Turning News Into Knowledge

While we’re always more than ready at MediaMiser to stand by our measurements, sometimes it’s nice to have our methods echoed by outside parties.

The second day of the International Conference on Market Research in the Mobile World in Atlanta, which concluded last week, did just that.

Thomas Brailsford, a consumer understanding and insight manager at Hallmark Cards, spoke about his company’s experiences using social media as a listening tool. We agree in particular with two points put forward by Mr. Brailsford:

-  That social media data is largely qualitative and unstructured; and

- That making any sense of this kind of data requires clear, specific questions.

Finding the right question to ask is always extremely important – and difficult. Asking a relatively broad question such as ‘What are consumers saying about my brand?’ is easy, but won’t produce much meaningful data or analysis.

No, a good question in these circumstances is a specific question. Something like: ‘Did event X affect how social media users view my brand?’ will more than likely give solid, actionable data.

This leads to two other presentations from the day, both of which focused on sentiment analysis. Specifically, to answer the above question – or any such similar query – we have to know whether people are speaking about the brand in a positive or negative manner.

The sheer volume of social media data, especially if you’re immersed in a popular topic, precludes manual toning. This places emphasis on developing a reliable, computer-driven methodology.

But the problem with using computers to analyze languages is that they aren’t (however much we might wish it so) thinking machines. Even the grimly brilliant supercomputer Hal, from the film 2001: A Space Odyssey, would probably have had trouble with the amount of sarcasm and irony used in social media.

To reliably analyze natural language, a program has to examine not just individual words within the body of data, but how people are communicating within the context of their posts.

David Johnson, CEO of Atlanta-based Decooda and Michael Wolfe, a senior director at the agency BBDO’s Atlanta office, each explained to conference-goers how they each use linguistic analysis to determine sentiment.

Both their sentiment analysis methods are backed by strong research, and rely on natural language and vernacular as used in social media.

Using linguistic science, both Decooda and BBDO are breaking down how language is used within social media and within a specific topic.

This goes beyond counting words predefined as positive or negative, and looks at how these words are used within the common language of social media.

They were fascinating presentations that reinforced for us that with strong research questions and effective language analysis, important and actionable knowledge can be gleaned from the mass of unorganised data that is most social media. On this, we seem to be in agreement!

Thinking of putting out a survey?

Better get out your mobile phone.

That’s because surveys conducted on mobile devices garner much higher response rates than traditional methods, according to research companies at this year’s International Conference on Market Research in the Mobile World.

Mobile marketers say demand is growing for location-based surveys, conducted through mobile devices such as an iPhone.

The conference, which I’m attending along with MediaMiser’s director of media intelligence Michael Sloboda, boasts an impressive roster of keynote speakers including Ravi Parmeswar, managing director of global insights for Citigroup, and Edelman Digital senior vice-president and global editor Brian Reich, among others.

The experts say, however, that for marketers to realize this increased response rate, users must also see a benefit – that’s where more robust user experiences come into play. By using mobile technology respondents don’t have to go to the survey, either physically or online.

Instead, they can fill out the survey wherever and whenever it’s most convenient.

Many of the presentations over the first day of this year’s conference, happening July 19 and 20 at Kennesaw State University in Atlanta, Georgia, discussed what’s happening across the spectrum of mobile market research.

Much talk was devoted to the uses of mobile technology for qualitative research like surveys.

Most experts say location-based surveys are a growing demand amongst mobile marketers. Examples of these include the tracking of people – with their permission, of course – through a period of time to gauge their habits, or triggering survey notifications when a survey panelist enters a specified location.

But there are barriers to doing location-based surveys. One is the battery power consumed by standard GPS tracking. Some operating systems, such as Apple’s iOS, provide a less-accurate version of GPS data collection that consumes less power, but is less effective for surveys conducted in urban areas.

After all, when apartments, stores and other buildings are squeezed together in an urban environment, an accuracy of 1.5 miles isn’t all that useful.

