PolyU has developed an “implicit emotion classification” method for deducing emotions hidden in the expressions in the context of emotion-related events.

Emotion classification is a key aspect of emotion analysis and a hot research topic in natural language processing. Studies often use keywords such as “happy” and “sad” in performing emotion classification, but some expressions contain no such words although they do convey emotions; these are referred to as “implicit emotion expression”. For example, an expression like “I have become big goat. Whoever a little goat met, it always said that, ‘you see, I have grown up.” does not contain any emotion keywords, but it carries the emotion of “joy”.

 

Prof. Huang Chu-ren

From the perspective of linguistics, emotions and events are interwoven – an event can trigger a specific emotional state, and an emotion can cause a particular event. To understand this interrelation, Chair Prof. Huang Chu-ren and Dr Sophia Lee Yat Mei at PolyU’s Department of Chinese and Bilingual Studies partnered with Dr Li Shoushan at Soochow University to conduct a study of effective emotion classification method, and develop a stochastic model for automatically identifying and classifying emotions and events.

The researchers first deleted emotion keywords from expressions used during emotion-related events, and then identified five emotions implicit in the resultant expressions: joy, anger, sadness, fear and worry. Statistical calculations and consistency and accuracy tests proved that the “implicit emotion expression” classification method was able to effectively comprehend the emotions hidden in the expressions without emotion keywords.

In addition, the researchers found that the meaning carried in some expressions during emotion-related events differed from one sentence to another, making it particularly difficult to identify implicit emotions such as “anger” and “worry” when keywords were absent.

The implicit emotion classification method is nevertheless set to help other researchers further understand people’s emotional states and the cognitive principles behind the interaction of language and emotion. It can also help to analyse their views on triggering events, based on the assumption that those views and emotions are interrelated. In other words, the tendency to hold certain views and standpoint on events could be observed through the analysis of emotion expressions. This approach can provide useful tools for discovery. ♦