Sentiment Analysis in the Online Health Community

Bing Wu, Peng Yu2015 · DOI

Abstract

Nowadays, accelerated urbanization and improving living standard have brought some unexpected negative influences, making modern citizens more suffered from chronic diseases. On the other hand, with the innovation of information technology, Web 2.0 is able to provide an accessible and unobstructed channel of communication, which attracts more patients to seek support of health problems through the Internet. Therefore, increasing researches have been focusing on improvement of effective management of online health community. It should been mentioned that community activities reflect the relationship and communication of people, which involves more or less emotional elements. Further, this may influence the management of the entire community. In this research, we utilize social network analysis and sentiment analysis to understand the pattern of communication between patients, in order to provide theoretical and technical support.

Introduction

As chronic diseases like diabetes become increasingly prevalent, patients are turning to "Health 2.0" platforms—online health communities (OHCs)—to share experiences and seek peer support. While previous research has often focused on technical aspects like data extraction or topic identification, there is a critical need to understand the social and emotional dynamics that sustain these communities. Managing an OHC effectively requires more than just data organization; it necessitates an understanding of the emotional patterns of communication. This research bridges the gap between raw text analysis and social structure, investigating how sentiment influences the formation of "collaborative friendships" among patients.

Methodology

The researchers propose a dual-methodological approach to analyze OHCs, combining structural and linguistic analysis.

Social Network Analysis (SNA)

The OHC is modeled as a graph where patients are "actors" (nodes) and their interactions are "relations" (edges). This structural modeling allows for the measurement of network characteristics, such as the centrality of specific individuals and the formation of emotional subgroups.

Sentiment Analysis Pipeline

To quantify human emotions from unstructured text, the study employs a multi-step sentiment analysis pipeline. This process involves identifying subjective sentences, extracting features, and characterizing the emotional polarity and intensity of the content.

Flowchart of the multi-step Sentiment Analysis Pipeline.

The pipeline utilizes two primary tools:

  1. OpinionFinder (OF): Used for initial natural language processing, including part-of-speech (POS) tagging and identifying sentence polarity.
  2. SentiWordNet (SWN): To address the limitations of OpinionFinder regarding emotional intensity, SWN provides a granular measurement of sentiment indices for individual words.

The study tests three core hypotheses: OHC sentiment is generally positive (H1), subgroups share similar emotional intensities (H2), and these networks are highly centralized with few isolated subgroups (H3).

Key Insights and Results

The analysis reveals several critical patterns regarding the emotional health of online communities:

  • The Positivity Bias: Unlike many online spaces, OHCs are inherently positive. Their primary function as support systems results in a high "average sentiment score," which serves as a barometer for community health.
  • Positive Users as Structural Hubs: "Positive" users are not only more pleasant but also more active. They engage with a wider variety of topics and hold more central positions in the social network. These users act as "critical points" that drive the community's ability to provide emotional support.
  • Emotional Homophily: Patients with similar emotional attitudes toward their illness tend to communicate more frequently. This similarity leads to the formation of tight-knit emotional subgroups, suggesting that emotional state is a key factor in relationship formation.

Conclusion and Future Work

The findings suggest that OHC managers should identify and encourage "key users" who exhibit high positive sentiment, as they are vital to community engagement. Furthermore, the discovery of emotional homophily opens possibilities for "friend recommendation" systems based on emotional profiles rather than just shared symptoms.

A significant technical limitation remains contextual ambiguity; while SentiWordNet is robust, it can struggle with sub-word meanings when context is unclear. Future research could explore real-time sentiment monitoring to help managers maintain a healthy community atmosphere and investigate whether encouraging communication between patients with differing emotional states might provide necessary inspiration to those in negative states.

Summary updated

Abstract

Nowadays, accelerated urbanization and improving living standard have brought some unexpected negative influences, making modern citizens more suffered from chronic diseases. On the other hand, with the innovation of information technology, Web 2.0 is able to provide an accessible and unobstructed channel of communication, which attracts more patients to seek support of health problems through the Internet. Therefore, increasing researches have been focusing on improvement of effective management of online health community. It should been mentioned that community activities reflect the relationship and communication of people, which involves more or less emotional elements. Further, this may influence the management of the entire community. In this research, we utilize social network analysis and sentiment analysis to understand the pattern of communication between patients, in order to provide theoretical and technical support.
BibTeX
@misc{wu2015sentiment,
  author = {Bing Wu and Peng Yu},
  title = {{Sentiment Analysis in the Online Health Community}},
  year = {2015},
  doi = {10.2991/ic3me-15.2015.254},
}
Figure