• Publications
  • Influence
Modeling the impact of short- and long-term behavior on search personalization
TLDR
This first study to assess how short-term (session) behavior and long- term (historic) behavior interact, and how each may be used in isolation or in combination to optimally contribute to gains in relevance through search personalization finds historic behavior provides substantial benefits at the start of a search session. Expand
Guidelines for Human-AI Interaction
TLDR
This work proposes 18 generally applicable design guidelines for human-AI interaction that can serve as a resource to practitioners working on the design of applications and features that harness AI technologies, and to researchers interested in the further development of human- AI interaction design principles. Expand
Pairwise ranking aggregation in a crowdsourced setting
TLDR
This work proposes a new model to predict a gold-standard ranking that hinges on combining pairwise comparisons via crowdsourcing and formalizes this as an active learning strategy that incorporates an exploration-exploitation tradeoff and implements it using an efficient online Bayesian updating scheme. Expand
Here or there: preference judgments for relevance
TLDR
This work hypothesizes that preference judgments of the form "document A is more relevant than document B" are easier for assessors to make than absolute judgments, and investigates methods to evaluate search engines using preference judgments. Expand
Refined experts: improving classification in large taxonomies
TLDR
Methods that target the latter two problems of increasing sparsity of training data at deeper nodes in the taxonomy and error propagation by propagating up "first-guess" expert information in a bottom-up manner before making a refined top down choice are introduced. Expand
Dual Strategy Active Learning
TLDR
A dynamic approach, called DUAL, where the strategy selection parameters are adaptively updated based on estimated future residual error reduction after each actively sampled point, to outperform static strategies over a large operating range. Expand
Personalizing web search results by reading level
TLDR
It is shown how reading level can provide a valuable new relevance signal for both general and personalized Web search, and models and algorithms are described to address the three key problems in improving relevance for search using reading difficulty. Expand
Learning to Rank Using an Ensemble of Lambda-Gradient Models
TLDR
The system that won Track 1 of the Yahoo! Learning to Rank Challenge was described, which used a linear combination of twelve ranking models, eight of which wereagged LambdaMART boosted tree models, two ofWhich were LambdaRank neural nets, and two of Which were MART models using a logistic regression cost. Expand
Transformer-XH: Multi-Evidence Reasoning with eXtra Hop Attention
TLDR
Transformer-XH is presented, which uses eXtra Hop attention to enable intrinsic modeling of structured texts in a fully data-driven way and leads to a simpler multi-hop QA system which outperforms previous state-of-the-art on the HotpotQA FullWiki setting. Expand
Predicting short-term interests using activity-based search context
TLDR
This study developed and evaluated user interest models for the current query, its context (from pre-query session activity), and their combination, which is referred to as intent, and investigates optimally combining the query and its context by learning a model that predicts the context weight for each query. Expand
...
1
2
3
4
5
...