1 of 9

CANIS Data Visualization and Foreign Interference

2 of 9

CANIS - Foreign Interference

Problem

The goal is to identify and understand the dynamics of foreign interference in social media.

This involves analyzing when specific content may be considered foreign interference, identifying the platforms with the highest occurrence of such cases, and understanding the type of actors involved.

3 of 9

Methodology

The project was developed in three stages:

1. We employ the 5W2H methodology for a thorough understanding and practical development of the problem.

5W2H

Who?

Where?

Why?

What?

When?

How?

How Much?

CANIS - Foreign Interference

4 of 9

4. Named Entity Recognition (NER):

Words classification in categories such as people, countries, and cities.

Methodology

2. Information Retrieval and Analysis in Python:

1. Data gathering:

Tweets extraction with web scraping from specific accounts using the Selenium library.

2. Information Preprocessing:

Treatment of stop words, upper/lower-case, and lemmatization for a consistent structure.

3. Sentiment Analysis:

Tweets classification as positive or negative using Vader Sentiment Analysis.

5. Definition of foreign interference proxy:

Flag identifying whether the content is foreign interference.

Tweets are considered foreign interference if sentiment is negative and the US or Canada (and its cities) are mentioned.

CANIS - Foreign Interference

The second stage includes retrieving and analizing additional information from Twitter using Python. Then, we gather the tweets from the provided state media outlets or actors leveraging on scrapping with the Selenium library. (click) So, we cleaned the tweets by removing stopwords, changing them to lowercase, and lemmatizing the words of the tweets. (click) Afterward, we employ a Vader sentiment analysis to infer the sentiment of the tweets. Here, we want to highlight that we also tested a BERT large language model classifier, but the running time was much better using Vader. (click) We also employed a Named Entity Recognition method to identify people, countries, and cities on the tweets. (click) Finally, we defined a proxy for foreign interference by flagging the tweets with negative sentiment and mentioning the USA or Canada and its cities.

5 of 9

Methodology

3. Visualization (Web APP):

CANIS - Foreign Interference

6 of 9

Demo

Link: https://canis-hackathon-app.onrender.com/

CANIS - Foreign Interference

7 of 9

CANIS - Foreign Interference

Conclusions

35% of the tweets are considered foreign interference, leading to 47 (24.6%) potential actors.
Based on the number of followers, the top influential actors are CTGN, China Daily News, and Xinhua News.
Among foreign interference-like tweets, Joe Biden and Xi Jinping are the most mentioned figures.
While Facebook demonstrates significant influence in terms of followers, YouTube exhibits the highest foreign interference rate.
During notable international events (Brexit, US elections, EU migration crisis), foreign interference increases.

How do we show the scale of China’s influence on social media?

China's impact on social media is perceived not only by the number of followers. It is also highlighted by: the number of (1) views (2) reposts, the (3) reached worldwide locations, and the (4) attention that major international events generate.

8 of 9

CANIS - Foreign Interference

- Optimize the tweets’ extraction code for more extensive and efficient analysis.

- Analyze tweets using NLP algorithms trained on large datasets, i.e., Large Language Models (LLM).

Next Steps

- Explore data-gathering methods from social media platforms, such as APIs, web scraping libraries, etc. to examine the dynamics of each platform.

* NOTICE that the application is not online (active) the whole time. It is necessary to wait 2-3 minutes for the system to start the app the first time.

9 of 9

“Unlocking data analysis insights empowers us to unveil foreign interference patterns, fostering a safer and more informed world”

- Someone not yet famous