Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NepEMO: A Multi-Label Emotion and Sentiment Analysis on Nepali Reddit with Linguistic Insights and Temporal Trends

Sameer Sitoula, Tej Bahadur Shahi, Laxmi Prasad Bhatt, Anisha Pokhrel, Arjun Neupane

TL;DR

The paper introduces NepEMO, a Nepali Reddit dataset annotated for multi-label emotions (fear, anger, sadness, joy, depression) and three sentiment classes (positive, negative, neutral), spanning 2019–2025 in English, Roman Nepali, and Devanagari scripts. It combines linguistic analyses (TF-IDF, n-grams, co-occurrence) and topic modeling (LDA) with a broad benchmark of ML, DL, and transformer models, showing that multilingual transformers (e.g., XLM-RoBERTa) consistently outperform baselines. High inter-annotator agreement confirms annotation reliability, and temporal analyses reveal meaningful trends in mental health discourse on Nepali Reddit. The work demonstrates the feasibility and value of monitoring mental health signals in low-resource, code-mixed languages and points to future expansion to more emotions and lighter models for practical deployment.

Abstract

Social media (SM) platforms (e.g. Facebook, Twitter, and Reddit) are increasingly leveraged to share opinions and emotions, specifically during challenging events, such as natural disasters, pandemics, and political elections, and joyful occasions like festivals and celebrations. Among the SM platforms, Reddit provides a unique space for its users to anonymously express their experiences and thoughts on sensitive issues such as health and daily life. In this work, we present a novel dataset, called NepEMO, for multi-label emotion (MLE) and sentiment classification (SC) on the Nepali subreddit post. We curate and build a manually annotated dataset of 4,462 posts (January 2019- June 2025) written in English, Romanised Nepali and Devanagari script for five emotions (fear, anger, sadness, joy, and depression) and three sentiment classes (positive, negative, and neutral). We perform a detailed analysis of posts to capture linguistic insights, including emotion trends, co-occurrence of emotions, sentiment-specific n-grams, and topic modelling using Latent Dirichlet Allocation and TF-IDF keyword extraction. Finally, we compare various traditional machine learning (ML), deep learning (DL), and transformer models for MLE and SC tasks. The result shows that transformer models consistently outperform the ML and DL models for both tasks.

NepEMO: A Multi-Label Emotion and Sentiment Analysis on Nepali Reddit with Linguistic Insights and Temporal Trends

TL;DR

The paper introduces NepEMO, a Nepali Reddit dataset annotated for multi-label emotions (fear, anger, sadness, joy, depression) and three sentiment classes (positive, negative, neutral), spanning 2019–2025 in English, Roman Nepali, and Devanagari scripts. It combines linguistic analyses (TF-IDF, n-grams, co-occurrence) and topic modeling (LDA) with a broad benchmark of ML, DL, and transformer models, showing that multilingual transformers (e.g., XLM-RoBERTa) consistently outperform baselines. High inter-annotator agreement confirms annotation reliability, and temporal analyses reveal meaningful trends in mental health discourse on Nepali Reddit. The work demonstrates the feasibility and value of monitoring mental health signals in low-resource, code-mixed languages and points to future expansion to more emotions and lighter models for practical deployment.

Abstract

Social media (SM) platforms (e.g. Facebook, Twitter, and Reddit) are increasingly leveraged to share opinions and emotions, specifically during challenging events, such as natural disasters, pandemics, and political elections, and joyful occasions like festivals and celebrations. Among the SM platforms, Reddit provides a unique space for its users to anonymously express their experiences and thoughts on sensitive issues such as health and daily life. In this work, we present a novel dataset, called NepEMO, for multi-label emotion (MLE) and sentiment classification (SC) on the Nepali subreddit post. We curate and build a manually annotated dataset of 4,462 posts (January 2019- June 2025) written in English, Romanised Nepali and Devanagari script for five emotions (fear, anger, sadness, joy, and depression) and three sentiment classes (positive, negative, and neutral). We perform a detailed analysis of posts to capture linguistic insights, including emotion trends, co-occurrence of emotions, sentiment-specific n-grams, and topic modelling using Latent Dirichlet Allocation and TF-IDF keyword extraction. Finally, we compare various traditional machine learning (ML), deep learning (DL), and transformer models for MLE and SC tasks. The result shows that transformer models consistently outperform the ML and DL models for both tasks.
Paper Structure (28 sections, 20 equations, 8 figures, 15 tables)

This paper contains 28 sections, 20 equations, 8 figures, 15 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Data collection and annotation
  4. Data collection
  5. Annotation process
  6. Three-phase annotation
  7. Annotation guidelines
  8. Inter-annotator agreement and statistics
  9. Data analysis
  10. Dataset statistics
  11. Language distribution
  12. Word frequency distribution
  13. Unsupervised topic modelling and theme extraction
  14. Co-occurrence of emotions
  15. Temporal dynamics of emotions
  16. ...and 13 more sections

Figures (8)

  • Figure 1: Architecture of our methodology, including data collection, annotation, preprocessing, and analysis
  • Figure 2: Language distribution of the dataset: English vs Code-Mixed posts.
  • Figure 3: Number of topics compared to coherence score
  • Figure 4: Rejected topic modelling due to overlapping(no.of topics = 6)
  • Figure 5: Rejected topic modelling due to overlapping(no.of topics = 5)
  • ...and 3 more figures