Hulkshare Recommendation Algorithm Challenge
Dear @ZINDI!
Could you please clarify some moments about data? Here is a screenshot from StarterNotebook: https://disk.yandex.ru/i/rJ_OYRd1Nr67Kg
My questions are written in magenta color:
1. How could straight intervals in certain video and session have different slopes? Is there a kind of playback speed change?
2. How can the slope be negative (for ~850-950 frames)? It seems like reverse playback.
UPD:
3. What is vidTime? It has correlation = 1 with the 'frame' column, so it could not be the duration of the current session. Please fix the data description https://zindi.africa/competitions/hulkshare-recommendation-algorithm-challenge/data
4. There are some pairs (video_id, session_id) having more than 1k frames and (e.g (video_id='ID_00FMRXW972', session_id='459deba29f86115427c7ae2f36a0f72747f984e5' has 2k frames and vidTime points). What does it mean? Whether the data is corrupted (duplicated)?
Hi
Can't clarify everything, but yes, the playback speed may change. And also the user can skip part of the video, go back and forward.
It will be good if we can have more clarity on the data to understand best way to approach the problem. For the same vid id, two sessions showing the same linear pattern but with different slopes have different session duration and VidTime. For instance, one shows 3s and another 1200s.
Both show no pauses, forwards, rewinds. I'm wondering did the session of 3s actually listen to a song in 3s for a video duration, that should be around 1400s. Can we say both listened to the same song and completed the session without pauses, rewinds, forwards?
My thinking was for a linear listening pattern, the session durations should be similar. Is the video duration given determined for a specific playback speed?
Thank you for the clarification @Zindi @DmitriyM
session duration is the time that the user spent watching the video. The video may be 2000s long, but one user watched it for 3s and another one for 1400s.
Thank you very much for the clarification.
I think I understand better now, so either way a listening session is split into 1000 frames. I see the one with 3 seconds didn't finish the song.
Thank you.
Yes. Each session is split into 1000 frames. Each frame represents session_duration/1000 (s). So if the session_duration changes, the frame size also changes, but the amount of frames don't.
This helps a lot. Thank you for the clarity
1. How could straight intervals in certain video and session have different slopes? Is there a kind of playback speed change?
PLAY, FAST FORWARD, REWIND, PAUSE. The slope of the line represents a viewing pattern. The spot is a visual function representing mode per time.
your welcome to contunue the conversation on our discourd https://discord.gg/TwsnzK8k