September 20, 2017

Helping PR pros make smarter decisions

Man vs. Machine

Man vs. Machine

Nathan Burke’s ‘The Buzz on Buzz Monitoring’ article raises a couple of interesting issues: how accurate is the data provided by brand monitoring services and what should a brand do with that data once it has been gathered? What I’ve found in my work as Chief Technology Officer at Converseon is that those two questions are actually more related than they might first seem.
If the goal of brand monitoring is only to arrive at some sort of a “sentiment score” related to a brand’s conversation, it’s actually not outside the ability of a machine-based analytics system to pull some fairly good data about how people feel about a brand, despite the difficulties software has understanding sarcasm and even context in some cases. But it’s already obvious that setting that goal is a dead-end; what exactly do you do if you discover, for instance, that people have become 20 percent more negative about your brand over the previous three months? The only answer is to do more research to find out why they’ve grown more negative, and then decide the best way to respond. Thus the real goal here is to understand and engage in the conversation, not just measure the top layer of sentiment or influence; and the only way to accomplish that is to make sure the data you’re gathering is as detailed and accurate as possible – being able to tell, for instance, that the 20 percent drop was due to frustration over the high price and lack of color options in a new product launch. And that gets us back to the original issue: can current technology replace the analysis a human can provide when monitoring conversation?
Nathan mentioned sarcasm as one of the “problems” that machine analytics cannot solve, but there are many more at this stage in the technology, here are a few:
– Social media contains misspellings, poor grammar and punctuation, and colloquialisms; while these things are easily understood by a human, machines have very little ability to handle them
– Each brand conversation takes place within a larger context of products, issues, and terms associated with the niche that brand occupies; humans can quickly and easily learn this context and understand meaning, while machines need to be taught each time for each client, and furthermore have no ability to learn new terms or context on the fly (which is very important for emerging topics)
– Humans are very good at understanding that a long blog post titled “iPhone Review” will contain mostly information about an iPhone, though other products may be mentioned and compared throughout; making accurate inferences when the object of discussion can be far removed from the descriptive text is something machines cannot do well
– Perhaps most importantly, machines have a very difficult time accurately identifying the issues being discussed about a brand or product (color and price, for instance), particularly in unstructured text like social media; and without that level of analysis there is no possibility of actionability based on the data
Despite these problems, most brand monitoring services use machine analytics, hoping to balance the inaccuracies that result with raw scale, all the time accepting the fact that human analysis is the gold standard by which any machine solution is measured.
Until the technology improves, we believe our approach at Converseon is an intriguing way to solve this problem: create a way to scale human analysis, thus avoiding the accuracy vs. scalability problem altogether. Rather than investing one time in an analytics technology, we’re constantly improving the systems that support our human analysts instead, all the while refining the machine analytics that both handle the simpler data sets as well as pre-code the datasets our human analysts process.
Our firm belief is that there is no better way to analyze social media – a system constructed of human contribution on a layer of distributed technology – than by employing a solution that operates in the same way. As analytics technology improves, we will continue to incorporate it into this system rather than replace humans, letting the technology help scale the solution while humans continue to do the detailed analysis.
*Jeff Doak is the Chief Technology Officer at social media agency [Converseon](http://converseon.com). Jeff is responsible for all technology and software products related to Converseon’s services, including their proprietary Conversation Miner service and various customized social media software platforms.*

Ad Block 728

About The Author

Related posts

Ad Block 728
0 Shares