Abstract
In order to detect and recognize emotions an analysis must be done of human expressions
present in multimodal forms like texts, audio, or video. In this research, we plan to, provide a
solution for a content-based classification problem that is detecting and recognizing emotions
from textual data. Detailed research would also be done that how keystroke metadata can
detect emotion from the speed and harshness of keystrokes while typing a text. Our aim is to
develop a model that performs significantly well than baseline methods.
The context of dialogue can prove helpful in the effective detection of human emotions. The analysis gives us a
the promising result shows a substantial number of emotional states detected from textual data.
Textual data would be classified into six generic types of emotions given by human
psychology.
Data collection is subject to user permission and is collected from the social
media websites a user has on his mobile device, like Twitter. Pre-processing of textual data
includes removing stop words, lemmatization, spell checking and identifying and translating
data present in Roman Urdu and then tokenizing the data in order to further generate word
and sentence embedding.
The emotion would then be detected by rule-based, feature-based
and embedding-based models. By the end, a working, non-interrupting, and simple android
application would be developed to work as an interface for our work