Kaggle's Model Hub Feature Research

Project Type: Generative research and Conceptual testing
Project Length: 12 Months

Project Summary

Kaggle is one of the largest websites where data scientists can access models. However, Kaggle’s support for this is very lacking. We believe that by effectively improving Kaggle’s support for models, it will pay huge dividends in lifting up the entire ML resource ecosystem. An integrated, community-driven ML resource hub would enable those who use models to push forward the state of the art, innovate, and build solutions to do so more efficiently, economically, and responsibly.

By making these changes, Kaggle could become the go-to resource for data scientists who want to use models. This would benefit the entire ML community by making it easier for people to find and use models, and by promoting collaboration and innovation.

My Role

As the lead UX researcher on this project, I was responsible for the following:

Creating a research plan for product design.
Sending out surveys to collect user feedback.
Testing design concepts through rapid prototyping.
Gathering user feedback through interviews.
Identifying usability issues and improvement opportunities.
Advocating for the user in strategy meetings.

My work was essential to the success of this project, and my contributions helped to ensure that the final product was user-friendly and met the needs of the target audience.

The Problem

Kaggle has 14 million users around the world with a diverse set of needs. Some use Kaggle to learn about data science, while others use it to create and fine-tune models. Kaggle is also one of the top places to go for data science competitions.

To better understand their users’ needs, Kaggle conducted research on their current workflows and pain points. Kaggle found that many users struggle to find the right pre-trained model for their task. We also found that ML researchers were not publishing or interacting with Kaggle as much as we would like them to do.

Kaggle is using this research to improve the way pre-trained models are presented to users. Kaggle is committed to providing the best possible experience for their users. We believe that the new redesign will help them achieve this goal.

The Solution

The purpose of this feature is to increase retention among ML researchers. This feature enables ML researchers to upload their research and models they have trained. The other purpose of this feature is to bring more “accurate” models to data scientist practitioners to fine tune and make more accurate. The text is about the process of developing a feature for a machine learning platform. The feature allows ML researchers to upload their research and models, and data scientist practitioners to fine tune and make more accurate models. The feature was developed based on research, interviews, and feedback from users.

The research showed that ML researchers focus on advancing the machine learning field and need access to benchmarking and open sourced datasets. The interviews showed that ML researchers and data scientist practitioners have difficulty finding and using accurate models. The feedback showed that users want a feature that allows them to upload and share their models.

The feature was developed in an iterative process, with multiple rounds of research, interviews, feedback, and development. The feature was released and is currently being used by ML researchers and data scientist practitioners.

The Results

20% of users that view the list of models, click into the detailed page of a specific model. We expect that number to continue to increase over the next 6 months. Of that 20% that are looking at the details of the model, less than 1% are downloading these pre-trained models. We believe this is due to not having all models accessible and not allowing the community to publish their own models.

The model hub did a quick pivot a week before release. We decided not to open the model hub to the community right away because there were some concerns about the potential for NFWS and other vulgar images to be uploaded. The publishing flow has been tested and validated to ensure that it is safe and reliable before opening it up to the community. It is expected to be released by the end of the year.