Subscribe: Comments on: Blog influence: talking with Flemming Madsen of Onalytica
http://brendancooper.com/2007/06/11/blog-influence-talking-with-flemming-madsen-of-onalytica/feed/
Added By: Feedage Forager Feedage Grade B rated
Language: English
Tags:
analysis  automatic analysis  automatic  bias  blog cites  blog  blogs  cites blog  don  human  microsoft  onalytica  search  sentiment 
Rate this Feed
Rate this feedRate this feedRate this feedRate this feedRate this feed
Rate this feed 1 starRate this feed 2 starRate this feed 3 starRate this feed 4 starRate this feed 5 star

Comments (0)

Feed Details and Statistics Feed Statistics
Preview: Comments on: Blog influence: talking with Flemming Madsen of Onalytica

Comments on: Blog influence: talking with Flemming Madsen of Onalytica



Content, copywriting and strategy from Brendan Cooper, your friendly neighbourhood copywriter and content strategist



Last Build Date: Tue, 03 Mar 2015 09:10:53 +0000

 



By: The Friendly Ghost

Fri, 21 Sep 2007 18:31:08 +0000

I like a good ramble! :)



By: Stephen Davies

Thu, 14 Jun 2007 00:25:34 +0000

"But it’s not just direct influence - think about the indirect. Blog A cites Blog B is direct. But if Blog A cites blog B, who has cited Blog C, then blog A is indirectly citing blog C." Yeah I get your point but for all the millions of blogs there'll be a tiny network in each specific area that work like that. And any company who's clued up on this stuff will know who these key influencers are. But really, do you think there are a load of bloggers talking about all brands in detail and frequently? It's very tech focused. "how would you reference more nebulous concepts such as, say, ‘politics’ or ‘environment’?" I can't see this words as search terms serving any purpose as they're too general. Sure, you'll pull up a lot of results but what would you be looking for? The general consensus of politics is in the blogosphere? (which is made up of many nationalities and thus many political parties). Just don't see what the end result would achieve. "It’s not about what’s being said, but who says it, and in a similar way to the Google referencing algorithm, the software will rank citations according to influence. It really is a massive exercise in how outputs route back into inputs." My point is this. If you have, for example, a MySpace blog which says: "I much prefer Gmail to Hotmail. Microsoft sucks" isn't going to really mean a whole lot in the grand scheme of things because MySpace blogs aren't 'findable' in a search engine query; they aren't read by a whole lot of people (unless you're Tila Tequila); and they're only available in a blog search engine for a limited period. And the sheer number of these types of blogs is huge. The signal to noise... No, I mean, the noise to signal ratio is huge. It's very much Noise > Signal. Plus the fact that this is just a throw away comment and there is no depth, argument, revelation etc etc. "why should Onalytica’s systems be any different?" Sorry wasn't suggesting that Onalytica was being naughty. I just meant that I would love to see a search tool that could analyse a 800 word blog post for opinion, argument, bias, sentiment etc etc. The OpinMind url I linked to that only measures sentiment by the words that go along with the search query. E.g. Microsoft **sucks** or I **hate** Microsoft or I **love** my new Microsoft Zune or Microsoft Office 2007 is **great**. Anyway, just rambling now. :-)



By: The Friendly Ghost

Tue, 12 Jun 2007 22:35:13 +0000

You wouldn’t have to because no brand (excluding Microsoft, Apple and iPod) is being mentioned thousands of times a day online. But it's not just direct influence - think about the indirect. Blog A cites Blog B is direct. But if Blog A cites blog B, who has cited Blog C, then blog A is indirectly citing blog C. It's complicated. Furthermore there will be several terms for each topic. Microsoft, Apple or Ipod could be fairly explicit but how would you reference more nebulous concepts such as, say, 'politics' or 'environment'? It grows exponentially, so really we would be talking about thousands of documents. And don't forget, Onalytica will at any time be running analyses for many clients, each with many widely varied and diverse topics. But still, in my opinion, not all of these blogs matter because a) they have no audience and b) they don’t appear in search results. There's the trick. It's not about what's being said, but who says it, and in a similar way to the Google referencing algorithm, the software will rank citations according to influence. It really is a massive exercise in how outputs route back into inputs. But the programming in the automatic analysis has been written by a human. Like you say, humans have a subconscious bias. Hmmm, not sure that you could program bias into an algorithm? I really don't know how the Onalytica algorithms work but I expect it's a combination of recognising keywords in isolation, key phrases, semantically-linked phrases and so on (and to be frank I am still amazed that software can do this so I get where you're coming from). A programmer working on a financial system would not be able to bias the trader toward investing in a certain stock or type of stock (unless he was extremely naughty) - why should Onalytica's systems be any different?



