Gojek Work experience as Data Science Intern

Embarking on a thrilling adventure with Gojek as a Data Science intern, I’ve had the chance to explore the world of real-world data challenges and exciting insights. Looking back on this amazing experience, I can’t wait to share the most important parts of my internship.

I joined the Gofood Data Science team at Gojek, where there were many different projects to choose from. When I started, I had to decide which project I wanted to work on. After thinking it over, I picked a project related to computer vision, which is all about understanding images and how they can help with real business problems.

Being part of this team wasn’t just about picking a project; it was like discovering a path to unlock the secrets hidden in images. I got to work with the Gofood Data Science team, where I worked together wirth my manager to solve problems using data and cool technology.

As I learned and grew, I realized that teamwork and facing challenges were really important. Every time we solved a problem, it felt like adding a piece to a big puzzle. This internship not only made me better at using technical tools but also showed me how important it is to use data to solve real problems.

Being part of the Gofood Data Science journey has opened my eyes to new possibilities and taught me valuable lessons. As I keep moving forward, I’m excited to use what I’ve learned to tackle more challenges and make a positive impact using the power of data and technology.

Merchants use GoBiz to share their content on the GoFood App. This content goes live right away. Since merchants can change their menu information, there’s a chance they might put up the wrong or not suitable content without anyone checking it first.

One problem we want to fix is when a merchant takes a picture from another platform, like a different food delivery app or social media, and claims it’s their own on our app. This is not okay, especially if the picture has a logo from another app or website on it. We don’t want pictures like that on our app.

To solve this, I worked on a way to stop these kinds of pictures from being uploaded. I used different methods and technologies to make sure only the right pictures show up on our app.

Business Objective

Our project aims to create a smart system using data to keep our platform safe. This system will automatically recognize pictures with watermarks and mark them for removal. We want to do more than that – we want to figure out what kind of watermark it is, like from social media, a competitor, or somewhere else. This way, we can take down the right pictures and keep our platform clean.

Technologies

I used SQL because all our information is stored in a big storage place called a database. To get the right information, I used a language called SQL. Besides that, I did most of my work using Python programming language.

Methods

To tackle the issue of inappropriate content in images, whether it’s in text or logo form, I employed a two-step approach. First, I started with a traditional image processing method, and then I moved to a more advanced technique called YOLO (You Only Look Once), which was effective for our needs. To spot text, I utilized the EasyOCR model, a pre-trained system for recognizing characters. Both methods worked successfully in detecting and handling these problems. I have finetuned above both models on my dataset, make it work for our purpose.

Pros:

1. Accuracy: The combination of classical image processing and YOLO for object detection can lead to high accuracy in identifying inappropriate content in images.
2. Comprehensive Detection: Using YOLO allows for the identification of various objects and elements in an image, providing a comprehensive approach to detecting both text and logo forms of inappropriate content.
3. Efficiency: YOLO is known for its speed in object detection, making it efficient for processing a large number of images in real-time or near-real-time scenarios.
4. Pretrained Models: Utilizing pretrained models like EasyOCR reduces the need for building complex detection algorithms from scratch, saving time and effort.
5. Flexibility: The combination of traditional methods and modern deep learning techniques provides flexibility in addressing different types of inappropriate content.

Cons:

1. False Positives/Negatives: While the method is accurate, there might still be instances of false positives (innocent content flagged as inappropriate) or false negatives (inappropriate content not detected), especially in complex scenarios.
2. Training and Tuning: Fine-tuning the YOLO model and optimizing the image processing steps can be time-consuming and require expertise in image processing and machine learning.
3. Resource Intensive: Deep learning models like YOLO can be resource-intensive, demanding substantial computing power for training and inference, potentially affecting system performance.
4. Dependency on Quality Data: The accuracy of the models heavily depends on the quality and diversity of the training data, which can be a challenge to curate and maintain.
5. Adaptation to New Content: The system might need adjustments or retraining when faced with new types of inappropriate content or when the platform evolves.
6. Potential Ethical Considerations: Automated content moderation systems may inadvertently censor legitimate content or fail to identify subtle forms of inappropriate content, raising ethical concerns.

Overall, while the method offers a strong approach to content moderation, it’s essential to continually monitor and fine-tune the system to address any limitations and challenges that arise.

Conclusion

My time as an intern at Gojek has been an incredible journey of growth and transformation. Immersed in real-world projects and surrounded by seasoned experts, I’ve gained a treasure trove of practical skills and insights. Navigating through challenges, I’ve learned the importance of resilience and dedication, while the euphoria of crafting impactful solutions has been a true highlight. This internship has undoubtedly paved the way for a promising future in the field of AI.