One of the primary determinants of a reputation management issue is the buzz about a company / product / person. If the buzz in primarily positive, then the chances are that the online reputation is fine (although the negatives still have to be examined to determine whether they have the potential to overwhelm the positives in the future). If the buzz is negative, well then, there's an issue or multiple issues that need to be resolved.
Gathering the data can be done by a number of tools, but how do you assess the actual sentiment of each entry? Can it be done programmatically? Sure it can, to a point, but most programs can't do that good a job at sentiment analysis. Sure there are certain key phrases that can be looked for to determine positive or negative, with a "not" or "doesn't", etc reversing the sentiment. But how well can a program determine the difference between
"The biggest disaster movie since Titanic" and "The biggest disaster since Ishtar"?
Other issues include:
- Colloquialisms - how many sentiment analysis programs know whether "That movie was pants" (UK) is positive or negative? (for the record it's negative)
- Sarcasm - "Oh sure, I truly believe that the Eagles will win the superbowl this season *roll eyes*".
- Trolling - going into an Eagles fansite and writing the above statement, which will then unleash a stream of negativity towards the troll, not the team.
As an example, let me show you some results from the Summize sentiment analysis tool that ran through twitter posts and ranks them based on content (Summize was recently purchased by Twitter, and this tool does not appear to currently be available). Posts for the last 30 days are then aggregated and a score presented to show the overall sentiment for a keyword. However, if you look at this tool, sentiment on the current President is 'great' (the same score as Satan), and the Iraq war is swell, which ranks it as having the same sentiment as SEO. So either the approval polls are wrong, the majority of the audience that uses twitter is part of the 21% that still believes that the current President is doing a 'bang up job', while also approving of Satan, or the sentiment analysis just isn't cutting it.
So how can you determine sentiment if the automatic systems aren't up to the job? That's right... manually. A person is the best equipped to look at a comment, review the context, and determine sentiment. It takes time, and some of the results are still going to be subjective, as the written word sometimes isn't as forthcoming as it could be (depending on the author and the level of obfuscation that they manage to achieve either intentionally or unintentionally), but overall you're going to end up with the clearest picture of the landscape surrounding your keyword.
Oh, and just to throw a bone to Summize sentiment analysis, they did get one thing right, ask either of the people who saw "Meet Dave" (should they ever admit it). ;)