Knowledge-based Analysis of Open Data

Abstract

There is a shortage between the emerging availability of time stamped open data and the capacity of extents in the application sphere to turn it into a reliable source of knowledge without the help of data scientists. Time series now concern every aspect in our daily life, from environmental monitoring to music, or the telemetry of our actions on social media. This struggle of understanding the underlaying data behaviour deepens when considering the heterogeneity problem of the scattered series, that must be handled consistently to produce a reliable source of knowledge. The current approach proposes to target this bottleneck by generating a versatile and domain-neutral representation of data, containing statistics, designed to be consumed both by people (via a brief interaction) and by systems (via a standardized enriched structure). This method has the aim of converting tacit data into explicit artefacts, prioritizing time series knowledge representations. The paper will present a knowledge-based approach to time series interpretation, which takes a dataset from an open data source, without domain assumptions, transforms it into a common format, with semantic enrichment of metadata and includes analytical results. The open data are mapped onto a sub-part of the SemTS (Semantic Time Series) ontology, so that series characteristics can be stored and reused as knowledge artefacts. The final output is a knowledge specialized JSON, partitioned into insightful sections (e.g. dimensions, feature families), plus explanations in natural language, with the aim of reducing comprehension barriers for non-specialists and facilitating subsequent integration into knowledge management flows, like cataloguing, searching or comparing. The proposed format encompasses insights regarding the series, drawing a profile for the series that includes quality indications (e.g. missingness, gaps, outliers), trend patterns, seasonality (cyclicality). The contribution is a practical knowledge aspect for time series analysis, a tool to convert raw time-stamped open data into reusable knowledge artefacts, that could guide downstream decisions in the data discovery procedure. The intent is to assist with accessible diagnostics, rather than to replace analytical expert judgment.