Fuzzy searches based on numerical context in a document management system

BACKGROUND

A company has developed and internationally markets a Document Manager, capable of managing documentation and its workflows according to its internal procedures.

Problem

The Document Manager allows users to search for documents through a filter system, allowing to set search strings or specific values ​​for certain fields (such as the amount in the case of an invoice, or the delivery date in the case of a delivery note). However, sometimes users don’t know the exact values ​​to set the filters, making searches complex.

Benefit

The integration of a new module based on fuzzy logic concepts allows users to carry out fuzzy searches, such as “high amount”, “near delivery date”, etc., capable of retrieving information based on context, allowing the selection of documents without the need to establish specific values.

METHODOLOGY & results

Architecture: Cloud architecture deployed on its own servers, using SQL Server as a storage system.

Developing language: Java, Python.

ML techniques: Unsupervised learning algorithms (K-means, HCA), supervised learning algorithms (decision trees) and fuzzy representation models.

Results: Depending on the parameterisation, the system allows a successful recovery of the documents sought in between 90% and 99%.

This project was carried out before AI Shepherds’ foundation by its team members.
Scroll to Top