This is a implementation of deep patent landscaping introduced in the paper "Deep Patent landscaping Model Using the Transformer and Graph Embedding"
Among the vast patent data, selecting patents related to specific technology topics is a challenge. This task mainly is done by experts in the patent field and is very inefficient and expensive. The patent landscaping model using deep neural nets can be a clue to solving this problem.
We used search formula from KISTA and converted it to queries for BigQuery. Obtained data is too large, so we sampled data using heuristic method for training. You can download the full dataset and sampled data here.
| Acronyms | Full Name |
|---|---|
| MPUART | Marine Plant Using Augmented Reality Technology |
| 1MWDFS | Technology for 1MW Dual Frequency System |
| MRRG | Technology for Micro Radar Rain Gauge |
| GOCS | Technology for Geostationary Orbit Complex Satellite |
We used Word2Vec for abstract embedding and Diff2Vec for classification code embedding. You can download the pretrained embedding here.
- pretrained word2vec location
- /embeddings/w2v
- pretrained diff2vec location
- /embeddings/graph
Tensorflow==1.x
Keras==2.3.1
pandas==0.25.3
scikit-learn==0.21.3
python train.py --data mrrg