Recognition of emotional information is essential in any form of communication. Growing HCI (human-computer interaction) in recent times indicates the importance of understanding of emotions expressed and becomes crucial for improving the system or the interaction itself. In this research work, textual data for emotion recognition is used. The proposal is made for a neural architecture to resolve not less than eight emotions from textual data sources derived from multiple datasets using google pre-trained word2vec word embeddings and a multi-head attention-based bidirectional LSTM model with a one-vs-all multi-level classification. The emotions targeted in this research are anger, disgust, fear, guilt, joy, sadness, shame, and surprise. Textual data from multiple datasets are ingested such as ISEAR, Go Emotions, and Affect dataset. The results show a significant improvement with the modeling architecture with good improvement in recognizing some emotions.
TopIntroduction
Voice Assistants, chatbots, product reviews, or any form of digital conversations that embody users conversing with the digital assistant, include one or the other way of textual communication. Most of the communications surface only if there are appreciations, queries, or concerns. Understanding the emotions of such conversations whether the conversations happening or offline would facilitate tailoring higher solutions on a case-to-case basis. These solutions may be in multiple varied domains and their applications. Conventional methods of machine learning to understand human emotions may hardly help to extract complex, deeper, or indirect emotions, and always appreciate the sequential nature of the language and context, which adds to the complexness, and depth of information. A simple example of this may be, an individual’s brain that would perceive the irony in the earlier sentence, it’s onerous for a machine to understand.
In emotion recognition systems, the restricted variety of words with robust linguistic relations between them needs special attention. This is because of the involvement of both the language novelty and a wide range of feature prospects. Usually, sentiment analysis looks at the polarity of the text in a document or a sentence. Certainly, the focus would get on whether or not the expressed opinion within the text is positive, negative, or neutral. However advanced sentiment analysis techniques check up on the emotional aspects of the text that are proposed in our approach with the emotional indicators such as anger, surprise, sadness, guilt, shame, disgust, fear, and joy.
Contextual emotion detection in textual communication has been seen as gaining popularity and hence its importance. SemEval-2019 Task 3 Chatterjee et al. (2019) introduces a task to detect contextual emotion in conversational text. Emotion recognition from non-verbal communications such as text becomes more complex due to the usage of non-standard languages. To share a few examples, social/microblogging language having large usage of contractions (e.g. I’m gonna bother), elongations (e.g. a vacation tooooooo!), nonstandard use of punctuation (e.g. gonna explain you later...!). Moreover, incorrect spelling (e.g. U r), emoji’s, acronyms, abbreviations, typos, unnecessary spaces, special characters, etc., the text looks more gibberish than more structured if one goes by the book of dictionary. Still, the mass population tenders this data precious considering the volume and usage.
In this paper, the proposal made with a multi-head attention-based deep neural network with bidirectional LSTM for emotion recognition from text.
In this work, it is shown that, with the consideration of the sequential nature of the text and the context, with multi-head attention and bidirectional LSTM model, is better captured, and emotion recognition performance is also have proven to outperform all the models.
TopTextual emotion recognition is relatively one of the unexplored areas of research as the conventional machine learning techniques are quite not sufficient to capture emotions as the sequential nature of the language adds to the complexity.