Active embedding search via noisy paired comparisons (bibtex)
by , , and
Abstract:
Suppose that we wish to estimate a user's preference vector w from paired comparisons of the form "does user w prefer item p or item q?," where both the user and items are embedded in a low-dimensional Euclidean space with distances that reflect user and item similarities. Such observations arise in numerous settings, including psychometrics and psychology experiments, search tasks, advertising, and recommender systems. In such tasks, queries can be extremely costly and subject to varying levels of response noise; thus, we aim to actively choose pairs that are most informative given the results of previous comparisons. We provide new theoretical insights into the benefits and challenges of greedy information maximization in this setting, and develop two novel strategies that maximize lower bounds on information gain and are simpler to analyze and compute respectively. We use simulated responses from a real-world dataset to validate our strategies through their similar performance to greedy information maximization, and their superior preference estimation over state-of-the-art selection methods as well as random queries.
Reference:
Active embedding search via noisy paired comparisonsG. Canal, A. Massimino, M. Davenport and C. Rozell. In International Conference on Machine Learning (ICML), June 2019. (Acceptance rate 23%)
Bibtex Entry:
@InProceedings{canal.18d,
      author = 	 {Canal, G. and Massimino, A. and Davenport, M. and Rozell, C.},
      title = 	 {Active embedding search via noisy paired comparisons},
     booktitle =	 {International Conference on Machine Learning ({ICML})},
     year =	 2019,
	 address = {Long Beach, CA},
	 month = jun,
	 note = {(Acceptance rate 23\%)},
	 url = {https://arxiv.org/pdf/1905.04363.pdf},
	 abstract = {Suppose that we wish to estimate a user's preference vector w from paired comparisons of the form
"does user w prefer item p or item q?," where both the user and items are embedded in a low-dimensional
Euclidean space with distances that reflect user and item similarities. Such observations arise in numerous
settings, including psychometrics and psychology experiments, search tasks, advertising, and recommender
systems. In such tasks, queries can be extremely costly and subject to varying levels of response noise;
thus, we aim to actively choose pairs that are most informative given the results of previous comparisons.
We provide new theoretical insights into the benefits and challenges of greedy information maximization
in this setting, and develop two novel strategies that maximize lower bounds on information gain and are
simpler to analyze and compute respectively. We use simulated responses from a real-world dataset to
validate our strategies through their similar performance to greedy information maximization, and their
superior preference estimation over state-of-the-art selection methods as well as random queries.}
  }
Powered by bibtexbrowser