Developing an intelligent algorithm for automatically labeling digital videos - BC-413
Preferred Disciplines: Electrical Engineering, Mathematics, Computer Science, Information Management or Statistics (Masters, PhD or Post-Doc)
Project length: 4-12 months (1-2 units)
Approx. start date: January 2019
Location: Vancouver, BC
No. of Positions: 1
Preferences: Prefer Vancouver-based candidates
BroadbandTV Corp (BBTV), a digital entertainment company which exists to empower creators and inspire audiences. Today, BBTV generates over 33 billion monthly impressions and is the 3rd largest video property in the world in terms of unique viewers, following only Google and Facebook
Summary of Project:
Having an automated method for labeling videos is of great importance for “advertising” purposes. The labels can be used for tagging videos and then later used for matching the videos against ads. The solution should be flexible enough to address the needs of various advertisers.
In this research project, we want to develop a scalable cost-efficient method based on digital signal processing, machine learning and deep learning to automatically label the content of the video using metadata or selected images available from the video.
The main idea of the research includes:
- Generate highly informative features of a YouTube video. This includes but not limited to text feature (title, description, tag, transcript), visual feature (thumbnail and selected frames) and audio feature.
- Train a model that leverage machine learning or deep learning techniques to predict the most likely labels of the video content based on some guidelines provided by the user (the guidelines might differ from one use case to another). Each video could be labelled by few topics.
- Aggregate the video level results to get the main topic for the entire YouTube channel.
- To be determined upon discussion with the professor and intern
Expertise and Skills Needed:
- Working knowledge of machine learning algorithm such as classification, and prediction
- Know how to build deep learning models like RNN, CNN, DNN and other neural network models
- Know how to design an experiment for testing the performance of a machine learning algorithm
- Familiarity with Python
- Familiarity with one of the data science platform such as Tensorflow, Theano or Pytorch
- Good knowledge of numpy, pandas, sklearn, scipy, seaborn.
- Ability to multi-task
- Ability to work in a fast-paced environment
- A pragmatic and solution-oriented mindset is absolutely necessary
- Good attention to details
- Critical Thinking
- One of the following data analysis tools: Excel, SAS, R, etc
For more info or to apply to this applied research position, please