Detecting Correlation Between Emotional Transition and Vocabulary Usage in Individual Social Media Posts Using NLP



Summary and goal

  • Emotion modeling of event-based mental states changing using topic modeling
  • Understanding word attribution and vocabulary usages for emotional shifts
  • Discovering discernible changes in writing style depending on emotional shifts in online communication and social media posts
  • Envisioning a conditional random field (CRF) that captures transitions in emotional states to create an empathetic language suggestion model that generates comforting and supportive responses to users

Phase 1 - Analysis of the Correlation Between Emotional Fluctuation and Usage of Words

In today’s digital age, online communities like Reddit play a significant role in providing emotional support and a sense of belonging to individuals dealing with various life situations and struggles. Users often express their feelings, emotions, and experiences on different subreddit communities and are provided with empathetic responses. Harnessing the valuable data generated within these communities is crucial in gaining insight into human emotion.

Our study centrally focuses on exploring the relationship between an individual’s emotional state and writing style on social media platforms, specifically on a conversational platform like Reddit. Writing style is not just limited to one’s choice of words but also includes tone, voice, length of sentences, sentence structure, complexity, and other stylistic elements. However, we mainly focus on vocabulary usage to simplify our experiments in this study. We aim to discover whether emotional state shifts manifest as discernible changes in the usage of vocabulary and content of their online posts.

This paper adopts the term “user” to signify a Reddit participant, and our comprehensive proposal encompasses the following key facets:

  • Emotional state prediction: By analyzing a user’s posts, our objective is to predict their corresponding emotional state, contributing to a nuanced understanding of the emotional landscape within the online community.
  • Vocabulary-emotion relationship: We aspire to establish a robust connection between a user’s choice of vocabulary and the classification of their emotional state, unraveling the nuanced ways in which language reflects and shapes emotional experiences.
  • External event impact: Investigating the impact of external events, ranging from elections and sports events to the pervasive influence of global phenomena like COVID-19, allows us to uncover how these occurrences influence fluctuations in users’ emotional states.
  • Probabilistic emotional transitions of users: Our research aims to unearth the cyclical patterns and transitions in users’ emotional states, providing insights into the temporal dynamics of emotional experiences within the digital realm.

We aim to leverage this research in developing an empathetic AI suggestion model for platforms like Reddit. This model aims to offer supportive responses to users facing diverse challenges, including mental health issues, financial struggles, career, and relationship challenges, fostering a more empathetic and supportive online environment. The overarching objective is to enhance the emotional well-being of users and improve existing empathy systems.

Hypothesis

H1: Connection between word usage and emotional shifts
H2: Impact of external context on emotional shifts

Main findings

Temporal Predominant Topics Analysis

  • Macro-scope temporal analysis like one-year-long or 2-year-long reveals more general and broader topics that may be associated with societal issues. On the other hand, topics from a narrower timeframe are more personal and subjective themes that are related to personal life.
  • Societal issues or events may be deeply connected with the emotional peak of the crowd. Further investigation of events happening in the world is required.

Vocabulary-Emotion Relationship

  • Similar emotions tend to link together, such as annoyance-anger and disappoint-disapproval.
  • Similar emotions also tend to group together, for instance, optimism-love-approval or caring-desire-gratitude.

Probabilistic Emotional Transitions of Users

  • Reddit users with emotions such as fear, excitement, desire tend to remain in the same emotion.
  • There is a general trend of emotional transitions from negative to positive emotions, such as sadness to admiration, love, gratitude, annoyance to admiration, disappointment to love.

Source code

https://github.com/saxenamansi/EmotionAnalysis_RedditUsers