Brandwatch

Brandwatch tracks your brand quickly and clearly, and then goes into detail - it tells you who is saying what about your brand and why. Where they are saying it and when.

It lets you see the context in which your brand is being mentioned and finds the issues raised in connection with it. Which topics are gaining strength and relevance? It can highlight who and what is influential, pick up the first signs of new markets, and draw out target audiences, profiles, and demographics.

Use brandwatch to gain an instant view of the public perception of your brand and compare it with other brands. Monitor how successful a campaign is. Find out immediately what the web thinks about your new product launch or whether your endorsement is a hit or miss.

Brandwatch can generate data for any brand across the entire internet, fast. We automate the process that a team working 24/7 would have had to do manually. It saves money and time and eliminates human bias. Our technology is transparent. And because we own it we can develop it in new directions. There are no illusions. It has been developed over time, by top engineers into a robust, simple to use system.

To see how Brandwatch works, take a look at our quick online demo. For further information on Brandwatch, please contact us at info at magpie dot net.

More about the technology behind Brandwatch

By Giles Palmer
Managing Director, Magpie

We group brands together into industries to help people compare the performance of two or more similar brands. Let's say, for example, we want to look at the online performance of brands in the health and beauty industry. What we need to start with is a (long) list of websites, blogs and forums that are important for the health and beauty sector as well as a list of health and beauty brands, companies and products.

Armed with this, the Magpie web crawler visits these sites daily (sometimes several times a day) and picks up the new pages. On average we are finding that we are picking up almost 2 new pages each day for each website we crawl

When we have the data, we analyse it and make sense of it.

The first thing we do is match each of the new pages we have found with our list of brands and products. Some of these can be done in a straightforward way with key word matching - for example a unique brand name like 'Pantene' will return high quality results, whereas a generic name like 'Boots' (the chemist) won't. One of the downsides to having a generic word as your brand name is that it is hard to isolate the discussions around your brand from other uses of the word. So we have to use some different matching techniques for difficult brand names.

Once we have the matches, or mentions, we do some further analysis on the sentiment being expressed, as well as try to understand the topic or theme that is being discussed.

Enter the world of statistics and automatic classifiers! The way they work is basically as clever comparison systems. We train the system using lots of examples of stories from the health and beauty sector, previously classified by humans as being positive, negative or neutral. The system uses these to learn about positive, negative and neutral stories in the industry so that when we pass it new mentions it can compare them to what it has learned from its training and make a good guess at their sentiment.

To get anything meaningful from our sentiment analyser, we need to pass into it only the bit of a text that refers to a particular brand....and that's not easy. We have developed a set of rules around page structure, sentence structure, occurrences of the particular brand and separation from other brands within the page to try to pull out the right bit of the mention automatically. We'll never be perfect at this as even two people are likely to disagree about which words in a piece are talking about one brand in particular, but we are getting pretty good at it. Rather, our system is getting pretty good at it.

We now have the nuggety bit of text which is hopefully the bit talking only about our brand, but it also contains lots of statistically noisy words like the, of, if, and, then and so on. So the system cleans these out and we are left with text which is probably meaningless to you and I, but our machines love it!

Remember we train our sentiment classifier. Well, the language that is used to talk about health and beauty brands and products is, as it turns out, very different to that used with Mortgages and Insurance. What this means is that to get good accuracy on sentiment for both of these industries, we need to train two different classifiers. Furthermore blogs and news sites use different languages too. So we end up with multiple classifiers.

As well as sentiment, we measure the credibility of a site. A very negative mention from a site that nobody reads is not as important as one from the New York Times for example, so we want to take this into account. The system uses properties of the individual pages and the websites they come from, to judge how credible it is within the industry.

Once we have matched a brand to an online mention and assessed its credibility and sentiment we create a score for that mention. The scores range from +10 for a positive mention on a very high credibility site to -10 for a negative mention from a high credibility site. A neutral mention from a low or very low credibility site would score a 0. We do this every day and at the end of each day, the system updates all the scores for each brand.

The final piece in the jigsaw is to build a website that allows people to access this data in as rich but simple a fashion as possible.

We hope you like it.