PaperlayerView exact revision on arXiv

Source-linked AI summary

Engagement Patterns of Peer-to-Peer Interactions on Mental Health Platforms

Ashish Sharma, Monojit Choudhury, Tim Althoff, Amit Sharma

arXiv:2004.04999v1cs.SI

TL;DR

Online peer-support platforms offer mental health support, but their engagement patterns remain insufficiently understood and user interaction is essential. This study models four indicators across TalkLife and Reddit to derive 11 patterns, finding mutual discourse associated with higher seeker retention on TalkLife.

  • Problem

    Engagement on mental health support platforms is not well understood, although users need to interact for peer support to work.

    Source: §1 · pp. 1–2

  • Method

    The study jointly models four attention- and interaction-based indicators across TalkLife and Reddit to derive 11 interpretable engagement patterns.

    Source: §1 · p. 2, §8 · p. 10

  • Results

    Mutual discourse is more important for seeker retention than other engagement indicators on TalkLife, and is associated with early seeker replies and negative sentiment.

    Source: §1 · p. 2, §8 · p. 10

  • Takeaways & Limitations

    The engagement-pattern framework supports informative self-evaluation and comparative assessment of online support platforms.

    Source: §8 · p. 10

    Interventions encouraging early seeker replies may carry unintended risks because mutual discourse often involves negative sentiment.

    Source: §8 · p. 10

Abstract

from arXiv · show

Mental illness is a global health problem, but access to mental healthcare resources remain poor worldwide. Online peer-to-peer support platforms attempt to alleviate this fundamental gap by enabling those who struggle with mental illness to provide and receive social support from their peers. However, successful social support requires users to engage with each other and failures may have serious consequences for users in need. Our understanding of engagement patterns on mental health platforms is limited but critical to inform the role, limitations, and design of these platforms. Here, we present a large-scale analysis of engagement patterns of 35 million posts on two popular online mental health platforms, TalkLife and Reddit. Leveraging communication models in human-computer interaction and communication theory, we operationalize a set of four engagement indicators based on attention and interaction. We then propose a generative model to jointly model these indicators of engagement, the output of which is synthesized into a novel set of eleven distinct, interpretable patterns. We demonstrate that this framework of engagement patterns enables informative evaluations and analysis of online support platforms. Specifically, we find that mutual back-and-forth interactions are associated with significantly higher user retention rates on TalkLife. Such back-and-forth interactions, in turn, are associated with early response times and the sentiment of posts.

1 Introduction

Limited access to mental healthcare motivates online peer support, but support depends on users engaging with one another. This study analyzes thread-level engagement at scale and finds that mutual discourse is associated with higher seeker retention on TalkLife.

1 Introduction

  • The study examines engagement because access to mental healthcare remains poor worldwide despite mental illness contributing 13% of the global disease burden.Most low- and middle-income countries have fewer than one psychiatrist per 100,000 people; 60% of US counties have none.Source: §1 · p. 1
  • Online support platforms enable asynchronous, geographically unrestricted, and anonymous peer conversations that may help address stigma around mental illness.Threads begin with a support-seeking post and may continue through replies and back-and-forth interaction.Source: §1 · p. 1
  • One third of TalkLife support-seeking posts receive no responses, while limited engagement can have serious consequences for vulnerable users.Prior literature links engagement with favorable outcomes including empathy, effective distraction, and overcoming cognitive distortion.Source: §1 · p. 1
  • 35 million posts across TalkLife and Reddit are analyzed to identify 11 interpretable thread-level engagement patterns from jointly modeled indicators.The framework uses attention and interaction indicators because no single indicator captures engagement dynamics.Source: §1 · p. 2
  • Prior studies examined engagement dimensions separately or used small annotated datasets and characterizations that overlooked interaction between users.The paper motivates a joint, thread-level analysis rather than examining indicators such as post and like counts independently.Source: §1 · pp. 1–2
  • On TalkLife, mutual discourse is more important for seeker retention than the other engagement indicators, and the framework supports platform self-evaluations and comparisons.The paper also reports that comparing TalkLife and Reddit highlights how design differences relate to engagement dynamics.Source: §1 · p. 2

2 Related Work

Prior engagement research often used context-specific indicators, while this work adds interaction-based measures and models them jointly to identify interpretable engagement patterns in online support conversations.

2 Related Work

  • The related work spans engagement studies, mental-illness and online-support research, and statistical models of conversational threads.Prior thread models examined growth, structure, user re-entry, and latent user roles.Source: §2 · p. 2, §2.3 · p. 3

2.1 Engagement patterns in online platforms

  • Prior engagement studies used varied context-specific markers, while this work introduces interaction-based indicators and jointly explores them with other engagement dimensions.The paper describes one interaction indicator, degree of interaction, as novel in online social support contexts.Source: §2.1 · p. 2

2.2 Mental illness & online social support

  • The paper develops a framework for discovering patterns of effective online mental health support conversations and evaluating and comparing platforms.The framework focuses on engagement rather than diagnosing mental illness or classifying support types.Source: §2.2 · pp. 2–3

2.3 Modeling of a conversational thread

  • The study jointly models multiple engagement indicators to discover meaningful engagement-pattern clusters with an unsupervised generative model.Its focus is thread-level engagement in online mental health support conversations.Source: §2.3 · p. 3, §2.2 · p. 3

3 Dataset Description

The study uses conversational threads from TalkLife and mental-health-focused Reddit communities, relying primarily on post text, timestamps, and author IDs while minimizing platform-specific metadata.

