Research: Relevant implementations

[JMitnik]

General

Resources

• Kaggle process with nice data preprocessing including Pipelines ❗New

• 32 Sklearn tips on Kaggle

  Learn-to-rank General

  Resources

• Dataquest Kaggle tutorial Slightly different dataset but still relevant

• Lucidworks ABC to Learning to Rank

Our dataset

Resources

• Kaggle: Expedia Personalized Sort Discussion

[Thomasdegier]

links:

• https://www.dataquest.io/blog/kaggle-tutorial/ (slightly different dataset but still relevant)
• https://lucidworks.com/post/abcs-learning-to-rank/
• https://www.kaggle.com/c/expedia-personalized-sort/discussion/6228

Learning-To-Rank algorithms (LTR's) (avoids confidence score predictive features?)

• pointwise (any single regression algorithms)
• pairwise (RankNet, LambdaRank)
• listwise (SoftRank, ListNet, LambdaRank)

Feature Selection:

Single Features for:

• prediction of bool_book (5)
• prediction of bool_click (1)
• neither (0) Highest correlating columns are prop_location_score2, promotionflag, position ( position needs to be predicted as well in testdata).

Other ideas/notions for features: Using median/average scores of numeric features per prop_id instead of looking at specific user_ids. Categorical features apart from prop_id don't increase performance significantly. Training over a subset seems adviseable when training takes long