On Automated Sentiment Analysis
Thursday, December 10, 2009 | 46 Comments
Tags: automated sentiment, automated sentiment analysis, radian6, Sentiment analysis, Social Media, Social Media Monitoring, social web, Technology
Posted in: Community, Features, Listening, Platform, Social Media, Social Media Monitoring
One of the most talked about capabilities of social media monitoring platforms is sentiment analysis, more specifically the automation of same. It’s a technology that’s important and can be valuable to companies’ social media analysis, but it’s critical to understand how it works, when it’s useful, and what its limitations are.
At Radian6, we’ll be publicly releasing our automated sentiment capabilities inside the platform early next week, and they’ll be immediately available to all current and new customers. As a bit of background, however, we thought we’d talk a bit about what we see as the role of automated sentiment in social media monitoring and engagement, and how Radian6 is approaching it.
What is Automated Sentiment technology, and How Does Radian6 Use It?
Automated sentiment analysis is a system for automatically determining the sentiment of a sentence or phrase. Sentiment refers to the thought or mood of a post and can be either positive, neutral or negative.
Radian6 automated sentiment reviews on-topic posts as they come in, determines the sentiment of the post at the sentence level, and aggregates a positive, negative, or neutral designation at the post level based on specified sentiment keywords and phrases. If a particular document or post touches multiple topics, sentiment can be determined for each separate topic.
Stay tuned on the Radian6 blog next week for more detail on Radian6’s automated sentiment capabilities.
What Automated Sentiment Can Help With
As a first pass, automated sentiment analysis can help streamline the workflow of processing a high volume of posts by providing preliminary determinations of sentiment for each post. Users can then follow up with review and manual adjustment as necessary. Automated sentiment also provides an initial snapshot of postive-negative-neutral ratios, and can help identify trends at a macro level such as sparklines or aggregate changes in sentiment over time.
Looking at ratios of positive to negative sentiment over time can sometimes indicate collective brand preferences as expressed online, or the overall mindset or mood of audiences. The unfiltered and unedited nature of the opinions expressed on the social web and tracked through sentiment analysis can sometimes offer a more realistic, less clinical view of how customers and communities are responding to companies and brands.
Armed with this high level analysis and trend information, Radian6 users can better craft engagement strategy, understand hot button issues and topics around their brand, and reach out to their customers informed about the pulse of opinion about the company and it’s work.
Automated Sentiment and The Human Factor
Sentiment analysis, whether automated or manual, is a subjective process and always needs to be considered in the context of business goals.
What’s read as positive for one person or in one context might be considered neutral for another, so businesses need to consider and outline criteria for positive, negative, and neutral definitions based on their goals for online presence and engagement.
In addition, the complexity and nuance of the English language combined with available technologies for text analytics means that sentiment analysis cannot currently achieve 100% accuracy. In fact, accuracy rates across sentiment analysis engines can be highly variable, as the criteria to define an “accurate” sentiment determination is also somewhat dependent upon human interpretation and context.
There will always be a need for human review and involvement to verify automated results, and ensure that sentiment levels are tagged within the context of individual and unique business goals and agreed upon criteria.
For More on Radian6 Sentiment Analysis…
Stay tuned for a post early next week with some additional detail about the technical features of Radian6’s automated sentiment, and information on how to set it up inside your Radian6 dashboard. And as always, if you have questions or feedback for our team regarding this or any other feature, we’d welcome your input and conversation.
Image by richkidsunite
![Reblog this post [with Zemanta]](http://img.zemanta.com/reblog_e.png?x-id=e0c4933d-e685-444a-96bc-977a418c28fc)
46 Responses to “On Automated Sentiment Analysis”
Amber Naslund on December 10th, 2009 at 4:40 pm
Hi Joe –
Not right out of the gate, no. But user adjustments and refinements will be tagged and captured in the system, and we’re talking about ways to use that information to help the system “learn” better in the future. Stay tuned for developments on that front. It’s definitely feasible, the trick (as with anything) is in doing it well, and accurately.
As Chris talks about, your clients can find value in automated sentiment as a first pass to trim workflow time, but ultimately they’re the determinants of how sentiment is tracked specifically for them. As they get accustomed to it, keep us posted on feedback!
jonnie Jensen on December 10th, 2009 at 6:17 pm
Automated sentiment is vital. The volume of content is impossible to manage otherwise. I am glad to see Radian6 introducing this. Without a whole level of additional cost and resource was required, not to mention the impact on time.
