From understanding what is said to measuring how much trust it deserves
Since 2009 I have worked on how a machine can read what gets published and understand more than words: sentiment, irony, emotion that does not match what is being said and, in the end, how much trust a piece of information deserves. These are the research lines along that path, summarised, with links to what I have published on each.
Five questions, one thread
They all start from the same question: what can be known about what gets published, at scale, beyond the words.
Disinformation
How falsehood spreads and how to measure how much trust a piece of information deserves. From detecting misleading headlines to FactCheck.
Read the summary → 2007 – 2022Sentiment and emotion
From reputation analysis to sentiment analysis and affective computing: what the writer feels and what the reader perceives.
Read the summary → 2007 – todayIrony and sarcasm
Why irony fools the systems that measure opinion, and how to start detecting it.
Read the summary →Cyberbullying detection
Detecting harassment of minors and vulnerable people on social networks, with Jimcrick.
Coming soonThreat detection
Active listening applied to security: spotting the impostor in a conversation.
Coming soonThe path, with its evidence
Most of the articles are in Spanish.
- 2009From reputation to sentiment analysis
Measuring what is said about a brand is not enough: you need to know the tone.
- 2011Technologies, branding and sentiment
Perception as something to be measured.
- 2013Irony, sarcasm and tone in measuring intangibles
The first serious obstacle: what is said is not what is meant.
- 2014Paper in Procesamiento del Lenguaje Natural (SEPLN), no. 53
Natural language processing and semantic technology in Brand Rain and Anpro21. With Silvia Suárez, Remei Bori and Oriol Flo.
- 2014The road to irony detection · Video reputation and crisis detection
From irony in text to reputation in video.
- 2017Fake news in politics: a challenge for democracy
The starting point of the disinformation line.
- 2022AI and affective computing · Measuring sentiment on social networks
Emotion as a signal, not as noise.
- 2023Detecting fake news I-III and the impostor technique
Semantic headline detection, two-stage classification and who said it first.
- 2024-2025The AI narrative in the Spanish and Portuguese press
59,324 articles from 942 outlets, with the University of Valladolid and the University of Beira Interior. In preparation.
- 2026FactCheck
A 0-100 trust score for a headline, with its evidence, built on the GeriAI analysis engine.
Publications
«Aplicación de tecnologías de Procesamiento de lenguaje natural y tecnología semántica en Brand Rain y Anpro21» (Natural language processing and semantic technologies: the application on Brand Rain and Anpro21)
O. Trabazos, S. Suárez, R. Bori, O. Flo · Procesamiento del Lenguaje Natural no. 53, pp. 201-204, Spanish Society for Natural Language Processing, September 2014.
In preparation
"The AI narrative in the Spanish and Portuguese press" (in preparation)
Study with Pilar Sánchez-García (Universidad de Valladolid) and João Canavilhas (Universidade da Beira Interior): 59,324 articles across 942 outlets analysed (June 2024 – June 2025, 15 variables), with TrawlingWeb as technology partner and GeriAI for the analysis. In preparation for Media and Communication.
Working on any of these lines?
Universities, media and institutions: let's talk about projects, data and talks.
Contact