Privacy is also a concern, but is perhaps not as large an issue as many believe – after all, people, particularly young people, already share a great deal of personal data willingly for no other incentive but entertainment. Just check out Facebook or location-based services such as Foursquare.

All in all, it was a terrific first day of the conference and we’re eagerly anticipating Day 2!

This article is re-posted from our most recent newsletter. Sign up for MediaPulse and be the first to get articles like this…

An effective communications strategy requires reliable analysis of issues, including visibility in the media.

The measurement of media visibility usually relies on three metrics: attention, tonality and prominence. Of these, attention is the easiest to both track and understand.

Attention refers to the number of stories on a particular topic, as well as the circulation or viewership of the media publishing those stories. Attention is measured by calculating the total circulation for the topic (the sum of the circulations/viewership for each story). This should be tracked on a regular and ongoing basis.

Tonality applies an emotional context to a topic. It measures how the topic is presented in the media– which quotes are chosen, and, in general, whether the writer is inclined to a favourable or unfavourable view of the issues.

While attention is a simple linear measurement, tonality is set on a scale running between negative and positive. A simple three-point scale (positive, negative and neutral) can suffice, but a more complicated 11-point scale (-5 to +5) provides a measurement of greater depth.

Prominence is the most complicated of these three metrics. It’s based on an algorithm that weights different aspects of articles to determine an overall score of how prominent the topic is across media.

MediaMiser’s approach to analysis can combine these three metrics to give an understanding of how the media is portraying issues and spokespeople.

To obtain reliable measurements, articles are first sorted through keywords pertaining to the issues and spokespeople under study. These articles are then used to produce quantitative data, such as coverage over time, regional coverage and the overall share of voice.

The following chart, taken from MediaMiser latest whitepaper on prominence measurement, shows how coverage over time is typically displayed for multiple issues:

A typical coverage over time chart shows the number of articles over the vertical axis, with time over the horizontal axis.

The collected articles are then toned, using the three- or 11-point scale as described. The tone is based upon an agreed upon method and viewpoint, taking into account messaging and attributes that are important to the communication department.

Analysis on toned articles can be run to show the overall sentiment displayed by the media. This can also be broken down over a variety of criteria, including average tone over time, or tone per region or per specific time period.

Only toned articles are used to determine prominence measurements.  Prominence is graded using a flexible scorecard that assigns scores based on multiple measurements of media coverage.

These metrics include placement of keywords within both the articles and the media source itself, word count, circulation and number of keyword mentions. Prominence scores can be measured on an entire issue, or on a per article basis.

While these three measurements are all useful independently, when combined, it presents a complete understanding of media coverage. An issue with a negative tone is rendered less problematic if it has both low attention and prominence scores, while a positive message greatly benefits from higher scores. It’s little use, after all, to have a key message portrayed favourably if few people view it.

By portraying prominence on a high-to-low horizontal axis and tone on a negative to positive vertical axis, all three metrics can be displayed in a single chart, as shown here:

Each issue/spokesperson is here portrayed by a coloured marker. The placement of the marker is based on a combination of prominence and tone, while the size of the marker displays the relative attention the topic gained.

If you’re interested in learning more about the use of prominence in media analysis, check out:

MediaMiser’s whitepaper on measuring prominence. (free download)

Since people started communicating online they’ve been searching for better – or just faster – ways of saying what they need to. The internet is full of acronyms (lol brb roflmao) and emoticons , but it’s also working to shape the way people communicate –giving rise to new words on what often feels like a daily basis.

Though languages are constantly evolving (how many of us still communicate in middle English?) the modern world has seen new technologies push the boundaries of speed –nothing more so than social media.

Not everyone is happy with this, as can be seen by the New York Times asking writers to stop using “tweet”.
The argument here is that Standard English and spoken English are two different entities, and only the former should have a regular place in news articles.

