Next Generation LtR Algorithms

• Unsupervised approaches to LtR, active learning for LtR, transfer learning for LtR;

• Incremental LtR, online, or personalized LtR;

• Embedding user behaviour and dynamic in LtR;

• Cost-Aware LtR;

• List-based approaches for result list diversification and/or clustering;

• Bias/Variance and other theoretical characterizations or ranking models;

• Feature engineering for ranking;

• Deep neural networks for ranking;

• Understanding and explaining complex LtR models, also via visual analytics solutions.

Evaluation of LtR Algorithms:

• Quality measures accounting for user behaviour and perceived quality;

• Quality measures accounting for models failures, redundancy, robustness, sensitivity, etc.;

• Evaluation of ranking efficiency vs. quality trade-off;

• Visual analytics solutions for exploring and interpreting experimental data;

• Reproducibility of LtR experiments.

Datasets:

• Measuring quality of training datasets: noise, contradictory examples, redundancy, difficulty of building a good model, features quality, coverage of application domain use cases;

• Creation and curation of datasets: compression, negative sampling, aging, dimensionality reduction;

• Contributing novel datasets to the community.

Applications:

• Application of LtR to verticals or to other domains (e.g., recommendation, news, product search, social media, job search, ...);

• LtR beyond documents: keyword-based access to structured data, multimedia, graphs, etc.