Recommender Solutions - A Brief Introduction

Overview

Recommender systems are powerful tools able to assist users in accessing information, media, products, and other assets. In an enterprise however, recommender systems are only helpful if they can address the specific requirements of the employees using them. Mainly driven by machine learning approaches, AI supports data-driven corporate decision making, but to decision makers machine learning is like a black box since the machine learning systems themselves do not support provision of the reasons for their results. By using knowledge graphs as a form of knowledge representation the system is indeed able to provide explanations of these results even if such decision making is a complex process. Semantic recommender systems allow users to make sophisticated matches between the concepts that have been defined by not only focusing on similarities between products or business items, but equally on both background and context.

PoolParty Semantic Recommender

The PoolParty semantic recommender is a knowledge-based recommender. The structured data we use is a knowledge graph (KG) with highly expressive taxonomic and ontological information to leverage the knowledge for high quality recommendations.

PoolParty knowledge graphs are based on W3C standards defining the Semantic Web stack of technologies. These standards include the Simple Knowledge Organization System (SKOS), the Web Ontology Language (OWL) and further languages to represent knowledge for specific purposes and use cases. Using these standards provides several advantages from interoperability and clearly defined semantics to standardized databases able to host and process such structured data.

The information space for the PoolParty Recommender is any corpus of documents of a domain represented in the knowledge graph and can therefore be analyzed by the PoolParty Extractor to identify KG entities. In contrast to content-based recommenders, PoolParty represents textual content as a graph structure based on extracted information and thereby directly integrates it into the knowledge graph.

The PoolParty approach to recommenders provides high recall as it is able to retrieve results not obviously related to the input, rather via ontology-based classifications and relations in the KG. It also maintains high precision due to the knowledge graph including explicitly defined facts and being based on a knowledge graph representing a curated domain model. With the adaptable architecture of the PoolParty Semantic Recommender solution it will be possible to link knowledge graph entities to statistical AI implementations (ML, LLMs, …) in the architecture thereby synergistically combining their functionalities.