Literature and methods Amir Tavanaei Dr. Vijay Raghavan RANKING PROJECT Literature and methods Amir Tavanaei Dr. Vijay Raghavan
Data Multi-attribute samples Goal: How? Numerical and ordinal Ranking the samples using Accumulating, Partial ordering, or Learning methods How?
Accumulating Method Each sample is represented by D values: The simplest method (just to mention) More advanced: The problem is to find the weights!
Attribute weights Hand-crafted weight values (heuristic) Correlation Clustering PCA-Varimax (It needs many pre- and post processing tasks) Partial Ordering (next slide) Learning and Regression (using comparable samples)
Partial Ordering can Find the Attribute Impact Partial order investigation for sensitivity analysis* Method: Original Hass diagram Removing attributes one at a time Finding new Hass diagram Difference between original Hass diagram and the new one describes the attribute’s importance *: Paola Annoni, R. Bruggemann, A. Saltelli, 2011 (I have a copy of their book in my documents)
Learning To Rank Learning refers to weight adjustment to show the variable impacts. Supervised Classification Ordinal categories are used Unsupervised There is no class label
Unsupervised Learning Unsupervised Learning to rank is the project goal. What we have: Dataset containing multi-criteria samples What we need: Ranking the samples using unsupervised learning Probable methods: Using partially ranked samples Clustering as pre-processing
Project Requirements An introduction describing unsupervised learning to rank methods. Following a method and implementing it. Finding a way (by going through literatures. New ideas are also welcome) to find the ranking performance. Evaluating the model. Report
Contact Me! What I can give you: I am available: Papers explaining learning to rank (Supervised and Unsupervised). Dataset I am available: Monday, Thursday, Friday at CVDI Lab, Abdollah Hall Tuesday, Wednesday, Saturday at CACS, #356 tavanaei@louisiana.edu