Crowdsourced Audio and Video Data Collection for a Leading Social Media Company
Crowdsourced Audio and Video Data Collection for a Leading Social Media Company
Evaluating machine learning models presents significant opportunities for technology companies to enhance user experience and ensure fairness in their algorithms. With the evolution of technology and social media platforms, having a diverse and balanced dataset is crucial for gaining a competitive edge. A robust and representative training dataset is critical in significantly improving model accuracy.
The Challenge
Our Client, a major social media entity, approached Us to evaluate the fairness and robustness of their machine learning algorithm. Their goal was to create an open-source dataset, which required the collection of an extensive range of audio and video samples from diverse participant groups.
The Answer is Us
Our Client entrusted Us to design and implement a strategic plan for acquiring and managing the required data. We performed the following to ensure consistency and avoid bias in our Client’s machine learning models:
- Utilized our Crowd Recruitment Strategy: We developed a multi-tiered approach to aggregate 13 diverse groups of participants from India, the Philippines, and Vietnam, facilitated through tailored messaging and marketing activities.
- Initiated Crowdsourced Data Collection: We leveraged our TaskVerse freelancers and organized in-person events in India and the Philippines to captivate a wider participant pool.
- Prioritized Data Quality: We executed manual and automated checks to ensure the integrity of the submitted materials.
- Promoted Data Management and Security: We ensured all data adhered to stringent privacy and security protocols.
The Results
Collecting data from a wide range of users within a limited timeframe while maintaining high-quality standards is crucial for the project’s success. Employing dynamic, innovative, and quality-driven strategies led to the following results:
- Achieved diverse representation by covering 13 different demographic groups
- 86.8% participant approval rate
- 11,056 videos vetted and forwarded to the Client, showcasing the project’s extensive scope
- Received over 11,800 sign-ups on TaskVerse through organic and strategic marketing initiatives
Discover our unique operational framework for machine learning models.
Download Case StudyReferences
