LearnITy™ Knowledge Engine (LKE) is a deep document understanding platform built on AI and NLP technologies that ingests organisational born-digital document assets (PDF, Word, Excel) and incrementally converts them to actionable corporate knowledge with provision for validation and augmentation by human experts.
The ability to process and extract key information from large and complex documents is extremely valuable for a number of industries
Gain speed and agility in your business processes dealing with such unstructured born-digital documents in PDF and Doc formats
Extract information from documents, validate document data against business rules, compare 2 or more documents, categorise and search documents based on concepts and key terms
- Information Extraction
- Information Extraction is the foundational application of DU where structured data embedded in documents need to be extracted based on various attributes of the data.
LKE’s powerful engine supports extraction of embedded document data based on different types of attributes (and their combinations)
Extract components of document (sections, pages, paragraphs,sentences, clauses)
Extract Key-value pairs (e.g., Invoice Number)
Extract Tables with row-column info for data items
Extract word(s) that are mentions of Named Entities (built-in entities such as Person, Date, Location, etc. or custom entities meaningful for your business domain, e.g., Quote )
Extract word(s) that are part of Relationships (either built-in such as is-a, part-of, etc. or custom relationships appropriate for your domain, e.g., TreatedWith[disease, medicine] )
- Document Categorisation and Search
- Document Categorisation and Search supports users who professionally search for information in document assets of their organisation (e.g., Patent applications, VoC).
A variety of approaches is supported by LKE so that you may adopt one that suits your business conditions
Define business rules and patterns for categorizing documents. The patterns are defined using a large collection of linguistic annotations that are obtained by processing of the documents
Use machine learning/deep learning techniques to train your custom categorization model if you have enough documents that may be used for training
Going beyond keyword-based search support is provided for semantic search (match a query with a document even if they do not use the same words) improving the efficiency of search processes by increasing the accuracy and relevancy of search results
Support advanced techniques like Approx. Nearest Neighbour
- Document Validation
- Document Validation (e.g., KYC/AML) checks for consistency of data elements in a document based on business rules and reference master data.
Checking the completeness and consistency of documents is a key capability of LKE
Both intra & inter document validation checks may be defined
Identify the data elements in a document for which you wish to define validation checks; these could be anything from key-value pairs to page headers
Define the validation checks using patterns and signatures combining the large number of annotations that are available (e.g., regular expressions, named entities, custom dictionaries)
Use external reference data sources, 3rd party databases, and APIs for validation
Escalate to human-in-the-loop in a maker-checker environment
Additionally define document stylometric and sentiment checks
- Document Comparison
- Document Comparison (e.g., comparison of Company Annual Reports) matches different data elements of 2 or more documents based on defined comparison criteria
LKE implements a very general and powerful approach to document comparison
The documents being compared (may be more than 2) need not have the same (or even similar) structure
You define the set of data elements to be used for comparison; these may be of different types for different documents
Element A (may be a sentence only to be found in page 1) in document 1 may be compared with Element B (may be a phrase only to be found in a table) in document 2
The comparison criteria may be defined as a (boolean) combination of various conditions with support for custom comparison functions, textual entailment and keyphrase matching
Apart from string comparison (using algorithms such as Cosine similarity), comparison of numeric values and dates is supported
