This repository contains experiments and proofs of concept for Edoardo Riggio's Master's thesis "API Scout: An Information Retrieval System for OpenAPI Specifications."
.
├── .run
├── data
├── notebooks
├── out
│ ├── latex
│ ├── models
│ │ ├── universal-sentence-encoder
│ │ └── doc2vec.model
│ ├── pdfs
│ └── plots
└── proposal
For the vectorization of the specifications,
I've experimented with both the doc2vec model by gensim (trained on my data),
and with the Universal Sentence Encoder model by Google
(only used to vectorize the documents, no training was necessary).
The notebooks elasticsearch.ipynb and migration.ipynb require the "Universal Sentence Encoder" module from Google.
This model can be downloaded by running the following command:
mkdir ./out/models/universal-sentence-encoder
curl -L "https://tfhub.dev/google/universal-sentence-encoder/2?tf-hub-format=compressed" | tar -zxvC ./out/models/universal-sentence-encoder