Understanding and Modeling Viewers' First Impressions with Images in Online Medical Crowdfunding Campaigns

Title of the Paper

Understanding and Modeling Viewers’ First Impressions with Images in Online Medical Crowdfunding Campaigns

Paper Information

• Research Area: Artificial Intelligence, Computer Vision, Social Computing
• Keywords: Online Medical Crowdfunding, First Impressions, Computational Modeling, Image Features, Donation Behavior

Research Background and Problem

• Problems and Challenges:

  1. Online medical crowdfunding platforms (OMCPs) serve as tools to provide financial support for patients, but viewers’ first impressions significantly influence donation decisions.
  2. Images play a crucial role in shaping first impressions, yet fundraisers often lack the expertise to select appropriate images, resulting in images that fail to effectively convey key emotional and reliability-related information.
  3. There is currently a lack of effective tools to help fundraisers evaluate the visual impact of images, as well as a lack of clear practical guidelines.

• Significance:

  • A successful first impression can enhance viewers’ empathy, trust, sense of justice, influence, and attractiveness, all of which are closely linked to donation intentions.
  • Understanding how images impact crowdfunding success can provide valuable insights for platform design and fundraising practices.

• Related Work:

  • First impressions have been studied in domains such as crowdfunding and tourism to explore behavioral decision-making.
  • Literature indicates that image content (e.g., gender, emotional expression) correlates with first impressions and performance, but the effects and predictive capabilities of visual elements alone have not been fully explored.

Proposed Solution

• Method Overview:

  • Propose a data-driven computational approach to assess whether images in online medical crowdfunding effectively convey appropriate first impressions.
  • Collect crowdfunding image data, design experiments to gather viewers’ perception ratings of images, and link visual features to first impressions and donation intentions.

• Innovations:

  1. Systematically define five key first impression dimensions related to donations: Empathy, Credibility, Justice, Impact, and Attractiveness.
  2. Systematically analyze the impact of visual features on these dimensions.
  3. Propose modeling first impressions as a multi-class classification problem, leveraging machine learning and deep learning models to evaluate predictive performance.

• Implementation Steps:

  • Dataset Construction:

    1. Collect 7,039 crowdfunding images from the GoFundMe platform, extract cover images, and conduct thematic classification analysis.
    2. Select 450 images for experimental use.

  • Experiment Design:

    1. Use a crowdsourcing platform to collect participants’ first impression ratings for each image.
    2. Employ five different participants to rate each image and ensure data quality.

  • Feature Analysis:

    • Extract visual features (content features, color features, texture features, compositional features) and analyze their correlation with perception ratings.

  • Classification and Prediction:

    • Train models using algorithms such as KNN, SVM, and Random Forest to predict whether an image exceeds the “average level” in a specific impression dimension.

Research Findings

• Experimental Results:

  1. Consistency in Human Perception:

     • Participants consistently formed first impression ratings based on images (reliability above 0.6).
     • Perception scores across the five key dimensions showed significant positive correlations with donation intentions.

  2. Image Feature Analysis:

     • Image content (e.g., age, number of people, emotions) significantly influenced first impression evaluations.
     • Color and compositional features (e.g., brightness, contrast) were closely associated with dimensions such as empathy and attractiveness.

  3. Modeling Performance:

     • The proposed classification algorithms achieved an F1 score of 0.727 in predicting first impressions.
     • Random Forest performed best across all evaluation dimensions, particularly excelling in predicting empathy, credibility, and attractiveness.

• Advantages:

  1. Provides a practical method to help fundraisers evaluate the potential impact of images in advance.
  2. Reveals the significant influence of visual features on human perception, with potential applications across various domains.

• Limitations and Future Directions:

  1. Cultural Differences: The study is based solely on data from U.S. viewers; future research should explore cross-cultural differences in first impression perceptions.
  2. Dataset Size: The current dataset is limited; future work could expand to larger image samples.
  3. Interaction Effects: The study primarily analyzes the effects of individual features; future research should focus on relationships and complex interactions between features.
  4. Practical Tool Development: Propose the development of user feedback systems for OMCPs to guide fundraisers in selecting images.

Conclusion

• The authors conducted an in-depth study and modeling of first impressions of images in online medical crowdfunding, systematically exploring for the first time how visual elements influence first impressions and potential donation outcomes. The findings provide theoretical foundations and practical references for platform optimization and image selection guidance.