On the Research Gap Regarding the Study of the Content, Dissemination, and Life Cycle of Memes

     There are numerous avenues for additional research on the topic of how the content of memes affects the how they spread, live, and die. In my lit review, I discuss one experiment by Barnes et al. (2021) in which the researchers use 129,000 memes scraped from reddit and fed into machine learning models for training so they could use it to predict what factors in the text and images of memes make them most likely to go viral. Yet, their entire sample was taken just on the first week of the COVID-19 pandemic in the US (mid-March 2020). The research team acknowledged that the timing, extraordinariness of the moment, and short span of data collection were all limitations that potentially affect the generalizability of their results. 

    Another study using machine learning by Ling et al. (2021) used a much larger sample of 160 million meme images collected over a period of about 10 years, but the place they gathered their sample from deletes posts periodically, making replicability impossible. 

I would want to propose an experiment similar to Barnes et al (2021). but with a sample gathered similarly to Ling et al, (2021) except with a sample location that isn't ephemeral. Doing so would overcome the issues of both samples and provide a much bigger sample overall to train the machine learning models to use for predicting what meme content factors impact viral spread. 

There is just one major problem. I'm not qualified/don't know anywhere near enough about:

  • how to scrape and cluster big data samples
  • how to use or train machine learning programs
  • how to do advanced analysis on results that would be necessary.
Basically, if I were to do this or even plan/propose this experiment, I would need to collaborate with one or more experts in these areas. Conducting the experiment would be an even bigger challenge.

However, I also propose another meaningful study option:

This would likely be a mixed method study in which known memes taken from Know Your Meme are matched with their Google Trends data showing the times and popularity of memes as well as the geographical distribution of where memes were being looked up as ways to measure meme spread and lifecycle. The data would be aggregated, and statistical methods would be used to analyze trends and patterns. 

On the qualitative side, I could examine certain interesting cases illustrative of, perhaps, variations on meme life cycles. Memes of interest would be singled out for in depth qualitative analysis on how their content affected their virality and life or death, etc. I mention a study in my review, Jones et al., (2022) that gave me the idea for both aspects of the study. They did a case study of just 3 memes and their histories, including Google Trends data. They didn't aggregate or do quant analysis on their memes.

doing both approaches in one study (quant first) allows one to flag which memes and their histories are worthy of case study. The two in combination allow us to learn broadly about meme life cycles and specifically about how content plays a role in meme uptake and life. after all, there is still much to learn about why certain memes are more successful than others, or why, after a while we hardly ever see them again. 

I probably know enough to be able to propose and even conduct this study. It is also somewhat original. I haven't seen anything like this so far in the research I've conducted or seen referenced. 

 Let me know what you think about which direction I should go. 

Comments

  1. Hi Mike,

    I really like your idea on qualitative study here that kind of mirrors Jones et al. (i.e., "On the qualitative side..." paragraph). If it's me, I'd replicate Jones et al.'s qualitative approach but on a diff. set of data (i.e., another context/set of memes/etc.) to add to the literature that is specific to the field of TWDR. You can also get mileage out of this topic and become a known scholar on memes with more studies for you to conduct after your first findings, this time with quantitative analysis in collaboration with a statistician (I wish Utah Tech has an operating Stat Center), and so on. Of course, if you'd like to design a method involving both quant. and qual. approaches, that would be good too. Let me know if you want to Zoom -- am here. :)

    ReplyDelete

Post a Comment

Popular posts from this blog

Reflections

Prewriting and 10 tentative sources

Revisions, future direction, and peer response