K-means Clustering of Instagram Reels

Based on @thorfinn_wisdom's content

---

Intro

K-means clustering is an unsupervised machine learning method that groups data into clusters based on similarity. It works by minimizing the distance between data points and the 'center' of their assigned cluster. In simple terms, it tries to put similar videos together into the same group.

For this dataset of 196 reels, we applied K-means clustering on three key performance variables:

• likesCount
• videoPlayCount
• videoViewCount

Although the silhouette score recommended 2 clusters, we chose 3 clusters to better capture the variation in performance.

---

Variables Used for Clustering

Cluster Averages (Main Features)

Cluster 0: Likes = 3.66, Plays = 165.49, Views = 12.69 (n=108)
Cluster 1: Likes = 9.34, Plays = 171.66, Views = 18.77 (n=65)
Cluster 2: Likes = 10.13, Plays = 574.30, Views = 209.04 (n=23)

Interpretation

• Cluster 0 represents the majority of videos: low likes, modest plays, very low views.
• Cluster 1 shows moderate likes and slightly better views, but still limited plays.
• Cluster 2 stands out as the high-performance cluster with both high plays and high sustained views, meaning these videos held audience attention much better.

---

Variables Not Used for Clustering (for context)

Video Duration & Style

• Durations across clusters are almost identical (~6.6 seconds). Thus, length is not driving differences.
• Font sizes are consistent, suggesting stylistic choices like text formatting are not key differentiators.

Hour of Posting

• Cluster 0: Avg. post time ~14:22 (afternoon).
• Cluster 1: Avg. post time ~12:94 (around noon).
• Cluster 2: Avg. post time ~11:09 (late morning). This suggests earlier posting correlates with higher performance.

Song Usage

• Cluster 0 dominant songs: bossavibez.wav (20.4%), bossaguitar.wav (15.7%).
• Cluster 1 dominant songs: bossatamb.wav (15.4%), thorfinn1.mp3 (15.4%).
• Cluster 2 dominant songs: thorfinn1.mp3 (21.7%), springtheme.wav (17.4%). High performers lean on thorfinn1.mp3 and springtheme.wav, while weaker performers rely more on the 'bossa' songs.

Background Visuals

• Cluster 0: vin6.mp4, vin7.mp4 dominate.
• Cluster 1: vin4.mp4, vin6.mp4 appear most often.
• Cluster 2: vin12.mp4, vin1.mp4 are more common. This suggests stronger-performing videos favor different, perhaps more visually dynamic backgrounds.

Weekday Patterns

• Cluster 0: Most common = Thursday (16.7%), then Sunday and Wednesday.
• Cluster 1: Most common = Thursday (21.5%), followed by Wednesday and Saturday.
• Cluster 2: Spread evenly between Friday, Sunday, and Thursday (all ~17.4%). High performers (Cluster 2) don’t rely on a single weekday but perform steadily across multiple days.

Hashtag Usage

• Cluster 0: hardwork (49.1%), staydedicated (49.1%).
• Cluster 1: SelfGrowth (60%), Motivation (53.8%), VinlandSaga (46.2%).
• Cluster 2: hardwork (91.3%), staydedicated (82.6%), believe (65.2%). Strong videos (Cluster 2) use hashtags more consistently tied to discipline and belief, rather than broader motivation tags.

---

Cluster Profiles

• Cluster 0 – The Low Engagement Baseline: Most videos fall here. They gain some plays but fail to convert into likes or sustained views. Commonly use bossa-type songs and neutral backgrounds.
• Cluster 1 – The Moderates: Better likes-to-views ratio but still limited overall performance. Songs like bossatamb.wav and hashtags around self-growth dominate here.
• Cluster 2 – The Viral Set: Smaller in number but much higher in both plays and views. Characterized by thorfinn1.mp3 and springtheme.wav, backgrounds like vin12.mp4, and hashtags tightly focused on discipline.

---

Recommendation for Creators

This analysis shows that not all content is created equal—some combinations of songs, backgrounds, and hashtags drive much stronger results. A social media creator should:

• Prioritize songs like thorfinn1.mp3 and springtheme.wav, which consistently appear in high-performing reels.
• Use motivational hashtags tied to discipline and belief rather than generic growth terms.
• Experiment with posting earlier in the day (morning to late morning), as higher-performing videos tend to cluster around this time.
• Rotate backgrounds to include those linked to strong clusters (e.g., vin12.mp4), avoiding overuse of neutral visuals.

In short: Cluster 2’s recipe—focused hashtags, impactful songs, and earlier posting—provides a blueprint for boosting engagement and breaking out of the low-engagement baseline.