Found a cool medium article on applying knowledge graphs to the financial industry, modelling the balance sheet to hack the Pearson exams!

Why is knowledge graph so important and …cool?

It’s a way that allows you to not only effectively connect different types of datasets together but also allows you to perform reasoning and discover connections that you otherwise wouldn’t have been able to if you were to analyse these datasets separately.

Knowledge graphs provide:

  1. Context – provides context to algorithms by integrating information into an ontology and applying a reasoner to derive new knowledge

  2. Effectiveness – graph algorithms provide effective and low costs of computations

  3. Explainability – has a good tradeoff between accuracy and explainability

What is a balance sheet?

A balance sheet is a statement of assets, liabilities, and capital of a business at a given time. It contains a lot of entities, relations, and rules in different hierarchies. One major problem in dealing with balance sheets is to be able to normalise them and put different balance sheets into a single frame for comparisons and forecasting.

How did we apply knowledge graphs to MCQ of Pearson exams on balance sheet, income statement, and cash flow statements?
  1. Initialisation of knowledge graphs

  2. Query regarding liability is connected to equity (entity)

  3. Translation of request through fuzzy matching

  4. Graph computation and answers

Accounting rules are incorporated into a knowledge schema which allows the knowledge graphs to make logical reasonings. There are two types of inference:

  1. Type-based inference – automatic detection of data type

  2. Rule-based inference – logical rules to deduce new information

Essentially, KGs can derive new conclusions from undiscovered relationships and it’s able to abstract complex patterns into simple queries.

What is a graph embedding?

Graph embeddings aim to convert graphs into a low dimensional space where the graph information is preserved.

What is AmpliGraph?

It is a suite of ML models for relational learning, useful for knowledge graphs. Use AmpliGraph if you want to:

  1. Discover new knowledge from an existing knowledge graph

  2. Complete large knowledge graphs with missing facts

  3. Generate independent knowledge graph embeddings

  4. Develop and evaluate a new relational model



Data Scientist

Leave a Reply