/WordLink

An NLP Research project in which we try to establish a concrete mathematical relation between common word pair relations and then to improve the performance of word embeddings by integrating word pair relations.

Primary LanguageJupyter Notebook

banner

Notebook Links

Problem Statement

To establish a concrete mathematical relation between common word pair relations and then to improve the performance of word embeddings by integrating word pair relations.

Approaches

Modify an Existing Word2VecModel

  • Use Pre-existing Model to establish relations (one-one) with regression.
  • Establish relationship word tuples .
  • Convert said word tuples to vector tuples using multiple word2vec models.
  • Find a subspace using regression that satisfies this relation.

Create Custom Embeddings

  • Created a model with set word relation in mind.
  • Injected word pairs of that relation while training our own word2vec model.