For my week 8 synthesis I thought I would describe my experience with a particular algortihm. I choose YouTube’s Watch Time algorithm and started to investigate ways in which this particular piece of code helps to recommend videos for its users.
I decided to not use my original YouTube account as that one had been overrun, with video games, gadgets and puppy videos. After creating a new account I ventured onto YouTube and started exploring topics. I thought perhaps I could go my usual route and watch videos on gadgets but instead opted to mix it up a bit. Instead I decided to explore videos that I would not normally watch. This way I would be able to see significant data for this analysis.
So after venturing onto YouTube, new name and all I tried to actually test and see how this algorithm actually works.
The Watch Time Algorithm attempts to recommend videos based on the duration that a particular video is watched. In other words, the more people that watch a particular video until the end of its run time, the more likely that video will be recommended ahead of others. I wanted to test this particular algorithm (or attempt to at least) by see how videos would be suggested based on how long I watch them.
I created my new account, (see Fig. 1) which pays homage to one of my favourite movies and afterwards was presented with a “What to Watch” front page filled with video suggestions which were linked or related to where I lived (see Fig 2). In case anyone was wondering I live in Taiwan, so the recommended videos will be those trending in the country.
Fig 1: The new account I setup for this experiment.
Fig 2: The What to Watch page which is presented as soon as I registered with YouTube
I decided to beginning my search with something I am interested in and that is mobile gadgets. Being an Asus fanboy watched a review video of the Asus Zenfone 2 (see Fig. 3). Once the video had completed I refreshed my home page and was given suggestions of a channel and videos related to what I had just watched (see Fig. 4).
Fig 3: The first video I watched was about a mobile phone.
Fig 4: Recommendation videos about the phone I searched were now on my What to Watch page. Specific channels are also presented.
Now I decided to try something I little more obscure (well for me anyway). I looked up videos on Barbie dolls. After typing in my search, I decided to watch one video and let it run all the way through to the end (see Fig. 5). The next step would be to refresh my home page and see which videos were recommended. To my surprise the list of videos did not necessarily include Barbie videos but had clips surrounding the topic of toys for girls (see Fig. 6).
Fig 5: A video about Barbie Dolls which was watched in its entirety
Fig 6: What to Watch page was revisited there were no specific Barbie videos but videos about other girls toys.
The next experiment was looking at what would happen if I watched a video for 20 seconds. For this test I decided to look up the topic of mascara (it was my wife’s idea).She helped me pick a video and I watched the mascara video for twenty seconds (see Fig 7) and after refreshing my homepage did not find anything relating to the channel I watched or even suggestions of other mascara related videos (see Fig 8).
Fig 7: Video about mascara that was watched for 20 seconds.
Fig 8: After watching the 20 second mascara clip, no make-up or specific mascara videos were recommended.
I decided to go back to the video and watched it through to the end (see Fig 9). After viewing the video, I knew exactly which was the best mascara to buy for under 10 pounds! Not only that, but I visited my YouTube homepage and found recommendations from this particular channel (see Fig. 10).
Fig 9: Video was watched a second time but this time until the end.
Fig 10: Videos were suggested from the channel that hosted the mascara video.
Now this playful experiment is not too conclusive but it does show some sort of algorithm at work here. Suggestions of videos were made for based on the short amount of time I spent on the website. My YouTube page evolved in a matter of minutes. From what I have seen, one could learn quite a lot from a person’s account and recommendation list. To be honest, I feel that YouTube knows me better than I know myself.
Oh well! Time to watch some more mascara videos.
On to Week 9.