Inspiration
As first-year students, we wanted to explore natural language processing (NLP) and machine learning. Since this was quite new to us, TechJam felt like the perfect opportunity to try things out, learn something new, and gain more exposure to this exciting field.
What it does
Our project takes in online reviews and labels them as relevant or irrelevant. The idea is to help filter out spam, advertisements, and off-topic comments so that review systems can stay useful and reliable for consumers.
How we built it
- Because of the time crunch and limited computational power, we relied on pre-trained models.
- We chose BERT, since it balances performance and resource use better than architechtures like RNNs, which we found too intensive.
- To prepare our dataset, we experimented with one-shot classifiers to generate initial labels before fine-tuning.
Challenges we ran into
- The hardest part was building a clean dataset.
- Our first attempt with keyword-based labeling didn’t work well because context matters a lot in language.
- Finding a way to label data efficiently without losing important details took effort and iteration.
Accomplishments that we're proud of
- Completing our very first hackathon project on time.
- Actually getting a functional NLP pipeline running despite being beginners.
- Working together as a team and learning new technical and collaborative skills.
What we learned
- Starting in a brand-new domain isn’t easy, but it’s doable with persistence.
- Keywords alone aren’t enough for NLP tasks — context is key.
- Open-source APIs and pre-trained models are extremely helpful for getting things off the ground.
- On the teamwork side, we learned a lot about splitting tasks, organizing files, and coordinating better.
What's next for Barbie
- Experiment more with different models and hyperparameters.
- Expand training to larger and more diverse datasets.
- Keep learning more about AI and how it can solve real-world problems.
- As a team, we look forward to continuing our journey in science and AI together.
Log in or sign up for Devpost to join the conversation.