A Bootstrapping Technique to Annotate Emotional Corpus Automatically

In computational linguistics, the increasing interest of the detection of emotional and personality profiles has given birth to the creation of resources that allow the detection of these profiles. This is due to the large number of applications that the detection of emotion states can have, such as in e-learning environment or suicide prevention. The development of resources for emotional profiles can help to improve emotion detection techniques such as supervised machine learning, where the development of annotated corpora is crucial. Generally, these annotated corpora are performed by a manual annotation process, a tedious and time-consuming task. Thus, research on developing automatic annotation processes has increased. Due to this, in this paper we propose a bootstrapping process to label an emotional corpus automatically, employing NRC Word-Emotion Association Lexicon (Emolex) to create the seed and generalised similarity measures to increase the initial seed. In the evaluation, the emotional model and the agreement between automatic and manual annotations are assessed. The results confirm the soundness of the proposed approach for automatic annotation and hence the possibility to create stable resources such as, an emotional corpus that can be employed on supervised machine learning for emotion detection systems.