10 December 2007
Matthew Hurst has written an interesting post about sentiment detection at http://datamining.typepad.com/data_mining/2007/12/sentiment-minin.html.
He comments on the differences between automated and manual approaches, SentimentMetrics uses an automated approach and we are currently at an 80% accuracy which is considered good in the industry, however I sometimes get clients asking why something is negative when it should be positive which occasionaly happens.
Maybe we need to consider an human review stage in our service. The reason we haven’t so far is this would result in a delay in getting the data to the clients and also would inflate the cost of the service. At present our clients seem happy with the service but it is definately something to consider for the future.Get a Demo