WITI

Who Influences The Influencer

An experiment to understand the propagation of political influence on TikTok leveraging Artificial Intelligence.

Project Background

Why We Are Building the Tool

"Who Influences The Influencer" (WITI) is a collaborative initiative with Media Monitoring Africa (MMA) aimed at developing a proof of concept (POC) platform. This tool is designed to harness the power of artificial intelligence to analyse social media content, with a focus on political narratives and influence across platforms like TikTok.

The goal is to uncover the strategies and narratives that shape public discourse and opinion, providing a more transparent and informed social media landscape.

Methodology

Technical Approach and Methodology Overview

The WITI platform leverages cutting-edge web technology and AI to deliver insightful analyses of political content. The technical process involves several key stages:




  1. Data Collection and Upload: Content from selected TikTok accounts is manually uploaded to AWS.
  2. Data Conversion and Preprocessing: Using AWS services, audio and video data are transcribed into text, and images are processed to extract textual content. This text is then cleaned and normalised for analysis.
  3. Sentiment Analysis and Entity Recognition: The platform utilises various AWS AI tools for sentiment analysis and custom entity recognition, identifying political entities and overall sentiments from the text.
  4. Keyphrase Extraction: This process identifies mentions of potential sponsorships and political affiliations, cross-referencing with a predefined list of relevant keywords.
  5. Data Visualisation: The results are aggregated and presented on a user-friendly dashboard that displays political entities mentioned, sentiments, affiliations, and mentions of sponsorships, among other analytics.

Caveats & Limitations

  • Accuracy and Reliability: The model’s accuracy is contingent upon the quality and completeness of the training data, which may have inherent biases or gaps. Therefore, absolute accuracy cannot be guaranteed.
  • Data Quality and Diversity: The current dataset is limited to three TikTok accounts and content from the last year only, potentially limiting the scope and diversity of insights.
  • Model Interpretation and Errors: The nuanced nature of political discourse means that the tool might generate false positives or negatives, making interpretation challenging.
  • Work in Progress: The platform is continually evolving, with future versions expected to include additional features and capabilities that are not currently available.
  • Cost and Efficiency: Managing AWS cloud usage costs effectively is crucial to maintain financial viability of the tool.
  • Liability: Users bear full responsibility for how they use and interpret the tool’s outputs. MMA or its affiliates are not liable for any direct or consequential damages arising from its use.

In conclusion, the WITI project is a significant step towards understanding and deciphering digital influence in political spheres, leveraging advanced AI techniques to analyse complex social media data. However, stakeholders should be mindful of the tool’s current limitations and the iterative nature of its development.