PhilGo20

I’d be hard pressed to guess if social media monitoring tools have finally crossed the chasm but they’ve clearly been riding a solid buzz for the last 18 months. Looking at the client list of Radian6, Sysomos or ScoutLabs  you soon realize that many major brands have jumped on the bandwagon to be able to monitor the buzz and sentiment that their brand, products and competitors generate.

Some voices are questioning the real value and possible ROI of these platforms, sometimes in opposition with free solutions. And it totally make sense when monthly fees easily range in the 500-1000$/month for the main solutions with a decent set-up.

One of these, entititled “The problems with Social Media Monitoring technologies“, seems to have decent web findability as it was sent to me a couple times by friends and clients in the past weeks. I’ll take a moment to go over these particular points one by one.

1. The technology is fairly stupid. It either don’t do what it says on the tin or do it quite badly. It is supposed to be simple: based on keyword configuration, the software scrolls the social web and collect mentions of your keyword(s). It then supposes to analyse themes and influence ranking and sentiment etc, which it does with limited accuracy . For example, will pick up any header, ad sense or footer mentions of your keywords even if it’s in the totally irrelevant context. If your brand name is pretty generic you are in deep sh*t. Hours of configuration and exclusions awaiting you.

You have the same problem when you search with Google. Building up good search queries takes time, analysis, trial & error and a good understanding of the brand, product and industry . Nobody said social media monitoring would be easy or obvious. The web is getting noisier everyday and good queries are more valuable than ever.

Unreliable data. The most important thing to understand is that the software simply provides you with piles of data. Before you can extract anything meaningful from this data you have to go through hours and hours of spam filtering which can be very tedious if you are dealing with 1000s of mentions every week/month. In some occasions I had over 50% irrelevant data coming through my dashboard. Additionally, the spiders cannot access all social spaces and sometimes the most important conversations are blocked.

There’s a clear need to be able to classified results by social engagement, source, language, sentiment and geo-demographics.  There’s spam in Google results as well obviously and you need good tools and techniques to cut through the noise.

Sentiment analysis is flawed. Again, this is part of the limitation of the technology. The software analyses keywords, not human emotions and, on average, the software gets it wrong 30% of the data because human emotions are subtle and complex and not easily categorised by software – we are not there yet.

I agree. And one of the reason is that we are usually measuring the message instead of the effect. Sentiment analysis should be mainly used as an extra filter. 60-70% reliability is better than nothing and probablu much higher for tweets.

Good monitoring tools allows for user to overwrite the results of sentiment analysis. Results needs to be reviewed, but it’s still much faster than evaluating everything manually.

Region specific data: for global brands, social media have very strong global element as well as clear regional bent (forums, blogs, networks etc). This is tricky especially if you are working with a regional client (e.g Huggies UK). Problem is for the software it’s not about where you are but which domain are you using. So reliable geo / regional analysis is, in many cases imposible to carry or not complete so need to be complemented with manual search.

True, Twitter localization is only based on what the user enters as location in but will soon be changed by the new API opt-in feature.
The ability to drill down the query with localization-related keyword helps obviously.

Influence analysis is flawed. Well, the concept of influenced is flawed so of course technologies of measuring it are flawed as well. Similar to sentiment, the technology is just not as clever as they want you to believe. It is based either on bogus metrics or just irrelevant, obsolete ranks .

Agrred, most implementation are. That’s why I always ask before evaluating or using. Sysomos has their own Postrank-style ranking. Social Mention (free) as added PostRank to their results.

If I were to design a full platform, I’d be using Postrank to rank results.

Time consuming. Because all of the above, the reality is that while thess companies provide you with piles of data and funky visualisations the profound unreliability of the software means you have to sit for hours and days and configure the dashboard, refine the data, correct the scores, filter the spam, get rid of irrelevant data AND THEN, AND ONLY THEN you can start making some meaningful analysis.

It does take quite a lot of time to set-up and it’s not easy. After a lengthy initial process, the work involved somewhat decrease but  you will still have to filter out some spam.

The analysis will always take time. That’s why many brands work with analysts to help them make sense of it.

Price. This varies significantly but the fact is that you pay just for the data and license fees to use the software. For the level of service you don’t get value for your money.

I’d expect to have access to an API to integrate with client workflow and platforms or to customize views and results, the ability to query on past period and modify queries at any time.

Conclusion

Social Media Monitoring platforms are not be-all end-all solutions, more of a building block of brand’s modern tool set for effective market research, marketing and customer support. They require investment both in time and money and more often than not, some support from people who have done it before and are comfortable using these particular tools.

