A Guide for Learning from YouTube
Whether it’s coding, cooking or calculus, more people are using YouTube and other video sharing websites to learn new things. Computer science Ph.D. student Jingxian Liao, part of associate professor Hao-Chuan Wang’s group, is trying to make this experience better and easier by creating a structured learning experience from a list of video search results.
Especially since the pandemic started, learning from online videos has become a popular alternative to formal education. It’s free, accessible and can provide a wealth of knowledge on an extremely wide range of topics. However, video search results tend to be a disorganized list that doesn’t indicate which videos are quality or good for beginners, which means people need to spend extra time combing through results to find what they’re looking for. Beginners may also not know how to self-organize these results, which can make learning harder.
Liao and her collaborators created ConceptGuide, a map-based organizational tool, to address this challenge. The program automatically analyzes the transcript of videos in each search domain, extracts key words and cross-references them with text materials and academic literature to develop a recommended “web of learning.” The web organizes videos by increasingly advanced concepts, recommends an order to watch them in and shows how the different key words and videos are related to each other.
“YouTube can be very noisy, so we’re trying to have a general, scalable system to help,” said Liao. “We proved that this kind of automatic system works and may encourage people to learn more because they’re focusing more on the content instead of searching and filtering to get what they need.”
The project—led by Liao, Wang and professor Wen-Chieh Lin at National Yang Ming Chiao Tung University in Taiwan—builds on their previous successes. By using crowdsourcing to identify and timestamp important concepts in videos, the team found that they could help people learn better from single educational videos. ConceptGuide expands on this idea by creating an automatic program that does this for multiple videos to create a concept map for learners.
Liao says the most difficult part is extracting a meaningful structure from video transcripts, as some transcripts are auto-generated and some are about emerging topics, which makes them less reliable. It also requires a lot of research into different technical topics as well as educational strategies, and figuring out how to impart that knowledge to the system so it makes good recommendations. The team was able to overcome these challenges and developed a proof-of-concept system.
Her paper outlining the concept, titled “ConceptGuide: Supporting Online Video Learning with Concept Map-based Recommendation of Learning Path,” received the best student paper award at The Web Conference 2021 (WWW ’21) this spring, beating out 354 other submissions. Liao is happily surprised because the paper took four submissions to get accepted.
“It took a long time to get here, to get accepted and to get this award, so the whole team is quite excited and I really appreciate everyone’s hard work,” she said.
In the future, her team plans to continue to improve the mapping system while optimizing the algorithm to work for as many topics as possible without sacrificing accuracy. As the program improves, so does the likelihood that it can be deployed to help online learners.
“I hope that it will help people and give them a better experience to learn from YouTube,” she said.