Data Provenance

Data Provenance: On the origin of Data

by Ernest Field

The term Data Provenance refers to the subject of tracing the origins of data and its transfer between datasets. Determining the origins of scientific databases is a major concern.

Data Provenance (DP)

DP is defined as:

Meta-data that describes the history of a data set starting from its original sources. A web search revealed scores of articles on this subject. All the examples concerned large datasets who’s DPs were difficult to assess. These datasets arose from workflows and transfers from other datasets. Some datasets may have indeterminate origins. The complexity of the generation process can make it difficult to assign a quality rating to many of these datasets.

Several articles report on the role of DP in the validation of scientific databases. One issue they discuss is the development of the languages needed to describe data generation histories. One article introduces the idea of developing provenance stores or vaults that contain quality assured data. Another reports a major study on DP undertaken by IBM in collaboration with the EU. There are many more applications of DP reported that apply to specific subjects.

OR Supports Data Provenance

The quality of data is vital in all practical OR projects and underpins the quality of the final results. Increasingly data generators are being used to develop new techniques and databases but good quality data is still required. The many datasets that appear in OR projects fall into five main Categories, each with their own DP characteristics.

  1. Existing data on an organisation’s operations.
  2. Similar data on other organisations.
  3. New data produced by observation of the operations under study.
  4. Data produced by analysis and modelling.
  5. The results of literature surveys to check on other people’s relevant work.

There is a Yorkshire proverb that says:

"Never believe anything th’as not seen for thee sen, and then don’t be too sure”.

This cautionary principle should be followed in assessing the quality of data.

A more formal approach to DP would include providing for each data element meta-data, such as the last date of update. In the case of a database this would involve adding new DP fields.

Value Of Provenance

The benefits arising from DP are likely to be application specific, although the development of new DP languages will have general benefits. DP assists in identifying datasets that are unreliable because of missing or inaccurate data. Most OR studies find existing datasets need updating or upgrading for use by the OR study. The benefits accruing to the organisation from the improved data should be claimed as an OR result.

To Summarise

  • Data provenance is important to all users of large datasets.
  • The complexity of DP requires new languages to record the metadata.
  • OR must continue to use high provenance data
  • The subject of data provenance should be kept under review.
September 2009: Inside OR

Latest News

October 2021

FiveThirtyEight publishes its sports forecasting formulae

Nate Silver’s predictive analytics journalism website FiveThirtyEight has published the maths behind its sports forecasting model, making it possible for you to download the files and and run your own NHL league!

Read More

September 2021

Using data science to improve wildfire forecasting and modelling

A new partnership between the University of California San Diego and the USA’s Los Alamos National Laboratory aims to use thermal, mechanical and fluid dynamics systems modelling to help meet the global threat of wildfires.

Read More

September 2021

Benefits of data analysis not felt in companies’ marketing functions

A report, entitled Navigating Blind Spots: The State of Digital Marketing, reveals that 73 percent of 200 companies studied were caught off guard by new customer trends, despite having access to significant amounts of data on customer behaviour, purchase processes and preferences.

Read More