Sentiment analysis uncovers a pervasive cultural issue

Monday, 7 January 2019

When a senior executive from an Australian-based retirement fund provider abruptly resigned, the circumstances of his departure raised some questions. The organisation’s human resources department scanned the executive’s emails and found suspicious conversations about gambling. Worried that this executive may have also been engaging in fraudulent activity, the department turned to KordaMentha to analyse the executive’s emails and other unstructured data.

KordaMentha didn’t pinpoint any evidence of fraud. But using Relativity, exposed a more pervasive organisational issue.

Going Beneath the Surface with Relativity

Based on the initial review of the executive’s emails, KordaMentha helped the client collect a wider data set representing the suspected employees involved. KordaMentha processed the data in a customised workspace using Relativity Processing.

During the initial review, KordaMentha found instances of the executive and some of his colleagues using language affiliated with sexist sentiments. So, they armed the workspace with libraries of specific terms and phrases associated with misogynistic sentiments in addition to customised visualisations and an application that connects with third-party Natural Language Understanding artificial intelligence (AI) platforms.

Next steps for KordaMentha entailed taking those libraries’ terms and phrases and transforming them into search terms reports. Then, a deep dive into the search terms reports and looked at documents where certain phrases appeared, so we could better understand how those words were used in context.

While the team was at work, the third-party Natural Language Understanding AI platforms assessed the data that hit on key custodians and time periods. The platforms then analysed that data in relation to aspects such as sentiment, relationships, and entities.

With the project coming full circle, KordaMentha conducted a targeted review and evaluated the documents containing negative sentiments and key terms from the various libraries.

Applying this functionality further illustrated that the executive and a multitude of other employees throughout the organisation were using language associated with sexist sentiments in their emails and messaging platforms.

“The analysis we conducted in Relativity made it clear that the misogynistic language was happening throughout the organisation,” Craig Macaulay, executive director of KordaMentha’s forensic technology team.

KordaMentha met with the client to present the findings, using Relativity’s custom-built visualisations to illustrate the extent of the issue, showing the key players and conversations.

The client was floored—they recently conducted an organisation-wide survey on sexism and “passed with flying colours,” according to Craig.

“What people would say in interviews or to leadership was much different than what was being communicated throughout the organisation on a day to day basis,” Craig added.

The client drilled down into specific documents to see how the identified phrases were used in context.
 

“Culture is not traditionally thought of as something you can analyse with technology,” said Grant Whiteley, executive director of KordaMentha. “Our client was astounded in our ability to use the data to show them key insights about their organisation.”

The Investigation Expands

The client gave the green light to expand the search to the myriad individuals they identified in the executive’s emails. This increased the scope of the matter to nearly 1.5 million documents.

Armed with this information, the client could act to remedy the issue and mitigate their risk. They fired a handful of employees, issued sanctions to several others, and mapped out necessary organisational changes.

First published by Relativity