User-centric pattern mining on knowledge graphs: An archaeological case study

Abstract

In recent years, there has been a growing interest from the digital humanities in knowledge graphs as data modelling paradigm. Already, this has led to the creation of many such knowledge graphs, many of which are now available as part of the Linked Open Data cloud. This presents new opportunities for data mining. In this work, we develop, implement, and evaluate (both data-driven and user-driven) an end-to-end pipeline for user-centric pattern mining on knowledge graphs in the humanities. This pipeline combines constrained generalized association rule mining with natural language output and facet rule browsing to allow for transparency and interpretability—two key domain requirements. Experiments in the archaeological domain show that domain experts were positively surprised by the range of patterns that were discovered and were overall optimistic about the future potential of this approach.

Introduction

Digital humanities communities have shown a growing interest in the knowledge graph as a data modelling paradigm [1]. Already, this interest has inspired several large-scale international projects – amongst which are Europeana,¹ CARARE,² and ARIADNE³ – to actively explore the creation and publication of knowledge graphs in their respective domains. These knowledge graphs, and many others like them, have been made available as part of the Linked Open Data (LOD) cloud – a vast and internationally distributed network of heterogeneous knowledge – bringing large amounts of structured data within arm’s reach of humanities researchers, who are now looking for ways to analyse this wealth of knowledge. This presents new opportunities for data mining [2].

Data mining is the process of identifying valid, novel, potentially useful, and ultimately understandable patterns in data [3]. These patterns describe regularities in a dataset which can help researchers gain more insight into their data. Researchers can then use this insight as a starting point to form new research hypotheses, as support for existing ones, or simply to get a better understanding of their data [4]. This entire process can take weeks or even months of hard work in the traditional setting. However, by incorporating data mining into the workflow, much of this time can be saved through the automatic discovery of potentially-relevant patterns. This makes data mining interesting as a support tool for humanities researchers.

Of course, the idea of using data mining as a support tool in the humanities is, in itself, not novel. There have been various attempts before, for instance to classify coins [5] or to cluster cultural heritage [6]. However, the majority of these studies involved mining unstructured data, most commonly in the form of text mining, whereas mining structured data has thus far been largely limited to tabular data and tailored to specific use cases. With the growing popularity of knowledge graphs in the humanities, mining patterns from these structures becomes ever more important to researchers in this domain.

This work present the MINing On Semantics pipeline (MINOS) for pattern mining on knowledge graphs in the humanities. Its aim is to support domain experts in their analyses of such knowledge graphs by helping them discover useful and interesting patterns in their data. To this end, the MINOS pipeline places users in the centre by letting them guide the mining process towards their topics of interest and by letting them focus the results via a facet pattern browser.

Under the hood, MINOS employs generalized association rule mining (ARM). An association rule is an implication of the form $X⟹y$, where the presence of a set of items $X$ implies the presence of another item $y$. These implications are learned by iterating over a dataset of examples, called transactions. Generalized ARM works largely the same, except that the antecedent $X$ holds the item classes rather then the items themselves.

This method was specifically chosen to help overcome two key issues with technological acceptance in the humanities, namely transparency and interpretability [7], [8]. With transparency, we refer to the ease with which a method and its underlying theory can be understood: a black box method, for example, is less transparent than a glass box one. With interpretability, we mean how easily one can interpret the results of a method: it is, for instance, typically more difficult to interpret an $n$-order tensor than it is to interpret a set of symbolic statements.

Generalized ARM satisfies both of these constraints: it employs basic statistical know-how to produce human-readable rules in an overall deterministic process. A limited background in statistics, which most humanities researchers possess, therefore already suffices to understand how these rules map back to the input data and to check whether they are valid. This allows humanities researchers to put their trust in both the method and its results [9].

Of course, this trust is only gained if the produced rules provide useful and interesting patterns which can help these researchers to get a better understanding of their data. We call this the effectiveness of the approach. To assess this effectiveness we conducted experiments in the archaeological domain, specifically on data from various excavations, during which domain experts were asked to evaluate a set of candidate rules on interestingness.

By placing domain experts at the centre of both the pipeline and its evaluation, as opposed to data scientists, we contribute to an as yet largely unexplored niche in this intersecting field of data mining, knowledge graphs, and humanities. Concretely, our main contributions are (1) insight into some of the challenges and possible solutions for introducing data science tools to the humanities, (2) a pipeline design for pattern mining on knowledge graphs which is tailored to domain experts rather than to data scientists, and (3) a user-driven evaluation of our design choices instead of only a data-driven one.

With these contributions, our research aims to add to the interdisciplinary field of the Digital Humanities. For this reason, we will refrain from developing an ARM algorithm from scratch, but instead focus on how we can augment such an algorithm with complementary components to make it into an effective tool for Humanities researchers.

A concise overview of related work is given next, followed by an overview of the pipeline, the dataset, and the experimental setup. This paper then discusses the results from the user-driven evaluation, and concludes with a reflection on the chosen approach in light of these results.