The TEA Nets framework combines AI and cognitive network science to model targets, events and actors in text

The TEA Nets (Target-Event-Agent Networks) framework, introduced by Sebastiano Franchini and colleagues from the University of Trento and other institutions, provides a computational approach to extract subjects (Agents), verbs (Events), and objects (Targets) from natural language texts. Built as an open-source Python library, it sits at the intersection of artificial intelligence and cognitive network science. The researchers tested TEA Nets on three distinct case studies to demonstrate its versatility.

In the LOCO conspiracy corpus, TEA Nets analyzed 4,227 texts. High-conspiracy narratives connected personal pronouns ("I", "you", "we") with the same actions twice as frequently as low-similarity conspiracy narratives. They also linked person-focused elements ("you", "people") to actions eliciting anger significantly above the random baseline (z=2.63, p<.05). In contrast, low-similarity conspiracy narratives emphasized scientific actors ("researcher", "scientist"). For the HOPE and CounseLLMe datasets, which include 212 human and 200 LLM-based psychotherapy transcripts respectively, TEA Nets revealed emotional nuances. When expressing feelings, Claude 3 Haiku, GPT-3.5, and humans all used sad words more frequently than random expectations, but Haiku expressed sadness with lower emotional intensity than humans (U=1243.5, p=.036). These findings have implications for LLM-simulated patient training in psychotherapy.