Related articles by Zemanta

• How to Create a Share of Voice Report (socialmediatoday.com)
• Do Social Media Marketers Dream Of Monitoring Tools? (threeminds.organic.com)
• Collecta releases API to spur development of real-time search apps (venturebeat.com)
• Twitter Works to Improve Location ID Service (marketingpilgrim.com)
• Google Gives Sentiment Analysis a Multistar Rating (mikemoran.com)

Ranking social media users can be seen as an ego trip, but with communication and information channels growing and evolving so fast, it simply make sense to use a somewhat automated way to filter out users based on their influence. While PostRank gives me a nice way to evaluate social engagement around content, many apps are starting to express the need to use a similar filter to evaluate Twitter users by interest and activity.

Getting a score for a user is not so useful in itself, but is (we’ll be) a building block and filter-classifier of many upcoming tools and services.

I haven’t yet found a tool that fully satisify my needs but here’s the best of what I’ve tried so far.

The Tested Ranking Tools

1. Twitterank
2. Twitter Grader
3. Retweetrank
   

The References

To perform a real test, I needed some kind of references, so I used the following profiles:

The Normal Guy

@PhilGo20 : My own profile. I consider myself a fairly active Twitter user, with an ok following (around 560) and a close to 1:1 follower/following ratio. I retweet, been retweeted a couple times and am mentionned a couple times a week in reply.

The VIP

@SebProvencher : Seb is a Praized.com co-founder (Mtl startup) and Product Development Manager. I like his tweets and read all of them (thxs for TweetDeck group feature). He has close to 1500 followers,  but he’s only following 480 persons, giving him what is seen as an excellent follower/following ratio.

The Popular

@Caterina : Caterina is Cofounder of Flickr and Hunch and a  social media startup star. She has over 9000 followers but just over 200 tweets.

The Not-So Active

@PhilipBoum : Philippe is one of the NOFOLO/Percute Technologies guy in Quebec City, involved in the business side of these web services small companies. I don’t know if he’s using an automated follow back script but he has over 1000 followers with only 14 tweets and have not tweetted once in the last month. I am guessing he has a life outside Twitter, and that’s good !

The Spammer

@entrepreneur949: Sorry if you read this and you are not, but by your tweets and your link, if I were to design a spam tool for Twitter, you would be targeted.

The Inactive (used as a reference for the test)

@philippegauvin : I am squatting my own name domain. Only 2 tweets redirecting to my real account. 0 followers / 0 following.

Each reference is pretty self-explanatory, except the difference between Caterina and SebProvencher. I wanted to have both because I think they have different type of influence. To me most of SebProvencher influence comes from his activity on Twitter while Caterina’s comes from who she is. Without being very active, she has an important following. Praized is not know on the street while Flickr is (pick your street).

The Results (on August 27th 2009)

Retweetrank (0-100 percentile)

• PhilGo20 : 91.4
• SebProvencher : 98.59
  
• Caterina : 98.6
  
• PhilipBoum : 0:0
  
• PhilippeGauvin : 0.0
  
• Entrepreneur949 : 0.0
  

Twitter Grader (0-100 I suspect it’s a percentile)

• PhilGo20 : 95.7
  
• SebProvencher : 99.7
  
• Caterina : 99.4
  
• PhilipBoum : 96
  
• PhilippeGauvin : 25
  
• Entrepreneur949 : 89
  

Twitterank (not normalized, 0-200+)

• PhilGo20 : 16.93
  
• SebProvencher : 94.97
  
• Caterina : 32.17
  
• PhilipBoum : 0
  
• PhilippeGauvin : 0
  
• Entrepreneur949 : 0

First, despite the usefulness of TwitterGrader and some well thought features (I like the tag cloud), it is no yet usable in my sense as it rank inactive and spam account way to high.

Retweetrank does a pretty good job, but it’s very harsh if you haven’t tweetted in the last month. I would think PhilipBoum deserves better than 0. Having read the way they rank users, it seems to be only looking at the last month of activity. There is pros and cons to that approach in my mind. It does a very good job at filtering out Spam account though. What I am wondering is this : Could only one spam supporter account  retweeting the main account be sufficient to put Retweetrank at wrong ? Have not tested it yet. Also, it does not really separate significantly me, Caterina, and SebProvencher. That’s why you cannot be using a percentile as an actual rank.

TwitterRank does a good job too, but also harsh on PhilipBoum account. Maybe that’s the way it should be, but maybe the guy took a month off Twitter ! On the other hand, this is the only tool, that seems to normalize the value (not using the percentile as a rank) and the only one that shows the difference between Caterina and SebProvencher’s influence.

Conclusion

I think TwitterRank is doing the best job by far here and their simple API is also a big plus. There’s is going to be a large number of ranking tool for social media users coming up in the next months, so it’s futile to bet on the future of this service but the need for it is obvious to me. Mash it !