3 Dataset Description

  • TalkLife threads begin with a user's support-seeking post and may receive peer responses that lead to back-and-forth conversation.The initial poster is the seeker and responding users are peer-supporters within that thread; either role may differ across a user's threads.Source: §3 · p. 3
  • The Reddit dataset covers threads from 55 mental-health-focused subreddits archived on Google BigQuery from 2015 to 2019.The subreddit list was compiled by Sharma et al. (2018).Source: §3 · p. 3
  • TalkLife is a mental-health peer-support platform, whereas Reddit uses topic-focused subcommunities and less than 0.1% of its activity concerns mental health.Both platforms rely on volunteer peers, usually untrained, and rarely on professionals.Source: §3 · p. 3
  • The analysis uses thread posts, timestamps, and author IDs while minimizing platform-specific metadata to encourage methods and findings that could generalize.TalkLife data was obtained with license and consent, personally identifiable information was removed, and the work received institutional review approval.Source: §3 · p. 3

4 Indicators of Thread-Level Engagement

The paper treats thread-level engagement as multidimensional, separating attention received from interaction between seekers and peer-supporters, and operationalizes both with quantitative indicators.

4 Indicators of Thread-Level Engagement

  • The indicators are intended to provide a complementary joint view of engagement without analyzing the content of sensitive individual posts.The paper notes that engagement and helpfulness are not interchangeable and calls for future work on their relationship to outcomes.Source: §4 · p. 4
  • The framework distinguishes attention received by a thread from the nature of interaction between seekers and peer-supporters.It focuses on engagement patterns rather than outcomes, and a more engaging thread is not necessarily more helpful.Source: §4 · p. 4

4.1 Attention-Based Indicators

  • Thread length and number of peer-supporters measure attention, distinguishing short versus long threads and isolated, two-party, or multi-party conversations.The generative model determines short and long thresholds rather than requiring manually chosen cutoffs.Source: §4.1 · p. 4

4.2 Interaction-Based Indicators

  • Response timing distinguishes quick from slow threads, while degree of interaction distinguishes single interaction, repeated seeker interaction, and mutual discourse.Mutual discourse occurs when a peer-supporter responds again after the seeker replies.Source: §4.2 · pp. 4–5

5 Modeling Thread-Level Engagement

The model jointly analyzes thread length, response timing, and user roles to identify interpretable engagement patterns. It uses cluster-specific distributions and parameter inference to learn these patterns from threads.

5 Modeling Thread-Level Engagement

  • Jointly modeling engagement indicators captures thread dynamics that separate analyses can miss, while avoiding manual thresholds and implausible combinations.The framework represents thread length, time between responses, supporter count, and interaction degree together.Source: §5 · p. 5

5.1 Modeling Assumptions

  • Threads are assigned to a single cluster, with interaction-based roles distinguishing first, new, and existing peer-supporters from seekers.The four roles help represent both supporter count and interaction degree; consecutive posts by the same user are combined in preprocessing, which may lose temporal effects outside the study’s scope.Source: §5.1 · p. 5

5.2 Generative Process

  • Each cluster generates thread length, reply roles, and response delays using cluster-specific distributions.Thread length and time between responses are modeled with Beta distributions; roles are modeled categorically with Dirichlet-distributed parameters, and isolated threads are identified separately rather than used in learning.Source: §5.2 · p. 6, §5.1 · p. 6

5.3 Parameter Inference

  • Gibbs sampling estimates role distributions, method of moments estimates Beta parameters, and the Elbow method selects 20 clusters for each platform.The authors report that 20 clusters also yielded the most diverse and interpretable clusters.Source: §5.3 · p. 6

6 Inferred Engagement Patterns

The learned clusters yield hierarchically organized engagement patterns across TalkLife and Reddit, spanning isolated threads, single interactions, repeated seeker interactions, and mutual discourse. Their role and timing profiles vary across patterns and platforms.