While this garnered a lot of attention, of more interest is how the language of social media is used
Ben Zimmer of Visual Thesaurus examined the phenomenon of un-words

One of the hallmarks of social media is the ability for users to register their interest in something they see. But what if you change your mind? Then you can always undo the action that you’ve made.
LiveJournal, a virtual community of bloggers and diary-keepers, has been a pioneer in this type of usage. LiveJournalers were among the first to make friend into a transitive verb to describe the act of adding someone to an online list of acquaintances. (Other social networks like MySpace and Friendster soon got in on the act too.) To remove someone from friend status requires defriending or unfriending. Either the de- or the un- prefix works as a “reversative,” indicating the undoing of a reversible act (though un- seems to be winning out over de- as the preferred prefix these days).

Zimmer’s post is well worth the read, highlighting what is less an erosion of language than a new way of speaking. Social media users have simply invented a way of saying things English simply wasn’t equipped for.

In his post, Zimmer references a column by lexicographer Erin McKean in The Boston Globe.

In it, McKean examines not just the proliferation of tw– words used for everything twitter, but how twitter effects writing itself. And she sees it as far from a bad thing.

…it’s not just the twords that make Twitter interesting, it’s the character limit, the implicit constraint of being interesting, witty, informative – in short, of being worthy of the limited attention of your followers. The best tweets of Twitter (some of them collected on the occasionally not-safe-for-work site Favrd.com) are more epigrammatic than newsy. Twitter demands writerliness in a way that instant messages, text-messaging, and even blogging don’t.

McKean doesn’t see social media communication as a destruction of language, but instead as another step in the constantly-evolving world of communication.

Just as “blog” has become a common, and accepted, term for a new written form, terms like “tweet” and “unfriend” are part of the constant struggle to maintain communication in a constantly changing online world.

It’s hard to see anything wrong with that.

Jen Hogan is a Media Analyst at MediaMiser, follow her @jenogan

If you’re new to analysis, terms like qualitative and quantitative analysis may be just a little confusing. To look at the differences between, and benefits of, the two systems of measurement, a working definition is needed first.

According to WikiAnswers, qualitative analysis focuses on non-numerical data, such as words, pictures, or artefacts. Conversely, quantitative analysis focuses on numerical data.

Boiling it down, quantitative analysis will tell you what topics people are following, as in trending topics on Twitter.

The strength of each form of analysis can be seen in a recent report by MediaMiser’s own Samantha Ingram about media coverage of the 2010 Winter Olympics.

This chart, taken from the Olympic report, used quantitative analysis (counting the number of times each paper ran a story) to show the papers that contained the most coverage on the popular Olympic mittens.

Using quantitative analysis lets us see which areas are producing a lot of coverage, and which aren’t. Without this measurement, we wouldn’t have known that some of the top coverage was coming from outside Canada, although the mittens were only available here.

While quantitative analysis looks at what people are talking about and how much they have to say, qualitative analysis is more interested in why people are talking, and the contents of their conversations.

The above graph, also from the Olympic Snapshots report, looks at the overall tone of one of the Olympic coverage issues. Using a simple three-point positive/neutral/negative system (and rated by people, not computers), a large amount of articles can be sifted for some basic information.

Though it looks like quantitative analysis on the surface, the tone graph isn’t simply counting who said what. Instead, tone is focused on how people feel about a topic, and the broader content outside of basic keyword analysis.

Qualitative analysis is also used in open-ended surveys, looking at the reasons behind people’s concern rather than just the concerns themselves.

While each will give effective information on their own, without qualitative measurement, pure quantitative analysis can lead to number overload. Knowing how much people are talking is good, but knowing how they feel about a topic makes the information much more valuable.

Likewise, knowing people’s, perhaps customers’, concerns is important, but the information is far more useful if you can also look at which problems are getting the most attention.

Perspectives is a blogging series written, researched and compiled by teams of MediaMiser staff from Client Services, Sales, and R&D. This post was contributed by Jen Hogan, Sophie Jodouin and Daniel Enright.