Once it was enough for companies keen to know what people thought about them to subscribe to a cuttings service or commission market research. But the information revolution has changed all that. The number of daily tweets worldwide increased over the year to September from 60 million to 230 million, according to Twitter. Try asking a cuttings service to track all those.
Factor in Facebook, YouTube and all other social media sites and blogs and the dilemma for companies becomes clear. Exactly how do they attempt to keep track of what is being said about them in this information deluge?
'In the 20th century, corporate communicators suffered from the problem of a paucity of data,' says Philip Sheldrake, founding partner of venture marketing company. Meanwhile and author of The Business of Influence: Reframing Marketing and PR for the Digital Age. 'In the 21st century, they have the opposite problem. There's way more data than anyone thought we'd ever see and it's increasing every year at a massive rate of knots. We call it the Twitter fire hose and the problem is that we are going to have a few hundred of these fire hoses by the end of the decade.'
Monitor and respond
So how can such a welter of data be monitored and responded to, if appropriate? It certainly can't be tracked properly by Internet search engines such as Google, says Andrew Smith, director of digital communications agency Escherman.
'The problem with Google is that it is not necessarily real time,' he adds. 'If someone tweets about BT, the company wants to be able to respond as quickly as possible. Google is not necessarily going to be able to help with that. And it is not skewed to social media.'
For a growing number of organisations, from BT Group to Cisco Systems, American electronics retailer Best Buy and South Korea's Kia Motors, automated sentiment analysis offers one answer.
At its simplest level, this is a kind of Internet search allowing organisations to find Internet and social media posts mentioning them or their products and assess whether the references are positive, neutral or negative.
'In the old days we would have called it opinion polling. It's really about understanding what people think and how they feel about organisations and then having an input to try to change that on behalf of an organisation,' says Smith.
'The difference is that, where in the past, opinions could be taken over weeks and months, now clients want it in real time so they can get as much understanding as possible of what people think and respond accordingly.'
A complication is that there are several different types of sentiment analysis and the cost varies accordingly.
The most basic method uses dictionary-based technology, which attributes connotations and sentiments to words that it finds in stories about clients. Such offerings are relatively inexpensive, with one provider, www.socialmention.com even giving it away on its website free of charge. Company names or other words are simply entered on the site, which quickly supplies analysis of to what degree they are being used positively, neutrally or negatively on Twitter, Facebook, YouTube and other social media sites. As the lack of a price might indicate, Social Mention's is a basic service, however, and can have difficulties in detecting irony and slang. If someone tweets 'great' for example, it is not necessarily clear whether this is positive or sarcastic.
A second approach is natural language processing which uses sophisticated software and algorithms and claims much better results in both detecting meaning and sentiment from the context of a phrase and dealing with high volumes of content. A growing number of sentiment analysis firms, such as uberVU, Radian6, Alterian and Attentio operate along this spectrum, with Glide Technologies, a British firm operating at the very top end of the market and charging clients, which include Sony, between £15,000 and £100,000 a year.
Glide chief executive Sam Phillips says that some of the problems encountered by automated sentiment analysis sites are simply to do with context. The phrase 'banker's bonus,' for example, currently carries negative connotations with the general public, while continuing to be extremely positive for the bankers concerned.
Phillips says Glide's software combats such issues by using a 'head noun approach' that identifies the active ingredient in a sentence. 'Banker's bonus attacked' would then be classed as a negative mention, although that would also depend on whether the customer conducting the search was indeed Barclays, or a rival bank with no UK bonuses to defend. Phillips says Glide's technology can also take account of emoticons, such as smiley faces in Facebook posts. The firm claims accuracy rates of more than 80 per cent for its system - about the same level, notes Phillips, as the accuracy quotient in the average statement of a human being.
'Sentiment analysis is not new as a field but there have been a lot of false dawns with it because of its purported accuracy and whether that met expectations,' he says. 'We feel we are significantly ahead of the marketplace, because we have developed this contextual approach.'
