I’m reluctant to throw another stink-bomb into the debate about sentiment analysis, but I have a view of the tech that I don’t often see from others (I’ve worked extensively with sentiment analysis tech, in both development, while at Infonic, and as a power user of dozens of enterprise social media monitoring tools at The Conversation Group).
So here they are: the two reasons I think sentiment analysis automation has no place in social media monitoring:
- Granularity. Here’s the reason that people often find that their experiences of sentiment accuracy don’t match the claims on the boxes: training and testing of the tech is done against a brief that doesn’t match the typical user needs. It used to be that testing (and hence reported accuracy) was done at a document level (ie the sentiment of a piece of text as a whole), not at an entity level (ie the sentiment towards the brand, person, or concept that a user is trying to track). I suspect it still is, but because vendors are so fluffy (defensive?) in the way they market sentiment analysis, I can’t tell you for sure. Reasonable expectation is that the sentiment reported by a platform relates to the keyword(s) or search string in which sentiment is reported, not just towards the aggregate sentiment in the whole document returned by that search. For typical users of enterprise social media monitoring tools, it’s hard to see how document level sentiment is of any use at all – it makes the score given to any multi-entity documents not just inaccurate, but probably actively misleading (and rules out comparison of sentiment towards competing brands). I guess monopolies ought to be good to go though
- Fuzzy definitions. Here the problem is on the client side. As it’s currently sold in enterprise applications, sentiment analysis means something quite specific – in a nutshell the presence of words and phrases deemed culturally positive or negative in a body of text, and a judgement of the overall tendency within that body of text. It feels like customers of monitoring platforms frequently have a slight cognitive blind-spot: they see “positive” and read “that which benefits us”; they see “negative” and read “that which harms us”. But as anyone working in comms worth their salt knows, those two concepts really aren’t the same thing at all, or press embargoes wouldn’t exist and Nudge would never have been written
This isn’t to say that I don’t believe in sentiment analysis of social media – I do, in fact I can’t think of a recent case in which I haven’t recommended that sentiment is tracked as part of a core set of metrics. I just believe that technology doesn’t cut it for enterprise social media search applications, and I’m sceptical that it ever will. It’s simply too context-dependent to work for every (any?) individual case straight from the box. I’m supportive of tools like ScoutLabs, which include a user-override for sentiment scores, so it can be used a production console for human coding, but I wish the machine generated sentiment score wasn’t there to start with.
So where is it useful? In high volume, high stakes text mining, where the potential payoff of doing-clever-stuff-with-data can justify the huge effort needed to massage and contextualise the data and the technology to suit specific project circumstances. The technology was originally commercialised for use in HFTS type applications after all, not marcomms and customer service support work.
Pingback
by monitoring the media monitoring
13 Mar 2010 at 02:58
[...] More here: Why automated sentiment analysis shouldn't feature in social media … [...]
Trackback
by uberVU - social comments
14 Mar 2010 at 09:30
Social comments and analytics for this post…
This post was mentioned on Twitter by christhomasuk: Just posted: Why automated sentiment analysis shouldn’t feature in social media monitoring tools (http://bit.ly/aNYDdv)…
by Synthesio
16 Mar 2010 at 13:01
Couldn’t agree with you more, Chris. Nothing replaces human analysis of human behavior. Whether companies analyze social media monitoring data in-house or out-source it, there should be someone – if not a team – dedicated to analyzing the information at various levels to make it relevant to the company’s strategies and objectives.
Nathan Gilliatt actually just posted a great metaphor for automated sentiment analysis comparing it to a mood ring ; it’s the shiny new feature that everyone wants but not everyone knows why or how to use it. Like any other form of social media monitoring analysis, it is only as relevant as the qualitative information behind it.
It’s been a while since we last caught up, but I’m glad this caught my eye!
Best,
Michelle @Synthesio
by Chris Thomas
17 Mar 2010 at 12:06
Hey Michelle, thanks for the comment.
It’s an interesting metaphor (although I had to resort to Wikipedia to find out what a mood ring is
).
I frame the issue slightly differently. Human sentiment analysis is a very established part of traditional media content analysis (around since the 70s / 80s). This is what many customers – particularly the ones with a PR / comms background – think they are getting: an automated; cheap; effective substitute for expensive human content analysis.
But as you guys know, it’s not the same thing at all – automated sentiment analysis was developed to do something different, follows a different logic path and produces very different results.
by Maria Ogneva
17 Mar 2010 at 22:12
Thanks for starting this discussion. Sentiment analysis is a hot topic for sure. Both human and automated sentiment have their place.
I mostly wanted to address what you say in #1 re: vendors not disclosing if their analysis is at a topic or article level. I can’t speak for other vendors, but at Biz360, we do both, and our accuracy is slightly higher at a topic (aka entity) level analysis. I also think that if you are a brand, topic-level analysis is a lot more useful, because there are more articles that deal with several brands at the same time. Our topic level accuracy is 65%, while 62% accuracy at article level (we have 4 sentiments BTW vs. some other vendors’ 3). We trained our data via Mechanical Turk until incremental increase in accuracy was <1%. Users have a chance to override sentiment, as well as toss an article or a source altogether from your results.
Automated sentiment is useful when you have 50,000 articles – you will never be able to read them all! If you want sentiment of an individual article – well, you should probably read it
Cheers!
@themaria @biz360
by Nathan Gilliatt
18 Mar 2010 at 01:57
Y’know, I wondered how many people wouldn’t know about mood rings–so I linked to the wikipedia article, too.
Actually, my point in using the mood ring metaphor is to say that sentiment is more interested as a filter than as a standalone metric. Watching the sentiment line go up and down is like watching a mood ring change color–interesting, but lacking in depth.
http://net-savvy.com/executive/measurement/sentiment-analysis-is-not-a-mood-ring.html
by Mark Evans
19 Mar 2010 at 14:48
Chris,
I think there is a role for automated sentiment within social media monitoring given the amount of data/conversations that need to be tracked and reviewed. It’s important to remember that it’s still early days for sentiment technology so its reliability and accurateness will improve over time.
At the same time, social media monitoring services should still offer the ability for human editing. Within Sysomos’ MAP service, for example, users can manually adjust sentiment settings.
Mark
(I’m the director of communications with Sysomos.com)
by Chris Thomas
19 Mar 2010 at 16:04
Thanks for the input! Having a few more (and more constructive thoughts) about where I think the tech should go. Will post again over the w’end.
Pingback
by 5 Improvements for SMM Sentiment Analysis « Wholly Social
20 Mar 2010 at 10:59
[...] was pretty down on the tech in my last post. I had a few reasons, mainly wrapped up in the fact that I don’t think it does a good [...]