Brandon Cui ([email protected]), Na He Jeon ([email protected]), Matthew Kim ([email protected])
Pinterest deep learning/network analysis project with Professor Jure Leskovec
We only consider a small subset of the boards and pins due to the massiveness of the dataset. We only consider the first 9 million entries in the pins file and the first 100,000 boards when parsing, but only a subset of ~40,000 boards showed up in the first 9 million entries.
We will consider a few experiments for deep-learning
- random sampling
- KNN (based on pre-trained word embeddings)
- LSTM Encoding-Decoding frameworks
Right now the boards are parsed by timesteps. We will re-split the dataset into the appropriate train-test attributes.