CSE 575 Class Project Presentation Dynamic Tag Recommendation In High-Dimensional Systems Hans Behrens, 1211230537, hwbehren@asu.edu, 25% Yash Garg, 1206277315, ygarg@asu.edu, 25% Prad Kadambi, 1204813671, pradyumna.kadambi@asu.edu, 25% Yanyao Wang, 1001462080, ywang@asu.edu, 25%
Presentation Overview Ourselves Our Goals Our Motivations Our Methods Our Findings
Dynamic Tag Recommendation In High-Dimensional Systems Our Goals (1/4) Dynamic Tag Recommendation In High-Dimensional Systems
Dynamic Tag Recommendation In High-Dimensional Systems Our Goals (2/4) Dynamic Tag Recommendation In High-Dimensional Systems
Dynamic Tag Recommendation In High-Dimensional Systems Our Goals (3/4) Dynamic Tag Recommendation In High-Dimensional Systems
Dynamic Tag Recommendation In High-Dimensional Systems Our Goals (4/4) Dynamic Tag Recommendation In High-Dimensional Systems
Our Motivations (1/3) Practical Performance Relevancy Kernelizable Incrementally Updateable Relevancy Recommender Systems Neural Networks
Our Motivations (2/3) Useful To Users To Developers Improve ability to find relevant posts Get help faster To Developers Evaluate possibilities Highlight successes & failures
Our Motivations (3/3) Challenging Dimensionality Scale 75K unique words 47K unique tags Scale 50GB dataset 1 million posts
Our Methods (1/8) Dataset Preprocessing Data Cleaning Stop Words Numbers Tyops <HTML>
Our Methods (2/8) Characteristic Extraction Weight Matrix
Our Methods (3/8) Modelling Tensor Decomposition https://commons.wikimedia.org/wiki/File:Stress_transformation_3D.svg CC-BY-SA-3.0
Our Methods (4/8) Modelling Neural Network (Convolutional) http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf
Our Methods (5/8) Modelling SVM https://en.wikipedia.org/wiki/Support_vector_machine#/media/File:Svm_max_sep_hyperplane_with_margin.png
Our Methods (6/8) Modelling Neural Network (Perceptron) http://neuroph.sourceforge.net/tutorials/Perceptron.html
Our Methods (7/8) Modelling Neural Network (Multi-Layer Perceptron) Hidden Layers Input Layer Output Layer 𝑥 𝑛 𝑦
Our Methods (8/8) Modelling Ensemble Learning Max Pooling Averaging
Our Findings (1/1) [ Precision & Recall results will be included in the written report ]
Q&A Session
This page intentionally left blank
Team Information All the team members agree on the team members’ contributions, in terms of both (a) what they did and (b) the percentage. Hans Behrens, 1211230537, hwbehren@asu.edu, 25% Project proposal, project report, project presentation & slides Python/RDBMS integration; data cleaning & representation brainstorming Yash Garg, 1206277315, ygarg@asu.edu, 25% Network and ensemble modeling Prad Kadambi, 1204813671, pradyumna.kadambi@asu.edu, 25% Explored GPU acceleration and integration, Django web UI, convolutional neural networks Yanyao Wang, 1001462080, ywang@asu.edu, 25% Explored TF-IDF