Multi-label learning with millions of labels: recommending advertiser bid phrases for web pages

By Rahul Agrawal, Archit Gupta, Yashoteja Prabhu, Manik Varma. 2013

306 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information 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 Information Systems

Citations in the corpus

Large-Scale Hierarchical Text Classification with Recursively Regularized Deep Graph-CNN (2018, background): “Topical text classification is a fundamental text mining problem for many applications, such as news classification [18], question answering [28], search result organization [11], online advertising [2], etc.”
Structured Prediction Energy Networks (2015, methodology): “When y i = 0 , they are treated as missing data, whose values can be imputed using assumptions about the rank (Lin et al., 2014) or sparsity (Bucak et al., 2011; Agrawal et al., 2013) of the matrix of training labels.”
FastXML: a fast, accurate and stable tree-classifier for extreme multi-label learning (2014, methodology): “Unfortunately, such a hierarchy is unavailable in many applications [2, 35].”
Sparse Local Embeddings for Extreme Multi-label Classification (2015, methodology): “Tree-based approaches: Recently, tree based methods (Agrawal et al., 2013; Prabhu & Varma, 2014; Weston et al., 2013) have also become popular for XML as they enjoy significant accuracy gains over the existing embedding methods.”
Deep Learning for Extreme Multi-label Text Classification (2017, methodology): “Another category of efforts that has improved the state of the art of XMTC in recent years are the new variants of tree-based ensemble methods [1, 36, 41].”
Provable Inductive Matrix Completion (2013, background): “Finally, we provide empirical validation of our approach and demonstrate that alternating minimization is able to recover the true matrix for the above mentioned problems using a small number of measurements.”
Extreme Multi-label Loss Functions for Recommendation, Tagging, Ranking & Other Missing Label Applications (2016, background): “As such, the Hamming loss [4, 6, 9, 10, 13, 18, 21, 39, 42, 47] continues to be one of the most popular losses for extreme multi-label learning along with precision [3, 5, 17, 20, 21, 33, 44, 45] and the F-measure [4, 10, 14, 18, 19, 21, 39, 49].”
Comparison of Naive Bayes, Random Forest, Decision Tree, Support Vector Machines, and Logistic Regression Classifiers for Text Reviews Classification (2017, methodology): “A workflow model for review processing was established to compare Naïve Bayes, Random Forest, Decision Tree, Support Vector Machines, and Logistic Regression classifiers.”
Sampled Softmax with Random Fourier Features (2019, methodology): “Following the convention of extreme classification literature [34, 35, 4], we report precision at k (PREC@K) on the test set.”
Scaling Graph-based Semi Supervised Learning to Large Number of Labels Using Count-Min Sketch (2013, methodology): “A graph-based SSL method aimed at retaining only top ranking labels per-node out of a large label set is presented in [1].”
Open-vocabulary Object Retrieval (2014, background): “open-vocabulary problem that arises when natural language strings are the label space requires recognizing potentially millions of separate categories [4].”
Multi-label Adversarial Perturbations (2018, background): “Similar linearization of constraints have been proved to be effective in the multi-class attacking [12].”
The Emerging Trends of Multi-Label Learning (2020, methodology): “For example, [194] develops an MLC model to automatically recommend bid phrases to an advertiser from a given ad landing page; [195] approaches the item-to-item recommendation task on Amazon, which aims”
Large-scale Multi-label Learning with Missing Labels (2013, methodology): “, Yij = 1 or 0), or missing (Yij =?); several other works have considered another setting where only positive labels are known and are given as 1 in the label matrix, while negative or missing values are all denoted by 0 (Agrawal et al., 2013; Bucak et al., 2011).”
Large Scale Distributed Semi-Supervised Learning Using Streaming Approximation (2015, unknown): “ly storing and updating label distributions during semi-supervised learning with a streaming approximation. In addition, we also compare (in Section 5) against other relevant graphbased SSL baselines [30, 1] that use heuristics to discard poorly scored labels and retain only top ranking labels per n”