Human involvement is still vital. Automated sentiment is no good if it is not being managed and monitored – one man’s bad is another man’s good, as they say. This watching role helps improve the effectiveness of the sentiment scoring, without it having to be watched all the time.
I look forward to hearing more about how it works….
Evan Urbania on December 10th, 2009 at 6:20 pm
This is exciting. This is the next frontier in SMS. We have been waiting for this and can’t wait to roll it out with some of our government clients and political campaigns.
Bruce Eric Anderson on December 11th, 2009 at 11:51 am
Chris: thanks for the update on the upcoming feature. I agree with Jonnie’s comment, automating sentiment of content is critical, even if it’s not 100 percent accurate. From a competitive landscape, if five different brands are all looked at globally, at least the sentiment of all the discussion on those brands is an apples-to-apples comparison.
Looking forward to a demo of the tool.
@bruceeric
Edelman Digital
Austin, Texas
Simon Harries on December 14th, 2009 at 2:26 am
hi
This is really exciting. I’m currently using another tool with sentiment analysis. Have you been developed this in house or are you partnering with another provider
Simon
Barika CCroom on December 14th, 2009 at 3:28 pm
It would be really great to have an added feature to this tool that would automatically calculate the sentiment ratio for you based on your adjustments. But that’s just icing on the cake. Great tool and I love using it already and its my first day!
Barika
Arun on December 17th, 2009 at 2:21 pm
Hi
Thanks for the update. What will be the accuracy of the sentiment analysis? Other vendors talk of the high 70s to high 80s. Where does r6 stand in this regard?
Prabhu on December 17th, 2009 at 3:35 pm
Nice to know that R6 does automated SA now.
Wondering if sentiment is gauged based solely on the language or taking into account who it is directed against: the client or it’s competition.
Thanks.
Ryan Strynatka on December 17th, 2009 at 4:07 pm
Hi Arun: It doesn’t necessarily make sense to provide an accuracy number because it is so variable. For example, a topic that lends itself to sarcasm and irony will have a lower accuracy assessment compared to a topic that evokes strong emotions (e.g. a customer service disaster). Furthermore, even humans will disagree on post sentiment because many posts appear to display some sort of sentiment – but it may be expressed in an ambiguous manner.
Hi Prabhu – great question! One of the aspects of our implementation is that you can tune it based on “Sentiment Subject,” so it is indeed possible to check out what the sentiment is toward your competition as well. More details on this coming soon.
Prabhu on December 17th, 2009 at 4:28 pm
Thanks for fielding that, Ryan !
I have used Radian6 and built my own sentiment analysis module , hence the interest.
I was able to pit client versus competition using my approach. I found out that yielded better results. I took double negatives into account using a single regular expression and also do anaphora resolution (traversing all the way to earlier sentences until the first reference is obtained).
Let me know if you are interested.
Ef Rodriguez on December 17th, 2009 at 11:59 pm
Very cool!
I wish I had something more eloquent than that, but that’s all I’ve got at this hour.
LadyV on December 18th, 2009 at 12:22 pm
Hi,
I was wondering: is the SA available only for english or for other languages as well?
Ryan Strynatka on December 18th, 2009 at 12:36 pm
Hi LadyV: we certainly recognize the global need, but being new technology the initial focus is on English language support.
Ca Mortgage on February 1st, 2010 at 3:26 am
well I must admit you are correct, Thank for an eye opener article.
