eTREE: Learning Tree-structured Embeddings

Authors: Faisal M. Almutairi, Yunlong Wang, Dong Wang, Emily Zhao, Nicholas D. Sidiropoulos6609-6617

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate the proposed framework on real data from various application domains: healthcare analytics, movie recommendations, and education. This section aims to answer the following questions: Q1. Accuracy: Does e TREE improve the quality of embeddings for the downstream tasks? Q2. Interpretability: How meaningful is the tree structure learned by e TREE from an application domain knowledge viewpoint? Datasets: We evaluate e TREE and the competing baselines on the following real datasets: (i) Med-HF: (ii) Med-MCI: (iii) Movielens: (iv) College Grades: ... Table 1 shows the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of all methods on the different datasets.
Researcher Affiliation Collaboration 1Department of ECE, University of Minnesota, MN, USA 2Advanced Analytics, IQVIA Inc., Plymouth Meeting, PA, USA 3Department of ECE, University of Virginia, VA, USA almut012@umn.edu, {yunlong.wang, dong.wang, emily.zhao}@iqvia.com, nikos@virginia.edu
Pseudocode Yes Algorithm 1: Algorithmic Solution to e TREE
Open Source Code Yes 1Code is at: https://github.com/FaisalAlmutairi/e TREE
Open Datasets Yes Movielens: Movielens (Harper and Konstan 2015) is a movie rating dataset and a popular baseline in recommender systems literature.
Dataset Splits Yes We split each dataset into 5 equal folds. After training the models on 4 folds (80% of the data), we test the trained models on the held-out fold. The hyper-parameters of all methods are chosen via cross validation (10% of training data).
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments, such as GPU/CPU models, memory, or cloud instance types.
Software Dependencies No The paper does not provide specific software dependencies with version numbers needed to replicate the experiment.
Experiment Setup Yes in our experiments we set K = 5. ... we set it to η = 1000 in all experiments. ... R = 9, λ = 1, µ = 50 (more emphasis on the tree term), M2 = 27 (number of subcategories), and M3 = 9 (number of main categories).