A research team from the University of Jaén (UJA) has developed a system based on artificial intelligence to recognize the emotions of users on Twitter. This technology, applied for the first time to the Spanish language, perceives the state of mind of people who write messages on the social network, analyzes and classifies them.
As reported by the Discover Foundation, the researchers direct the information obtained to areas such as the detection of depression, anorexia and bulimia or abusive and violent language, among other areas.
Human language technologies are a branch of artificial intelligence that focuses on the study of computer systems that are capable of understanding and generating language. This area is related to machine learning, which is the ability of software or machine to identify and learn complex patterns in the form of mathematical algorithms autonomously.
Experts apply this technology to a set of data made up of tweets previously collected and analyzed by humans in order to detect emotions in the text. Further, they teach the machine how to interpret new terms in Spanish by incorporating dictionaries and lexicons To the system.
“This technology can be applied to different areas in order to detect mental health problems or verbal violence”, explained Flor Miriam Plaza, co-author of this study and researcher at the University of Jaén.
In the study entitled ‘Improved emotion recognition in Spanish social media through incorporation of lexical knowledge’ and published in the Future Generation Computer Systems magazine, experts train a computer system with a series of tweets already collected and previously interpreted in the Spanish language.
In this way, generates a language model that allows you to recognize emotions such as anger, fear, joy, and sadness. “It is a complex work because it is not a binary classification of negative and positive emotions. There are many nuances to detect joy, sadness or surprise, for example,” commented UJA researcher María Teresa Martín.
Once this basic information was integrated into the system, the researchers included new words from dictionaries and new words to expand the number of shades that it could perceive and increase its precision. This gradual language teaching, independent of the previously developed database, had the purpose of improving the effectiveness of the system.
After detecting the tweet, the system analyzes it and assigns an emotion based on the generated language model, in this case, Spanish. The experts observed in this study that the emotion most represented in the tweets was joy because it was easier for the system to detect than anger, fear or sadness, which have greater nuances.
The researcher from the University of Jaén Luis Alfonso Ureña has pointed out that “not a perfect process because this technology does not clearly perceive figures of speech such as irony, sarcasm or set phrases and, in addition, new expressions are continuously generated. “Therefore,” to perfect this system, the machine must be continuously ‘taught’ in a specific language, such as Spanish from Spain or British English “.
In previous studies, the Intelligent Access to Information Systems group focused on the detection of anorexia and bulimia and misogynistic and xenophobic language in social networks. Ureña explained that the research team focuses on human language technologies applied, among other areas, to the analysis of feelings in Spanish. In the future, the idea is “improve technology based on artificial intelligence and machine learning that we use to apply it to a wider variety of settings. ”
This research has been financed by the own funds of the research group Intelligent Access to Information Systems, by the European Regional Development Fund (Feder), the Living-LANG project and the Spanish Government Networks project.