Sentiment Mining

Sentiment mining is the application of text mining to infer the sentiment expressed in a piece of text about a chosen target.

Generally the source texts are short texts such as web log items or the 'tweets' created on Twitter.

Generally also the sentiments looked for are restricted to negative, neutral and positive. In this way it is fairly simple to create a popularity index for celebrities or products.

We've implemented sentiment mining using a combination of NLP and conventional text mining techniques. After pre-processing the data to isolate text associated with the target, and to handle negations within sentences the remaining processing is performed using XmlMiner Naive Bayes text mining algorithm.

A vital part of using such an algorithm is the development of a training set. This is a large set of examples of negative, positive and neutral texts.

The success of the sentiment mining algorithm is very much dependant on the quality of this set and how well it is marked up.

Our implementation makes it possible to improve performance by enhancing the set to cope with new usages and retraining.