Abstract
Online health forums are an important avenue for receiving support and learning about fellow patients' experiences with similar diagnoses. We seek to characterize the external information shared (via web links) on health forums as a proxy to participants' information needs. For this purpose, using a dataset of web links shared publicly on a lung cancer forum over a period of 16 years, we perform a comparative analysis with three different website typologies, uncovering a diverse ecosystem of websites. We also examine typological variations as this forum gains and then loses popularity over time.
Introduction
When facing a serious diagnosis like lung cancer, patients and caregivers engage in "collective sensemaking," bringing in external information—such as news articles, research papers, and government resources—to build a shared knowledge base. While the role of these forums in providing emotional support is well-documented, there is a gap in understanding the external resources users rely on over the long term. This study addresses two primary questions: what types of information needs are met through link sharing, and how do these needs evolve over the lifecycle of a forum from growth to decline?
Methods
To categorize the diverse array of shared web links, the researchers employed a multi-pronged approach:
- Link Extraction & Archiving: URLs were extracted from forum posts and archived using the Internet Archive’s Wayback Machine to preserve the context of the information as it appeared when shared.
- Triple Typology: Three frameworks were used for classification: the Open Directory Project (ODP) for general web classification, the Interactive Advertising Bureau (IAB) standard, and a custom Domain Specific (DS) typology tailored for health contexts (e.g., Medical Research, Health News).
- Quantitative Metrics: Topical diversity was measured using Entropy, while the concentration of attention was measured using the Gini Coefficient.
Dataset
The study analyzed a longitudinal dataset from a leading lung cancer nonprofit's public forum spanning 16 years (January 2003 to November 2018). The dataset included 369,284 posts across 38,209 threads, containing 28,302 total URLs from 4,645 unique domains. Links were found in approximately 25% of all threads, though they appeared in only about 3% of individual posts.
Key Results and Insights
The Health-Centric Core
Analysis revealed that 51% to 64% of shared links were health-related. The remaining 35-45% comprised a "long tail" of lifestyle content, personal blogs, and mainstream news. Notably, approximately 6% of links were "self-references" to other forum discussions, indicating that the community utilizes its own history as a knowledge repository.
The Shrinking Information Ecosystem
As the forum passed its peak popularity, the variety of shared information narrowed significantly. The community converged on a smaller set of established resources, while ancillary topics like government resources or general news became less prominent. This is evidenced by a steady decline in cross-topical entropy over the 16-year period.
Information Inequality and Density
Over time, the Gini coefficient increased, indicating that a few "super-domains" (such as major medical portals or the forum's own hosting site) captured a larger share of attention. Interestingly, while the total volume of posts decreased in later years, the density of links actually tripled, rising from 2.6% of posts pre-2011 to 8.8% post-2011. Under the custom DS typology, "Health News & Info" and "Medical Research" remained the most shared categories, while mainstream news (13.8%) and government/NGO sources (6%) played smaller roles.
Limitations
The study's focus on a single lung cancer forum may limit the generalizability of the findings to other health conditions. Furthermore, while the study observes a narrowing of the information ecosystem, the underlying cause—whether due to community burnout, information saturation, or external search engine algorithms—remains to be determined. Future research should compare these patterns across different patient populations, such as those managing chronic versus acute conditions.
This research was originally presented at ICWSM 2020.
Summary updated