Keith Woods-Holder, Glide's chief technology officer, adds that it is also important to understand precisely what sentiment does and does not provide. 'There's a lot of confusion between tonality, which is the polarity of a statement - whether it is positive, neutral or negative - and sentiment, which is about contextualisation, or where statements fit into a framework,' he says.
Adapting and lessons
Smith believes the market is taking time to adjust. 'Everyone is learning the best ways to adapt to the new environment,' he states, 'but the biggest challenge is how to respond to it.'
Indeed, Warren Buckley, managing director of BT Retail Customer Service, has a team of more than 30 staff devoted to referencing Internet and social media messages about BT. They track customers' issues and get in touch to offer help, sometimes not to everybody's delight. On one occasion BT contacted a Facebook account holder after detecting an uncomplimentary comment about the company on his status updates, only to be met with a diatribe about a breach of his individual freedom.
'Insight drawn from social media channels offers the opportunity to understand much more about our customers, what they like or hate, what drives them to buy or leave and in particular how peer recommendation works,' states Buckley. 'By analysing the sentiment behind a tweet, a post or a poke we can compare online commentary to offline activity and draw conclusions for business decisions. This is a brave new area for business if we are brave enough to respond positively to what it tells us.'
So where does sentiment analysis go from here? Woods-Holder sees accuracy rates increasing to a maximum of about 94 or 95 per cent. He argues that any higher is virtually impossible because of the speed at which language evolves.
That's the sort of level where Glide would like to think that sentiment analysis will become an almost obligatory budget item for most large corporations. 'The biggest issue is the complexity of it,' says Phillips. 'We have tried to a take the maths out of using it and come up with something as easy to use as Microsoft Office. The question will be What can't sentiment analysis do?'
With sentiment yet to become a fully developed tool and public relations budgets under pressure, a more pressing issue might be why companies should bother with it at all but Phillips doesn't see that as a problem. 'What's fuelling the demand for sentiment analysis is the massive changes we're seeing on the media landscape, which is becoming increasingly fragmented,' he says.
'If businesses are going to be able to respond to statements on social media sites they need to be able to know about them in real time. Our customers have long complained that they find research validation services such as opinion polls like looking in a rear-view mirror when they want to be dealing with current issues.'
Sheldrake goes much further. 'During most of the 20th century, organisations were very inward-looking but towards the end they became much more customer-centric,' he says.
'Now we are seeing the rise of the social enterprise, which seeks really meaningful relationships and emotional engagement with customers.
'There's a real competitive advantage in doing that. You need machinery to deliver those insights but if you don't and your competitors do, you're going to be disadvantaged by standing still.'
Understanding the concept
To understand how sentiment analysis works, there's no substitute for putting some words to the test and www.socialmention.com lets anyone do it for free.
CorpComms Magazine tried out the sentiment analysis firm's website search engine at the end of September and the limitations of this most basic approach to sentiment analysis were clear.
A search for BT Group, for example, brought up 558 'mentions' occurring at the rate of 13,315 seconds per mention, giving the company an 11 per cent likelihood that its brand was being discussed in social media at the time and a 36 per cent chance that individuals talking about the brand will do so repeatedly.
Some 421 of the mentions were deemed neutral, 111 positive and 26 negative, giving a sentiment ratio of 4-1, but beyond this headline the basis for those judgements seemed sketchy.
The top 'keyword' was 'group,' shedding little insight on what the 278 unique authors were referring to, and other top keywords - percent, Mahindra and Kabra - didn't help much either.
Nor did the picture improve much when CorpComms Magazine delved into some of the tweets mentioning the company's name.
Few seemed to actually be about BT Group, with the top item turning out to be about a boy band on X Factor.
Unfortunately, it seems that Britain's biggest telecoms company shares its acronym with the preferred short term of many Twitter users for the word 'but'.