By: Stephen Davies

Tue, 12 Jun 2007 10:49:30 +0000

FG said: "Human analysis is better but how would you cost-effectively analyse thousands of documents a day?" You wouldn't have to because no brand (excluding Microsoft, Apple and iPod) is being mentioned thousands of times a day online. For e.g Adidas, a global consumer brand was mentioned 241 times from Mon 11 June to Tue 12 June according to BlogPulse: http://tinyurl.com/258gpq Plus, BlogPulse pulls up loads of crap and duplicate posts. I know I've had to sift through it all before. But still, in my opinion, not all of these blogs matter because a) they have no audience and b) they don't appear in search results: http://www.prblogger.com/2007/06/not-all-blogs-matter-part-ii/ FG: "Automatic analysis achieves over 90% accuracy and whereas I agree that humans would be better, for the purposes of mass analysis well, it has to be automatic sentimenting all the way." But the programming in the automatic analysis has been written by a human. Like you say, humans have a subconscious bias. I agree though - what's positive/negative to one may not be to another. FG: "Say I harbour a secret desire for David Cameron (who doesn’t?)" Me Don't get me wrong. Companies need to WHAT the sentiment is but, more importantly, they need to know WHY too. And the why takes human analysis in my experience. It's a very interesting subject and Flemming probably knows a lot more than myself about it. Would love to know what he thinks.



By: The Friendly Ghost

Tue, 12 Jun 2007 07:55:57 +0000

Human analysis is better but how would you cost-effectively analyse thousands of documents a day? Automatic analysis achieves over 90% accuracy and whereas I agree that humans would be better, for the purposes of mass analysis well, it has to be automatic sentimenting all the way. Sure, automatic analysis might 'pop' some issues across from 'slightly negative' to 'negative', but it would be unlikely to swing from negative to positive. It's a 'good enough' solution. Plus, humans will introduce bias. Say I harbour a secret desire for David Cameron (who doesn't?), and I'm sentimenting political documents. Subconscously I could bias my results because of this. Then there are other factors such as my age, sex, nationality and so on. When you introduce automation, you immediately level that playing field. Also, I really don't know about the language issue and that's a good point you're making - and I wish I'd asked Flemming about it! I wouldn't want to second-guess Onalytica's approach but one approach could be to analyse, say, Western European languages as there is already a precedent in automatic translators such as AltaVista's Babel Fish and Voice to Screen solutions such as SpinVox. So can sentiment be tracked in Japanese or, more importantly in a few years' time, Chinese? And all the variants thereof? If Flemming's reading this then you never know, he might tell us...



By: Stephen Davies

Mon, 11 Jun 2007 22:43:00 +0000

FG said: "At the outset I had in the back of my mind that the information-gathering and evaluation process must of necessity include manual intervention, much in the same way some spam-filtering techniques work. Not at all. Algorithms really can identify sentiment in articles to a high degree of accuracy and - of crucial importance - impartially and universally." But can they (algorithms) identify the specific issues within the sentiment which are causing the discussion? E.g. 30% of all the online discussion relating product X was to do with the battery quality. Within this 30% - 70% was of negative tone with only 10% positive. The remaining 20% was of neutral tone. And 40% of the online discussion on product X was to do with the price. etc etc. There are tools out there that measure sentiment - www.opinmind.com being one - but if you want to get into the heart of the specific issues that are causing online chatter then it's human analysis all the way. Then there's the point that the blogosphere is made up of many languages. The English language only accounts for 34% (or there abouts) according to Technorati. Can sentiment be tracked in, say, Japanese?