In the realm of social media analytics, a fresh wind is blowing from the University of Massachusetts Amherst, where Assistant Professor Viviana Chiu Sik Wu has been making strides in refining AI's ability to trudge through the vast data swamps of Twitter/X. Her recent endeavor, as reported by UMass Amherst, demonstrates an "extreme boosting" model that combines the brute strength of machine learning with a touch of human discernment to better categorize content.
The increasing challenge has been how to sift through the deluge of tweets, retweets, likes, and shares that make up social media interaction—a task that often falls through the cracks of AI due to its tendency to misinterpret nuances or become encumbered by a variety of variables. Wu found that a majority of the 43 studies she systematically reviewed relied on manual coding, which is time-consuming and non-scalable. This stands in stark contrast to the nuanced detection promised by AI. Despite the limitations of unsupervised machine learning, which Wu points out can "introduce biases and noise into the results," her combined approach aims to provide a structured powerhouse capable of handling massive dataset analyses, according to UMass.
To put her model to the test, Wu embarked on a Twitter/X excavation, mining through 66,749 tweets from the platform's community foundations between 2017 and 2018. Upon personally dissecting 15% of these messages, she honed her AI model with this manually analyzed data to classify the remaining bulk of tweets. The focus was on pinpointing posts that fostered public engagement — a particularly slippery fish to catch due to their content overlap with seemingly similar topics like fundraising activities or grant announcements.
After the bytes had settled, Wu's AI flagged a pool of 6,331 tweets, all linked to public engagement, indicating a successful first dive. However, she stresses that further refinements are needed to bring the model's accuracy to its apex. "The findings can be extended to situations in other fields well beyond nonprofits to analyze massive observational datasets on social media," Wu elaborated on the broader applications of her research, as mentioned in the same news source. Despite her progress, Wu notes an uptick in difficulty for researchers accessing social media data, a consequence of tightening platform policies. This has sent academics scrambling towards alternative data sources like Reddit and TikTok, in search of a diverse stream of digital information.
The reverberations of Wu's work suggest a potential leap in how we utilize AI for parsing social datasets, which could transform not only nonprofit strategies but also a multitude of sectors swimming in social data. People interested in the granular details of Wu's study can delve into the full research published in the Journal of Chinese Governance.
