A Division Delta project
Studying online has turned education into a solitary experience. As a team of soon-to-be University students, we understand the value of study groups and collaboration. Thus, we were inspired to created Peerify – a game-based learning platform that turns study sessions into an exciting social experience.
What it does
We are all too familiar with grinding through AP FRQs and writing prompts with our classmates. The process of peer review and collaboration often brings out the best ideas from all parties, and makes a strenuous task enjoyable.
Our website facilitates this process online by allowing students to join online lobbies, create prompts, and review each other's responses in a style akin to online party games. Rooms are created with a unique id, which can be easily shared so people can join. Students have the freedom to ask any question, ranging from writing prompts to chemistry or geography problems. Communication is encouraged each round, and a point system is used to reward those with the best answers.
Following each round, we use machine learning to generate insights that allow students make the most of their experience. The feedback is analyzed and key words and topics are displayed as a word map, highlighting the subject areas that should be reviewed. This data can help educators plan future lessons and contribute to our dataset which guides future games.
How we built it
We built the frontend using React and used Node.js for the backend, using websockets to interact between the two. We hosted our backend using Digital Ocean, so we could each join a game from each of our computers.
We created a model using gensim's Word2Vec with the data of 300000 Amazon reviews. Based on these word vectors we were able to train a random forest classifier through sklearn and predict whether certain words had a stronger correlation with low ratings or high ratings. We started with this base dataset but as more rounds are played, the model can be updated every once in a while (let's say, every 500 user reviews).
Challenges we ran into
We had trouble initially implementing python machine learning script into our Node.js web app, but we figured it out in the end. We also went through quite a few iterations of our UI, but we are quite happy with the end result.
Accomplishments that we're proud of
We were proud of creating a project that not only had a robust website, but also integrated data science and machine learning to offer a unique solution to students. Being able to create a project that solves a problem that we often encounter as students was also incredibly fulfilling.
What we learned
We all learned a lot about web development and the theme of this hackathon also challenged us to build a machine learning model which was also integrated to the backend.
What's next for Peerify
As this project scales, we hope to continuously improve our AI model by using relevant data from users that interact with our website. We also hope to improve the user flow and round system depending on future user feedback.