Systems and methods for relation inference

A method for using a trained Recurrent Neural Network (RNN) model to generate relational information between terms of interest, the method comprising:

• receiving an input term;
• obtaining a feature representation of the input term;
• inputting the feature representation of the input term into the trained RNN model, which has been trained using, at least in part, co-occurrence of terms in training data to gauge correlation of the terms, that predicts a correlation factor between the input term and each term from a set of terms;
• and
• outputting terms that have a correlation factor that is above a threshold value.

A non-transitory system for correlation inference between terms of interest, the system comprising:

• a search processor communicatively coupled to a database, the search processor obtains documents that comprise terms that are included in a term repository;
• a data processor communicatively coupled to the search processor, the data processor identifies one or more terms in a set of documents;
• a data representation processor communicatively coupled to the data processor, the data representation processor generates features for the one or more terms;
• a feature representation library communicatively coupled to the data representation processor, the feature representation library stores at least some of the features;
• a sampler communicatively coupled to the data processor, the sampler performs the steps of:
• using the set of the documents to create document samples that comprise two or more terms;
• and
• determining relative sequential orders of the two or more terms within the document samples;
• and
• a Recurrent Neural Network (RNN) model that uses the feature representations and the relative orders to predict a correlation factor.