Visualizing posts and conversations around PCOS
Polycystic Ovary Syndrome or PCOS is a condition of hormonal balance in women and people with female reproductive organs that causes irregular menstruation and is often linked to infertility
Signs and symptoms of PCOS vary for different people. Some of the most common ones are:
- Irregular periods
- Polycystic ovaries
- Excess androgen (or male hormone) in the body
Even diagnosing PCOS is not straightforward!
It is very easy for people to miss early indicators or even misattribute some of these symptoms (e.g., hair fall, sudden weight gain) to other health conditions like Thyroid or simply to things like stress or bad lifestyle.
….What’s more?
Stigma!
People with PCOS often end up being stigmatized because of their health. What causes this stigma?
- PCOS being linked to infertility
- Women feeling “ugly” because of acne, hair fall, facial hair, excessive weight, etc
- Accompanying mental and emotional health issues
So people often turn to (anonymous or pseudonymous) online platforms to open up about these stigmatized topics. Such platforms provide a dedicated space to connect with people experiencing similar conditions, seek and provide support to one another.
Our paper on “Living with Uncertainty and Stigma: Self-Experimentation and Support-Seeking around Polycystic Ovary Syndrome” discusses how people with PCOS use a pseudonymous platform (unnamed for ethical reasons) to converse about sensitive details of their condition which they might not be comfortable sharing verbally. The community acts both as a social support platform and an information source. Both diagnosed and undiagnosed people use it for multiple reasons:
- Seek conversational support
- Share personal experiences
- Ask for advice
- Check the normalcy of their experiences
- Learn about other’s experiences
- Increase their aggregated knowledge of PCOS, etc
In this project, I take a closer look at the data we qualitatively analyzed. I created visualizations to understand the engagement on different posts and conversations around PCOS. Here are some of my guiding questions.
RQ1: What kind of posts are more popular? How does post engagement vary depending:
Who wrote it — Nature of user
What it talks about — Topics
How the user talks about — Type of post
What is the overall emotion of it — Sentiment
RQ2: How do the nature of users, topics, type of posts, and sentiment correlate with each other?
RQ3: Who are some of the most active users? What kind of posts do they make?
PRODUCT DESIGN
I explored the dataset and started by creating some basic pie-chart visualizations for the different coding schemes. Wanting to know more about the textual content of the posts, I created word cloud visualizations for each of the post categories. Next, I went on to make more complex visualizations to understand (a) the popularity of different posts, (b) the user interaction dynamics, and (c) the relationship between different post coding schemes.
Here are some of my initial product design ideas!
I iterated and reflected on each of the visualizations; discarding and refining them based on what I wanted to learn from the data. For example, here is an iteration of a scatter plot:
Two important factors I realized I had to be mindful of while creating digital versions of these were:
- Scalability — Might have to sample and visualize a portion of the data
- Accessibility — Felt that some of my designs might not be accessible (e.g., similar color usage can be problematic for people with low or no vision) and would need to think of ways to address that (e.g., using patterns instead of colors?)
BETA RELEASE & EVALUATION
Next, I created digital versions of the visualizations using Tableau and Matplotlib.
My goal — to generate multiple visualizations that tell a story about the data and help me draw nuanced insights.
User Feedback & Proposed Changes
I got three users to evaluate the beta version and enacted some of the feedback. Here is some of it!
Feedback: The color scheme lost consistency and users got confused between the sentiment and type of post encoding schemes
Fix: Reworked the color schemes to make them significantly different for different categories and consistent across different dashboards
Feedback:
- Users were skeptical about the placement and size of bubbles in the bubble chart. They weren’t sure if it was adding any new insights to the dashboard.
- Users were confused by the histogram visualization as one of the bars had 2 colors — pink (positive) and orange (neutral) together.
Fix: Removed both the confusing and redundant visualizations, clubbing the two sentiment-based dashboards into one!
Feedback: Users felt the scatter plot visualizations could be made more readable by reducing the size of the data points and intuitive by changing the shapes used for different sentiments.
Fix: Reduced the size of the data points changed the shape of the marks as shown!
FINAL RELEASE
I addressed feedback from the user evaluations; reworking the dashboards and creating interactive word clouds using Highchart.js.
Here is a video of my final release:
Some Key Insights
The goal of my project was to dive deeper into a PCOS-related online community and learn more about the kind of conversations happening on it. Here are some key insights (and thoughts) from the case study:
F1: Majority posts on all topics (except stigma/lack of awareness) were seeking recommendations
Are people okay living with the stigma, just expressing emotions from time to time? Or is there a different majority sentiment around stigma which maybe drowns in the forum’s support-seeking aspect?
F2: Majority posts seeking recommendations have a positive sentiment.
F3: The most common occurring post topic is physical, mental, and emotional manifestations of PCOS (e.g., effect on appearance because of weight gain, acne, hair fall etc, accompanying mental health issues or emotional outbursts).
So are people more interested in discussing the manifestations of PCOS rather than its medical details?
F4: Posts near-neutral polarity have more subjective content.
F5: Bell-curve of post polarity is almost balanced, slightly more towards positive sentiment
F6: The majority of posts seeking recommendations had lesser post score and number of comments
F7: More sharing experiences posts about staying active had people sharing positive experiences while diagnosis-related and stigma/lack of awareness posts had people sharing negative experiences.
Interesting to see that people share mostly negative experiences when it comes to medical details and diagnosis.
F8: Posts sharing positive experiences seemed to have slightly higher popularity in terms of score and comments as compared to those sharing negative experiences
F9: Posts sharing negative experiences may not be using “very negative terms” according to the sentiment classifier.
Are people in some way restraining themselves from venting or using very negative terms while doing so? The general dynamic of the forum seems to revolve around support seeking rather than venting. Do people want a space to vent out more?
F10: Positive and negative sentiment posts seemed to be talking about similar things but in different connotations.
Is this because people may be talking about both positive and negative experiences in their posts?
F11: Some symptoms (e.g., hair fall) that undiagnosed people associate as indicators of their undiagnosed PCOS may not be excessively spoken about by people who have already been diagnosed.
Looks like there is potential for using such forums for sense-making and even discovering/diagnosing conditions that haven’t been identified yet!
And many more…
These and many more interesting insights (and open questions) feed into my research on how we may improve existing online platforms to better support people’s support-seeking, sense-making, and self-experimentation practices.
CONCLUSION & WHAT’S NEXT
Visualizing the dataset helped me gain deeper insights into “what” and “how” users on a pseudonymous online forum communicate about PCOS. I got a big picture idea about which users were talking about what topics and in what manner. I also spotted trends in comments and post scores depending on the nature of the post. All the visualizations and dashboards tell a story about the nature of conversations happening around PCOS. The findings indicate the potential of tailoring online support platforms according to people’s needs. Future work can explore building statistical models that classify users and their posts with the aim of customizing support-seeking and experience-sharing in online communities.
References
[3] https://www.highcharts.com/
[5] https://textblob.readthedocs.io/en/dev/quickstart.html#sentiment-analysis