Mapping automatic social media information disorder. The role of bots and AI in spreading misleading information in society

Abstract

This paper presents an analysis on information disorder in social media platforms. The study employed methods such as Natural Language Processing, Topic Modeling, and Knowledge Graph building to gain new insights into the phenomenon of fake news and its impact on critical thinking and knowledge management. The analysis focused on four research questions: 1) the distribution of misinformation, disinformation, and malinformation across different platforms; 2) recurring themes in fake news and their visibility; 3) the role of artificial intelligence as an authoritative and/or spreader agent; and 4) strategies for combating information disorder. The role of AI was highlighted, both as a tool for fact-checking and building truthiness identification bots, and as a potential amplifier of false narratives. Strategies proposed for combating information disorder include improving digital literacy skills and promoting critical thinking among social media users.

Fig 1. Types of Information Disorder (ID).

Adapted from [21].

Fig 2. The PRISMA workflow.

Left panel shows the Prisma part of the literature review. Right panel reports the processing part of the workflow.

Fig 3. Evolution over time of academic production.

Misinformation, disinformation, and fake-news are investigated collectively.

Fig 4. Stratification of the documents collected.

Amount of documents analyzed per type (left panel). 94.8% Original Articles vs. 5.2% Reviews; Percentage of subject area to which the articles pertain (right panel).

Fig 5. Documents distribution per subject area.

How highlighted by Pareto diagram, ID topics are addressed for the most (∼ 70%) by four areas: Social Science (196 papers), Medicine (169), Computer Science (162), and Engineering (60).

Fig 6. Line plot implementing the elbow method for our corpus.

Sum of Squared Errors (SSE) vs Number of clusters. The absence of any clear elbow means that further explorative evaluations are needed.

Fig 7. LDA algorithm results.

The topics projection on the most informative subspace (ℝ²) of the space derived from the Principal Components; the size of the circles is proportional to the marginal topic distribution (i.e., the number of words/terms covered by the topic).

Fig 8. Bag-of-words relative to the six clusters identified.

Note a certain degree of terms and topics overlapping, reflecting the unclear behavior of the line plotted for the elbow method.

Fig 9. The documental corpus in an initial stage of the topic elicitation through the Obsidian software.

Note the cluster around tag #facebook formed by red nodes (papers containing the word “facebook”) and light-blue nodes (papers containing the word "twitter”).

Fig 10. The birth of topic 4.

The newly born Topic 4 emerges during the initial phase of topic elicitation in the knowledge graph.

Fig 11. Excerpt of the documents-social networks matrix.

The matrix (size 283X15) reports the strength of relationships between documents and social media platforms.

Fig 12. Excerpt of the documents-topics matrix.

The matrix (size 283X6) reports the strength of relationships between documents and topics.

Fig 13. The social media platforms-topics matrix.

Each cell reports the corresponding strength of relationship. The bottom row represents the total sums of the strengths per topic. The rightmost column reports a weighted score for each social media platform.

Fig 14. Correlation matrix between ID types.

The ID types, misinformation, disinformation, malinformation, are correlated with the most widely used social networks worldwide.

Fig 15. AI bots vs social media platform.

Correlation matrix between AI behavior addressed in the screened papers and the most widely used social networks worldwide.

Fig 16. AI bots vs topics.

Correlation matrix between AI behavior addressed in the screened papers and the major topics of dissemination of the ID.