Leading Social Media Platform Improves Content Relevance with Personalization
The company partnered with us thanks to our geographically and demographically diverse pool of raters, our ability ramp global operations quickly, and meet strict quality and data flow requirements
A leading social media platform working to keep on top of user demands for more relevant content needs a partner that can provide a geographically and demographically diverse pool of raters, ramp global operations quickly, and meet strict data flow requirements all while working within a complex quality system.
With the increasing demand from consumers for more relevant content—be it through social media, search engines or other websites—any site providing news feeds should develop a strategy for personalization. The algorithms behind the delivery of news content need to be trained to show users more of the content they want, and less of the content they don’t want. This requires collecting data from users that represent the profile of the site’s actual users—through their individual accounts—and feeding that data into the machine learning training model.
In this case, our client decided to conduct an initial pilot that required a large-scale ramp in which participants would provide very personalized input, with strict data flow requirements and a complex quality system. The company already had a similar project underway with another vendor, but the setup didn’t allow for geographic expansion, nor did it provide the demographic diversity that was required to adequately improve the company’s algorithm so that it accurately represented its user base.
The pilot started with 500 participants contracted for a period of four weeks. We developed a strong, scalable onboarding module with visual and interactive components, and in a matter of weeks, the client was receiving the data needed to improve its algorithm 24 hours a day, seven days a week. Participants rated each news item for a variety of elements, including the importance and impact of the content. This process guided users in determining an overall rating score for each item, which allowed the training model to fine-tune the algorithm to provide a more personalized experience for all its users.
While the client initially believed this would be a short-term pilot to fill in gaps, the flexibility, scalability, speed, and high quality of the data we provided, together with a strong partnership formed in those first weeks, allowed this project to go from a short pilot to an ongoing program. Since then, the pilot has expanded to new markets and additional experiments have been implemented successfully.
The geographic and demographic diversity of the rater pool proved immensely valuable to the training model since it mapped closely to the client’s existing users, allowing the client to deliver much more personalized content than it had in the past. Additionally, with our on-demand crowd model, the provider continues to receive a steady flow of data, even on holidays and weekends, allowing it to refine its algorithm much more quickly and consistently than it could in the past. Now that the client has a proven program for improving the personalization of its news feed, it can apply a similar process to addressing other areas such as reducing news SPAM for its users.
The client had strict quality thresholds that needed to be implemented and maintained throughout this project. The project also involved a subjective and complex task, and required minimal human QA. As a result, close collaboration was critical to the project’s success. Our team partnered with the client to develop task guidelines and quality management plans, and quickly ramped the number of participants needed to meet daily and weekly data demands.
Frequent changes to the requirements for this project meant that our project management team needed to quickly update complex processes. The team’s agility and effectiveness in responding to these changes were key success factors for this project.