danishahmed92 / rapid

Relation extrAction using Pattern generation and semantIc embeDdings

RAPID (Relation extrAction using Pattern generation and semantIc embeDdings)

RAPID aims to determine dbo relations from a given natural language text, and tries to disambiguate between multiple relations by means pattern generation and semantic embeddings.

Resources Required

External Libraries

• JWI 2.4.0 WordNet (https://projects.csail.mit.edu/jwi/)
• Stanford CoreNLP Models (https://stanfordnlp.github.io/CoreNLP/download.html): -- stanford-corenlp-3.9.1-models -- stanford-english-corenlp-2018-02-27-models -- stanford-english-kbp-corenlp-2018-02-27-models

External Data

• Word2Vec Model (https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/)
• Glove Model (http://nlp.stanford.edu/data/glove.840B.300d.zip)
• fastText Model (https://dl.fbaipublicfiles.com/fasttext/vectors-english/wiki-news-300d-1M.vec.zip)
• WordNet 3.0 (https://wordnet.princeton.edu/download)

Remaining Data

• All other data can be found at "data"

Installation Guide

1. Import "evaluation_all_embed_threshold.sql" MySQL database.
2. Set the required database configuration, and data paths in "dbConfig.ini" and "systemConfig.ini" respectively.
3. Execute "Application.java" to start REST API service.
4. To run Web-App demo, run command "node app.js" from directory "rapid-webapp" 4.1 Make sure the REST API is already executing.

System Requirements

1. Tomcat 8
2. MySQL 5.1.39
3. NodeJS 8.10.0
4. At least 18 GB of ram (Since all models of Word2Vec, Glove, and fastText are loaded simultaneously along with their respective Property Embedding Model - We used 32 GB of ram)

Useful Readings

1. Theoretical Document
2. Brief Overview
3. RAPID In Action
4. Contact Person

Credits

• Supervised by Dr. Ricardo Usbeck
• Submitted to Prof. Dr. Axel-Cyrille Ngonga Ngomo