6 Inferred Engagement Patterns

  • Isolated threads make up 32.43% of TalkLife and 27.53% of Reddit, while Single Interaction accounts for 30.57% and 7.64%, respectively.Single Interaction includes Short Slow Two-Party threads (20.30% and 0.08%) and Short Slow Multi-Party threads (10.27% and 7.56%).Source: §6 · p. 7
  • Repeated Seeker Interaction accounts for 18.6% of TalkLife and 21.4% of Reddit, with Multi-Party threads comprising 14.35% and 15.82%, respectively.The Two-Party totals are 4.25% and 5.58%; subpatterns include Short Slow Two-Party (3.39% and 3.99%), Short Quick Two-Party (0.86% and 1.59%), Short Slow Multi-Party (1.10% and 12.96%), and Short Quick Multi-Party (13.25% and 2.86%).Source: §6 · p. 7
  • Mutual Discourse accounts for 18.4% of TalkLife and 43.43% of Reddit threads; its Two-Party and Multi-Party shares are 8.86% and 9.54% on TalkLife, versus 22.08% and 21.35% on Reddit.The listed subpatterns include Short Quick Two-Party (8.11% and 21.99%), Long Quick Two-Party (0.75% and 0.09%), Short Quick Multi-Party (6.17% and 17.33%), and Long Quick Multi-Party (3.37% and 4.02%).Source: §6 · p. 7

6.1 Qualitative evaluation of inferred patterns

  • Role profiles distinguish the patterns: Two-Party threads have more first peer-supporters, Multi-Party threads more new peer-supporters, and Mutual Discourse more seekers and returning peer-supporters.Single Interaction patterns rarely contain seeker or existing peer-supporter responses.Source: §6.1 · p. 7
  • Seekers are more likely to return after first threads with higher engagement, and least likely after an Isolated first thread.The caption emphasizes variation in retention likelihood across engagement degrees rather than absolute values; error bars are bootstrapped 95% confidence intervals.Source: Figure 4 · p. 7
  • Short and Long thread lengths and Slow and Quick reply delays differ across TalkLife and Reddit.TalkLife Short and Long threads have mean lengths of 3.9 and 13.5; Reddit’s are 3.33 and 23.95. Median reply times for Slow and Quick threads are 7 and 1 minutes on TalkLife, and 75 and 16 minutes on Reddit.Source: §6.1 · p. 7

7 Implications of Engagement Patterns

The engagement framework is used to compare TalkLife and Reddit, examine how engagement patterns relate to seeker and peer-supporter retention, and investigate factors associated with mutual discourse. Mutual discourse is associated with higher seeker retention on TalkLife, while early seeker replies and response content are associated with its occurrence.

7.1 Comparative assessment of TALKLIFE & REDDIT

  • Reddit has more Mutual Discourse than TalkLife (43.43% vs. 18.4%) and very few Two-Party Single Interaction threads (0.08%).The authors suggest Reddit’s topic-based subcommunities may contribute to these differences and call for future study of community structure dynamics.Source: §7.1 · p. 8

7.2 Seeker Retention on support platforms

  • On TalkLife, seeker retention is highest after Mutual Discourse (0.75), compared with Repeated Seeker Interaction (0.71) and Single Interaction (0.66).Across joint patterns, retention varies mainly by degree of interaction, with limited variation among other indicators within each degree.Source: §7.2 · p. 8, Figure 5 · p. 8

7.3 Peer-Supporter Retention

  • Peer-supporter retention is higher in Two-Party than Multi-Party threads on TalkLife (0.81 vs. 0.79), especially in Long, Quick Mutual Discourse threads on both platforms.For Long Quick Mutual Discourse, retention is 0.95 vs. 0.84 on TalkLife and 0.84 vs. 0.62 on Reddit; in Repeated Seeker Interaction threads, Slow threads have higher peer-supporter retention than Quick threads.Source: §7.3 · p. 9, Figure 6 · p. 9
  • Among first-time peer-supporters, a slow first response is associated with higher retention likelihood than a quick first response.The authors suggest possible explanations include taking time to select threads, write, or learn the platform interface, and identify disentangling them as future work.Source: §7.3 · p. 9

7.4 Engaging in Mutual Discourse

  • Mutual Discourse is associated with more negative and less positive seeker responses, which average 17.16 words versus 13.89 in Repeated Seeker Interaction.The sentiment differences are reported for both platforms: average sentiment is 0.096 vs. 0.074 on TalkLife and 0.078 vs. 0.064 on Reddit; positive sentiment is 0.152 vs. 0.222 and 0.133 vs. 0.171, respectively.Source: §7.4 · p. 10
  • Threads are more likely to become Mutual Discourse when seekers reply immediately after the first peer-supporter (75.22% on TalkLife; 53.51% on Reddit).The likelihood tends to decrease when more peer-supporters reply before the seeker responds.Source: §7.4 · p. 10

8 Conclusion

Mutual Discourse is critical for seeker retention, particularly on TalkLife, while comparisons suggest topically focused sub-communities may make support platforms more engaging. The authors caution that the findings are correlational and that interventions involving early replies may carry risks.

8 Conclusion

  • Mutual Discourse was critical for seeker retention, particularly on TalkLife, and facilitating it is an immediate future research direction.Source: §8 · p. 10
  • Comparisons suggested that topically focused sub-communities, as found on Reddit, may help make online support platforms more engaging.Source: §8 · p. 10
  • The study is correlational and cannot establish causal effects; its framework also excludes post content, limiting use of theories that depend on prior post content.Source: §8 · p. 10
  • Mutual Discourse often involves negative-sentiment responses, so interventions encouraging early seeker replies may risk negative effects for vulnerable users.Source: §8 · p. 10
Loading 2004.04999v1…