In the first post of this UNICEF series, we covered how to run a social media optimization (SMO)
baseline analysis on UNICEF’s Facebook posts. We also introduced the concept of associations in that post.
From the 2,917 Facebook posts we analyzed, we extrapolated that the posts revolve around 3 thematic clusters: Family,
Refugees + Migrants, and
School+ Education.
In this post, we will focus on the Family cluster.
Top 3 clusters identified by our embedding models based on top words
In the next posts, we will also take a look at what the
Refugees + Migrants analysis and the
School + Education analysis showed.
Our goal is to identify exactly which associations drive successful social media posts and images within the Refugees + Migrants cluster. Why? Because, by knowing what works and what doesn’t,
SMO can reach its maximum.
Big Data, Machine Learning, and Social Media Optimization
In this article, we will take a look at the following, in order to establish strong and weak points of UNICEF’s existing approach to social media:
- Text analysis of this cluster
- Text associations that led to HIGH engagement
- Text associations that led to LOW engagement
- Examples of text+associations that were SUCCESSFUL
- Examples of text+associations that were UNSUCCESSFUL
- Image analysis for this cluster
- Image associations that generated HIGH engagement
- Image associations that generated LOW engagement
- Examples of Images
- SUCCESSFUL images
- UNSUCCESSFUL images
- Image analysis: Text on o image
- Image analysis: Activation maps
Let’s get started!
Text Analysis for Family Cluster:
From UNICEF’s 2,917 Facebook posts, 419 posts were classified as belonging to the Family cluster.
A text analysis from these 419 Facebook posts showed that certain associations were correlated with successful and unsuccessful posts.
Correlations of associations invoked in successful posts (represented in yellow bars) and unsuccessful posts (gray bars)
Text Associations with High Engagement
The most successful posts in the Family cluster were associated with the following major descriptors and related terms:
- Colorful: Colour, yellow, purple, color, blue, dark, and bright
- Goofy: Cartoon, kooky, lovable, anthropomorphic, funny, minion, and clown
- Glowing: Glimmer, shine, glow, glint, glisten, and shimmer
With this information, in the future, it is possible to draft posts that invoke these associations. For future posts, we recommend that UNICEF drafts text that specifically correlate with these high-engagement associations.
Text Associations with Low Engagement
However, unsuccessful posts in the Family cluster were associated with:
- Trauma: emotional, physical, mental, trauma, anxiety, depression, cognitive, physiological, emotion
- Evacuation: refugee, resettle, resettlement, migrant, idps, shelter, asylum_seeker, evacuate, impoverish
- Medical staff: medical, hospital, patient, doctor, pediatric, nurse, medicine, clinician, paediatric
- Malnutrition:malnutrition, famine, starve, poverty, malnourished, epidemic, hunger, malnourishment, impoverish
- Charity:volunteer, charitable, donate, underprivileged, unicef, donation, ngo, fundraising, nonprofit
From the associations above, Charity was the association that led to the least success and engagement.
With this information, you know what kind of associations to avoid since they are associated with low success and engagement. Such insights are crucial for an effective social media optimization strategy.
Examples of Successful Posts and Associations: Social media optimization success
Below, you can see how the highly successful associations (like Colorful, Goofy, and Glowing) are identified and measured in real-life Facebook posts categorized within the Family cluster:
The green highlights are used to visualize the strength of the goal association and the word. This is made possible via the in-browser
NEURO FLASH chrome extension. The extension works while you are logged in to your account. For your free 3-day trial,
sign-up here.
Colorful Association
The following three posts were significantly correlated with the colorful association:
Colorful association
Colorful association
Colorful association
Goofy Association
Next, the goofy association also worked well within the Family cluster. The following three posts were exemplify the goofy association:
Goofy association
Goofy association
Goofy association
Glowing Association
The glowing association is another example of associations that were highly correlated with success within the Family cluster. Below are text examples of posts that embody the glowing association:
Glowing association
Glowing association
Glowing association
Examples of Unsuccessful Posts and Associations
In order to have an accurate social media optimization approach, you need to know what kind of posts to avoid.
