This is an Unbiased Learning To Rank Algorithms (ULTRA) toolbox, which is still UNDER DEVELOPMENT. A user-friendly documentation can be found here.

Please cite the following paper when you use this toolbox:

Ai, Qingyao, Jiaxin Mao, Yiqun Liu, and W. Bruce Croft. "Unbiased learning to rank: Theory and practice." In Proceedings of the 27th ACM International Conference on Information and Knowledge Management, pp. 2305-2306. ACM, 2018.

Create virtual environment (optional):

pip install --user virtualenv
~/.local/bin/virtualenv -p python3 ./venv
source venv/bin/activate

Install ULTRA from the source:

git clone https://github.com/ULTR-Community/ULTRA.git
cd ULTRA
python setup.py install #use setup-gpu.py for GPU support

Run toy example:

cd example/toy
bash offline_exp_pipeline.sh

1. ClickSimulationFeed: this is the inpuyt layer that generate synthetic clicks on fixed ranked lists to feed the learning algorithm.

2. DeterministicOnlineSimulationFeed: this is the inpuyt layer that first create ranked lists by sorting documents according to the current ranking model, and then generate synthetic clicks on the lists to feed the learning algorithm.

3. StochasticOnlineSimulationFeed: this is the inpuyt layer that first create ranked lists by sampling documents based on their scores in the current ranking model and the Plackett-Luce distribution, and then generate synthetic clicks on the lists to feed the learning algorithm.

4. DirectLabelFeed: this is the inpuyt layer that directly feed the true relevance labels of each documents to the learning algorithm.

1. DLA: this is an implementation of the Dual Learning Algorithm in Unbiased Learning to Rank with Unbiased Propensity Estimation.

2. IPW: this model is an implementation of the Inverse Propensity Weighting algorithms in Learning to Rank with Selection Bias in Personal Search and Unbiased Learning-to-Rank with Biased Feedback

3. REM: this model is an implementation of the regression-based EM algorithm in Position bias estimation for unbiased learning to rank in personal search

4. PD: this model is an implementation of the pairwise debiasing algorithm in Unbiased LambdaMART: An Unbiased Pairwise Learning-to-Rank Algorithm.

5. PDGD: this model is an implementation of the Pairwise Differentiable Gradient Descent algorithm in Differentiable unbiased online learning to rank

6. DBGD: this model is an implementation of the Dual Bandit Gradient Descent algorithm in Interactively optimizing information retrieval systems as a dueling bandits problem

7. NA: this model is an implementation of the naive algorithm that directly train models with input labels (e.g., clicks).

1. Linear: this is a linear ranking algorithm that compute ranking scores with a linear function.

2. DNN: this is neural ranking algorithm that compute ranking scores with a multi-layer perceptron network (with non-linear activation functions).

3. DLCM: this is an implementation of the Deep Listwise Context Model in Learning a Deep Listwise Context Model for Ranking Refinement. (TODO)

4. GFS: this is an implementation of the Groupwise Scoring Function in Learning Groupwise Multivariate Scoring Functions Using Deep Neural Networks. (TODO)

1. MRR: the Mean Reciprocal Rank (inherited from TF-Ranking).

2. ERR: the Expected Reciprocal Rank from Expected reciprocal rank for graded relevance.

3. ARP: the Average Relevance Position (inherited from TF-Ranking).

4. NDCG: the Normalized Discounted Cumulative Gain (inherited from TF-Ranking).

5. DCG: the Discounted Cumulative Gain (inherited from TF-Ranking).

6. Precision: the Precision (inherited from TF-Ranking).

7. MAP: the Mean Average Precision (inherited from TF-Ranking).

8. Ordered_Pair_Accuracy: the percentage of correctedly ordered pair (inherited from TF-Ranking).

Create click models for click simulations

python utils/click_models.py pbm 0.1 1 4 1.0 example/ClickModel

* The output is a json file containing the click mode that could be used for click simulation. More details could be found in the code.

(Optional) Estimate examination propensity with result randomization

python utils/propensity_estimator.py example/ClickModel/pbm_0.1_1.0_4_1.0.json <DATA_DIR> example/PropensityEstimator/

* The output is a json file containing the estimated examination propensity (used for IPW). DATA_DIR is the directory for the prepared data created by utils/libsvm/prepare_exp_data_with_svmrank.py. More details could be found in the code.

If you use ULTRA in your research, please use the following BibTex entry.

@inproceedings{Ai:2018:ULR:3269206.3274274,
 author = {Ai, Qingyao and Mao, Jiaxin and Liu, Yiqun and Croft, W. Bruce},
 title = {Unbiased Learning to Rank: Theory and Practice},
 booktitle = {Proceedings of the 27th ACM International Conference on Information and Knowledge Management},
 series = {CIKM '18},
 year = {2018},
 isbn = {978-1-4503-6014-2},
 location = {Torino, Italy},
 pages = {2305--2306},
 numpages = {2},
 url = {http://doi.acm.org/10.1145/3269206.3274274},
 doi = {10.1145/3269206.3274274},
 acmid = {3274274},
 publisher = {ACM},
 address = {New York, NY, USA},
 keywords = {click model, counterfactual learning, unbiased learning to rank, user bias},
}