning which is of great interest to practitioners due to the recent emergence of real-time applications such as advertisement placement and online web ranking [36], [37]. One can employ the age penalty functions that grow quickly with respect to the age, such as th”
TEM: Tree-enhanced Embedding Model for Explainable Recommendation (2018, methodology): “The prior work by Facebook [22] has demonstrated the effectiveness of such a solution, which feeds the leaf nodes of a GBDT into a logistic regression (LR) model.”
Display Advertising with Real-Time Bidding (RTB) and Behavioural Targeting (2016, methodology): “To tackle this problem, non-linear models such as factorisation machines (FM) [Menon et al., 2011, Ta, 2015] and tree-based models [He et al., 2014] are studied to explore feature interactions.”
Low-Precision Hardware Architectures Meet Recommendation Model Inference at Scale (2021, background): “Normalized (cross) entropy (NE)[28] is an ofﬂine accuracy metric that is applicable to various models producing prediction probabilities on binary labeled datasets.”