Despite the advance of Twitter and Facebook, email is still the “killer” electronic communications medium particularly at work, where it long ago replaced the telephone as the preferred way for communicating between colleagues. But not all of the billions of emails handled each day by office workers are strictly work-related. According to a recent study, about 15% of the emails sent by corporate users contain information about a person or persons not among the recipients. In other words, they contain gossip.
Long before social media started giving compliance officers sleepless nights, office workers have been using email as an electronic rumor mill, to let slip unguarded comments on co-workers or badmouth corporate policies or their boss.
To be sure, they say positive things as well, but according to the researchers, negative gossip far outweighs the praise by a factor of 2.7 times.
But how did they sift good gossip from the bad? Using natural language methods, specifically a technique called “named entity recognition” to identify gossip and then performing sentiment analysis to sift the good stuff from the bad.
This novel use of Natural Language Processing (NLP) technologies to analyze office gossip opens up a vast new range of possibilities.
In the past, the difficulties of searching unstructured data such as emails have tended to limit initiatives in this area to compliance applications.
But it is one thing to monitor the emails of a group of salespeople suspected of collusion, quite another to monitor the mostly innocuous conversations of an entire organization.
As well as the obvious ethical and legal issues, conventional data mining techniques, even if they could handle the volume of data, would struggle to make sense of most gossip. But researchers say that there is value to be found in office gossip.
“Gossip gets a bad rap,” says Eric Gilbert, an expert in social computing at Georgia Tech in the US, who headed the research project. “When you say ‘gossip,’ most people immediately have a negative interpretation, but it’s actually a very important form of communication.”
To conduct his research, Professor Gilbert and his team examined nearly half a million emails exchanged by Enron employees in the five years prior to the energy company’s bankruptcy — the largest in the US — at the end of 2001. Many executives at Enron went to prison for their part in the company’s downfall after evidence emerged of widespread destruction and alteration of records to cover their tracks.
Two good things did come out of the Enron scandal, however: the Sarbanes-Oxley Act and a comprehensive five-year database of its email records, which has proved a goldmine for researchers.
The Georgia Tech team wanted to measure how gossip flows through a large organization; they did not set out to understand what the gossip was about, which in the case of Enron would perhaps have been even more revealing.
Professor Gilbert and his team divided the emails among seven layers of Enron hierarchy, from rank-and-file office employees all the way up to presidents and CEOs, and found gossip emails flowing within and among nearly every level, with the heaviest flow among the rank-and-file.
Despite the uniqueness of the Enron case, Professor Gilbert argues that the gossip contained in the Enron email database is not that different from that of any other large organization.
He says “a lot of the emails we’re looking at were from the rank-and-file, and it was the Enron CEOs – a tiny fraction of its employee population – who initiated and directed the actions that brought the company down. The average employee had no idea what was going on.”
Indeed, one of the most common words found in emails classified as gossip was “born” — while the senior execs were plotting the destruction of the company, the rank-and-file workers were sending each other emails about their newborn children — just like any other organization.