Over the past few weeks, much of the online conversation about social media monitoring has centered around the importance of pairing automated analysis with human expertise. Two ideas in particular caught our eye: First, humans should participate in data filtering to ensure accuracy and relevance. Second, as Jason Falls stated on Social Media Explorer, humans are needed to interpret and correctly utilize the insights that monitoring tools can provide.
While we agree with that second idea and offer consulting services of our own (as well as have excellent partners that offer strategic services), we’d like to†highlight the first point – that humans are necessary to filter out irrelevant data – with some analysis we’ve done about the conversation surrounding the new Bing search engine. Our algorithm incorporates training by humans for both sentiment and relevance, allowing us to catch things like the following two tweets:
Even though the first tweet could have been recognized by a simple English language filter, the second one contains a common misspelling of the word “being” that a computer would be hard-pressed to identify as such.
Indeed, a post by Marshall Sponder over at webmetricsguru.com makes the point that sentiment analysis is best done by people. At the Sentiment Analysis Symposium, he used several popular fully automated social media monitoring tools to study the conversation about Social Media Week, and obtained drastically different results from each tool.
Our algorithm is different in that it is neither keyword based nor based on natural language processing†– rather, the user teaches the algorithm to replicate human judgment, achieving a true “marriage between man and machine” – something that Social Media Today describes as social media monitoring at its best.
Tagged with