mgnlq_er

mgnlq_er

Entity recognition for mongo nlq

entity recognition based on word categorization

the word categorization contains a bitmap filter to retain only sencences which are homogeneous in one domain

entity recognition based on word categorization

Words are categorized according to an index (see mgnlq-model)

into

• "Facts",
• "Categories",
• "Domain",
• "Operators",
• "Fillers",
• "Any" (generic verbatim strings)

The word categorization contains a bitmap filter to retain only sencences which are homogeneous in one domain.

The word index is built by mgnlq_model

usage:

  var erbase = require('mgnlq_er');
  var words = {}; // a cache!
  var res = Erbase.processString('orbit of the earth', theModel.rules, words);

result structure is a set of sentences and associated errors

sentences are further pruned by removing: sentences containing Words containing identical strings which are mapped onto distinct entities, sentences containing Words containing distinct strings which are mapped on the same entity ( if a better match exists )

Test data

the tests run against recorded data in E:\projects\nodejs\botbuilder\mgnlq_testmodel_replay\mgrecrep\data\807d3ce983c2f3....

This data can be recorded by setting

SET MGNQL_MODEL_NO_FILECACHE=1

0.0.4 -> single result in checkOneRule

entity recognition mgnlq_er parsing mgnlq_parser1 querying