Non-Linear Label Ranking for Large-Scale Prediction of Long-Term User Interests

Authors: Nemanja Djuric, Mihajlo Grbovic, Vladan Radosavljevic, Narayan Bhamidipati, Slobodan Vucetic

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on a real-world advertising data set with more than 3.2 million users show that the proposed algorithm outperforms the existing solutions in terms of both rank loss and top-K retrieval performance, strongly suggesting the benefit of using the proposed model on large-scale ranking problems.
Researcher Affiliation Collaboration Yahoo! Labs, Sunnyvale, CA, USA, {nemanja, mihajlo, vladan, narayanb}@yahoo-inc.com Temple University, Philadelphia, PA, USA, vucetic@temple.edu
Pseudocode No The paper describes algorithms using mathematical formulas and textual explanations, but it does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code No The paper mentions using and modifying existing tools like Vowpal Wabbit and Budgeted SVM (Djuric et al. 2014) but does not provide a link or explicit statement about the availability of their specific modified code or the implementation of AMM-rank.
Open Datasets No The data set that was used in the empirical evaluation was generated using the information about users online activities collected at Yahoo servers.
Dataset Splits Yes Performance of the competing methods in terms of ϵdis, following 5-fold cross-validation, is reported in Table 1.
Hardware Specification No We note that, other than IB-Mal, the methods are very efficient, obtaining training and test times of less than 10 minutes on a regular machine.
Software Dependencies No We used Vowpal Wabbit package1 for logistic regression, Budgeted SVM (Djuric et al. 2014) for AMM, that we also modified to implement AMM-rank. ...used the default parameters from Budgeted SVM package for AMM-rank. The paper mentions software names but does not provide specific version numbers for these software components.
Experiment Setup Yes We set ν(i) = 1, i = 1, . . . , L, and used the default parameters from Budgeted SVM package for AMM-rank, with the exception of the λ parameter which, together with competitors parameters, was configured through cross-validation on a small held-out set; this resulted in k = 10 for IB-Mal.