Previous artificial intelligence developments were able to detect the emotional state of users after typing passages of fixed text, based on keystroke attributes associated with seven emotional states (anger, fear, shame, disgust, sadness, guilt and joy). And now, we learn that a student has designed an artificial intelligence (AI) learning system able to detect emotions and sentiments such as irony and sarcasm in emails and text messages. His artificial intelligence system could have applications in helping detect content that suggests "calls for help" or even suicidal intentions.
It is still easy to misinterpret emotions in an email or text message, despite the use of exclamation points and smiley faces. However, this may all change due to the efforts of a computer science student at the Technion-Israel Institute of Technology. Eden Saig developed an artificial intelligence learning system able to detect emotions expressed in a text message or email. The computerized learning system is based on recognizing repeated word patterns.
In a student project called "Sentiment Classification of Texts in Social Networks," Saig presents his AI system. The project won the annual Amdocs Best Project Contest. After taking a course in artificial intelligence at the Technion Faculty of Computer Science, Saig developed the system at the Technion's Learning and Reasoning Laboratory.
According to the computer science student, voice inflection and tone play an important role in our verbally communicated messages. They are conveying meaning. However, those nuances are lost in email and text messages. Writers who want to signify emotions rely on using images or "emoticons," to compensate.
As Saig said, "these icons are superficial cues at best" but they cannot express complex or subtle feelings present in the verbal communication. Pages intended to be humorous on social networks were titled "superior and condescending people" or "ordinary and sensible people."
According to Saig, such pages are very popular in Israel. Users submit suggestions for 'stereotypical sayings.' According to Saig, he was able to identify existing patters by observing posts to these groups. Based on his observations, he developed a method that enables the AI system to detect future patterns on any social network.
Saig realized that, since the content in these sections was everyday language, he can create a useful database with this content for collecting homogeneous data. The database was used to 'teach' a computerized learning system how to recognize slang words or patronizing semantics in text.
'Machine-Learning' algorithms were applied to the content, according to Saig, The results were used to automatically identify stereotypical behaviors found in usual social network communication. The computerized learning system examined 5,000 posts on social media pages in order to perform the quantification through statistical analysis. This way, the artificial intelligence system was able to recognize content structure identified as slang or condescending. Through text analysis, the repeated patterns were learned automatically.
According to Saig, his AI system can now "recognize patterns that are either condescending or caring sentiments." His AI system can even send a text message to a user if it thinks that the post may be arrogant."
Saig's AI system has many potential applications beyond clarifying feelings or emotions in interpersonal communication on social media. For instance, it may help detect content suggesting suicidal intentions when applied to other networking pages.