Below is the analyzed text of a few posts which showed to have low engagement scores, representing the following associations: temperature, illness, and water body.
Malnutrition association
Temperature association
Medical staff association
Charity association
Image Analysis for Family Cluster
To supplement the text analysis, an image analysis was also performed, using 419 images.
For the Family cluster, correlations of image associations appearing in successful posts (represented in yellow bars) and unsuccessful posts (gray bars)
Image Associations with High Engagement
As shown above, the image analysis revealed that the most successful posts within the Family cluster had the following associations:
- Legendary: Iconic, famous, pay_homage, heritage, masterpiece, timeless, contemporary, renowned, and tradition
- Adventure: Perilous, embark, journey, treacherous, adventurer, adventurous, explore, and expedition
The following associations were also correlated with successful, engaging posts:
- Workshop: workshop, participate, meeting, attend, attendee, facilitator, participant, session, training
- Cinematic: film, filmmaker, actress, screenwriter, documentary, oscar, filmmakers, biopic, motion_picture
- Instruments: piano, musician, violinist, musical, pianist, cello, virtuoso, guitar, saxophone
Image Associations with Low Engagement
Furthermore, the image analysis showed that the least successful images for the Family cluster were associated with:
- Rape: rapist, murder, kidnap, perpetrator, accuse, arrest, convict, victim, killing
- Tragedy: devastate, terrify, horrify, tragic, devastation, survive, aftermath, frighten, ravage
- Medical staff: medical, hospital, patient, doctor, pediatric, nurse, medicine, clinician, paediatric
- Trauma: emotional, physical, mental, trauma, anxiety, depression, cognitive, physiological, emotion
- Illness: illness, symptom, disease, acute, sufferer, pneumonia, debilitate, ailment, syndrome
From the unsuccessful associations, Illness is the association led to the lowest level of post engagement within the Family cluster.
Examples of Images
Now, let’s move away from abstract ideas and take a look at what these associations look like as images. In this section, we will see what successful and unsuccessful images look like.
Successful Images and Associations
Below are examples of images that were associated with high engagement and their respective associations:
Workshop association
Adventure association
Legendary association
Unsuccessful Image Posts and Associations
Below are images, and their respective associations, that showed to have the lowest levels of engagement:
Medical staff association
Trauma association
Illness association
Image Analysis for Family Cluster: Text on image
Also, we assessed whether images that have text on them were successful or not. As shown below, for the Family cluster, images without text performed significantly better.
For the Family cluster, images that did not have text on them were significantly more successful than images with text.
Image Analysis for Family Cluster: Activation
Activation maps
Additionally, we used activation maps to show that high activity was occuring at spots where touch was occuring. What does this mean? Since the three images above were strongly successful within the Family cluster, touch is a driving factor of that success. Thus, touch an additional positive element to take into consideration when drafting Facebook posts within the Family cluster.
Conclusion
Using big data insights and machine learning tools, we were able to pinpoint exactly what led to high success for UNICEF’s posts. By isolating the associations and stylistic factors that are associated with success, UNICEF’s SMO strategy will automatically improve.
Now, we know what factors contribute to the success of a social media post. For the Family cluster, specific guidelines can be established, defining which associations for texts and images are most effective.
Therefore, piece by piece, a detailed social media optimization strategy is forming which outlines which factors to use and avoid.
Related Blog Posts: Social media optimization case studies
For more info, check out these related blog posts:
- Baseline analysis: The first blog post of the UNICEF series which explains the baseline analysis and basic procedures followed.
- Refugees + Migrants: See our detailed analysis on the Refugees + Migrants cluster for the factors that led to success.
- School + Education: Check out the analysis on the School + Education cluster for a specific report on what worked best for this theme.
Top 3 clusters identified by our embedding models based on top words