Movers in Kentucky on March 11th, 2010 at 12:17 pm
Tweetbacks
AmberCadabra (Amber Naslund) on December 10th, 2009 at 6:07 pm
Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
chieflemonhead (Judi Samuels) on December 10th, 2009 at 6:10 pm
RT @AmberCadabra: Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
TomHasselman (Tom Hasselman) on December 10th, 2009 at 6:11 pm
RT @AmberCadabra: Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
DawnEva (Dawn Wilcox) on December 10th, 2009 at 6:12 pm
RT @AmberCadabra Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
EAAustin (Eleanor Austin) on December 10th, 2009 at 6:15 pm
Cool stuff …RT@AmberCadabra Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
twebinars (Radian6 Twebinars) on December 10th, 2009 at 6:16 pm
RT @AmberCadabra Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
chrisramsey (Chris Ramsey) on December 10th, 2009 at 6:35 pm
Radian6 announces Automated Sentiment Analysis in its social media monitoring measurement and engagement platform http://tinyurl.com/yeoe9aq
PatrickRedknap (PatrickRedknap) on December 10th, 2009 at 6:37 pm
RT @chrisramsey: Radian6 announces Automated Sentiment Analysis in its social media monitoring platform http://tinyurl.com/yeoe9aq
bitpakkit (Ben Watson) on December 10th, 2009 at 6:39 pm
RT @chrisramsey: Radian6 announces Automated Sentiment Analysis in its SM monitoring measurement platform http://tinyurl.com/yeoe9aq
andreatarrant (Andrea Tarrant) on December 10th, 2009 at 6:40 pm
Radian6 announces Automated Sentiment Analysis in its social media monitoring measurement and engagement platform http://tinyurl.com/yeoe9aq
tmrichardson (Tina Richardson) on December 10th, 2009 at 6:44 pm
RT@AmberCadabra Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
jamespoulter (James Poulter) on December 10th, 2009 at 6:53 pm
RT @chrisramsey: Radian6 announces Automated Sentiment Analysis http://tinyurl.com/yeoe9aq
mmanuel (Mike Manuel) on December 10th, 2009 at 6:56 pm
@radian6 congrats guys, sentiment analysis is one hellofa nut to crack, but glad to see you’re working on it: http://tinyurl.com/yeoe9aq
JTitus (JTitus) on December 10th, 2009 at 6:58 pm
RT @jamespoulter: RT @chrisramsey: Radian6 announces Automated Sentiment Analysis http://tinyurl.com/yeoe9aq
GrahamWolfe (Graham Wolfe) on December 10th, 2009 at 6:59 pm
RT @AmberCadabra: Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
Chaf (Chafic Haddad) on December 10th, 2009 at 7:05 pm
RT @TomHasselman: RT @AmberCadabra: Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
billrobbCisco (Bill Robb) on December 10th, 2009 at 7:20 pm
RT @mmanuel: @radian6 sentiment analysis is one hellofa nut to crack, but glad to see you’re working on it: http://tinyurl.com/yeoe9aq
jordanlutes (Jordan Lutes) on December 10th, 2009 at 7:32 pm
RT @chrisramsey: Radian6 announces Automated Sentiment Analysis in its social media monitoring platform http://tinyurl.com/yeoe9aq
plgraham (Patti Graham) on December 10th, 2009 at 7:43 pm
RT@AmberCadabra Radian6 and our CTO @cdnewt on automated sentiment analysis: http://bit.ly/4tbPVk
TweetRich (Richard Mcinnis) on December 10th, 2009 at 11:50 pm
What are your thoughts on auto sentiment analysis in social media Radian6 CTO @cdnewt shares some thoughts: http://bit.ly/4tbPVk
Tecnoetica (Davide) on December 15th, 2009 at 5:05 am
RT @chrisramsey: Radian6 announces Automated Sentiment Analysis in its social media monitoring platform http://tinyurl.com/yeoe9aq
aldeneaton (Mike Eaton) on December 17th, 2009 at 6:22 pm
Radian6 will be releasing automated sentiment analysis soon; should be interesting http://bit.ly/4tbPVk
u2elan (Tim Sears) on December 17th, 2009 at 6:40 pm
Radian 6 to start using auto-sentiment. Surprised they are not using a machine-learning technique. http://bit.ly/4tbPVk
Siegelo (Lori Siegel) on December 17th, 2009 at 6:52 pm
RT @aldeneaton: Radian6 will be releasing automated sentiment analysis soon; should be interesting http://bit.ly/4tbPVk
MattSF (Matt Nagel) on December 17th, 2009 at 7:18 pm
Radian6 is introducing automated sentiment analysis: http://www.radian6.com/blog/2009/12/on-automated-sentiment-analysis/
rdublife (rick wion) on December 17th, 2009 at 9:07 pm
checking out the new automated sentiment features on Radian6 http://www.radian6.com/blog/2009/12/on-automated-sentiment-analysis/
raidious (raidious) on December 18th, 2009 at 2:58 pm
@Radian6 Automated Sentiment Analysis = awesome. http://www.radian6.com/blog/2009/12/on-automated-sentiment-analysis/
Joe Boughner on December 10th, 2009 at 2:40 pm
Will the system fine tune itself based on end-user adjustments? Or is there some way for users to have a say in the configuration of the sentiment analysis (similar to the way users can tune the influence sliders)?
I’ve never played with an automated sentiment analysis tool before so I don’t know if such things would even be feasible. I know my clients were pleased to hear they had input in influence weightings; some may be wary (as I am) of automated sentiment analysis.