Snoonu, the fastest growing technology company in Qatar, was struggling with manual categorization of their marketplace items on the platform. Previously, Snoonu used to classify manually whatever items received from the merchants and upload them manually to the related category. To automate this process, Zero&One created a machine learning (ML) item categorization solution on Amazon Web Services (AWS) platform. This solution improved Snoonu's inventory management and customer experience by enabling precise item categorization. The solution also assisted in content moderation, ensuring a safe user environment by leveraging ML for accurate content categorization.

The Challenge:

Snoonu's previous process of categorizing online items relies heavily on manual intervention, leading to inefficiencies, errors, and inconsistencies in item classification. This not only consumes valuable time but also affects customer experience due to inaccurate categorization.
Also, with the growing volume of online items being added to the platform, managing and categorizing these items manually becomes increasingly challenging. Snoonu requested a solution that can handle a large number of items and adapt to changes in the inventory rapidly.

Moreover, to ensure that the images associated with the items are appropriate, in line with community guidelines, and free from explicit or offensive content is crucial for maintaining a positive brand image and providing a safe user experience.

Why Zero&One and AWS:

Zero&One's collaboration with AWS creates a potent force in the realm of Machine Learning (ML) solutions. With a profound understanding of ML algorithms and techniques, it is well-equipped to tackle complex data challenges and extract valuable insights from large datasets. By partnering with AWS, Zero&One gains access to a reliable and scalable big data infrastructure, which is vital for ML applications that often require substantial computational resources. AWS's cloud-based services provide the necessary compute power and storage capacity to support ML model training, evaluation, and deployment on a vast scale. Furthermore, Zero&One's successful track record in delivering ML-driven projects ensures that organizations can entrust the data analytics initiatives to experienced professionals. The combined expertise of Zero&One and AWS establishes a robust foundation for organizations seeking to leverage the power of ML to drive innovation and data-based decision-making.

The Solution:

Zero&One, has created an innovative solution tailored specifically for Snoonu, utilizing the power of Machine Learning (ML) to automate the process of categorizing and image moderating. The initial step in the solution is data labeling. This is performed using AWS SageMaker Ground Truth, which allows us to create high-quality training datasets quickly and cost-effectively. SageMaker Ground Truth employs a combination of human labelers (via Amazon Mechanical Turk) and machine learning to make the data labeling process more efficient. The data, both images and text, is sourced from an Amazon S3 bucket. The labeled data is then stored back into another S3 bucket for use in the next stages.

The process is then split into two parallel streams, one handling images and the other handling text descriptions. For the image stream, the labeled image data from S3 is used to train a custom image classification model with AWS Rekognition. AWS Rekognition uses pre-trained deep learning models, which can be fine-tuned with labeled data to fit specific classification tasks. Similarly, for the text stream, the labeled text data from S3 is used to train a custom text classification model with AWS Comprehend. AWS Comprehend utilizes pre-trained models that can be further fine-tuned with labeled data for specific classification tasks.

After training the models, they are then hosted on AWS to make them readily available for predictions. This involves deploying the models as endpoints. The services provide a fully-managed solution for deploying the machine learning models, making them accessible via a secure and scalable API. This way, the models can process incoming requests from the application, classify marketplace items based on their images and descriptions, and return the predicted categories and subcategories.

Finally, integrating AWS Rekognition content moderation API empowers the solution to dynamically assess user-uploaded images and text, leveraging advanced machine learning algorithms to detect and categorize potentially objectionable content, thereby enhancing platform safety and content quality.

Results and Benefits:

Through the collaboration with Zero&One, the implementation of AWS ML-based item classification models for image and text analysis, alongside content moderation, yield a range of impactful results and benefits for Snoonu. The integration of these advanced technologies will ensure accurate and consistent item categorization. Real-time content moderation powered by ML will proactively filter out inappropriate content. By automating traditionally manual tasks, Snoonu can optimize its operational efficiency and scale effortlessly as its offerings expand. This transformation translates into a host of benefits, including an enriched user experience that fosters higher satisfaction and conversion rates, resource optimization, streamlined workflows, brand protection, and a competitive edge in the e-commerce arena.