One of project Aiur’s main goals is to increase the quality standards of published research results. With the Knowledge Validation Engine we aim to build a machine learning-powered engine that can produce a report on the reproducibility and hypothesis trust levels of an input text or document.

In terms of what it does, the Knowledge Validation Engine identifies the building blocks of an input article, classifying them as part of a category, including problem, solution, evaluation method, results, etc. These building blocks are then backtracked to other building blocks in the scientific literature. If Article A has a building block supported by Article B, there is a need to split the building blocks in Article B, and keep doing this until a stop condition is hit (i.e. no more articles supporting the building blocks). The logic is very similar to the one used by the PageRank algorithm for links, but we will apply it to factual data.

Each building block has its own underlying hypothesis. With these identified a causality graph can be built, and based on the rank of the causality the trustworthiness of each hypothesis scored. For example focusing on whether for a given evaluation metric used (M), all the assumptions (A, B, C) and constraints (X, Y, Z) are outlined.

As a final step the Knowledge Validation Engine will build a report pulling results from every branch of the knowledge tree, highlight the problematic branches that present lower trust levels. Experimental research will be initially better suited for Knowledge Validation Engine scrutiny that, say, mapping studies.  More information is included in the ‘Aiur development roadmap’ section of the white paper.

Learn more about Knowledge Validation Engine on our blog.