Sentiment analysis
Sentiment analysis that delivers more than just positive or negative valuations with built-in sentiment scoring, topic identification and categorisation.

Talee of this service is to extract opinions from text. An opinion represents the topic an author is authoring and a sentiment score that classifies how positively or negatively the author feels towards that subject. Deep Linguistic Analysis is used to identify the subject the author is discussing. This could be:

? an entity (brand/ person/product/place?
? a concept (like ?global warming?, ?public policies? or ?financial crisis?).

The sentiment analysis service will also break the opinion right down to detect exactly which features or attributes or components of the subject are increasingly being discussed. For a product this may be the main components or accessories as for example, the ?screen? in ?the screen of the Galaxy Tab? or the ?case? in ?my new iPad case?. For a person this could be the activities or attitudes associated with them. For a place it could be the precise buildings or institutions located there.

When coupled with our categorisation service these features or attributes can be used to place the opinion in a category extracted from a taxonomy. This provides a powerful way to structure a set of texts according to what topics folks are discussing and how they experience those topics.

Sentiment scores are also based on Deep Linguistic Analysis. The more intense the feelings of the writer about the subject, the higher or lower the score. To do this, the analysis detects linguistic features like the strength of the vocabulary or the usage of intensifiers like ?really?, ?very? or ?extremely?. So a comment like ?Installing software on this machine is painful!? will be scored as less negative than ?Installing software on this machine is actually very painful indeed!?

Deep Linguistic Analysis accurately handles complex issues like negation: ?the new Nikon is really not bad at all?.

The service handles complex linguistic issues that play a significant role in sentiment analysis, such as negation or comparative sentences. Deep Linguistic Analysis automatically handles this kind of phenomena capturing the difference between opinions like:



? ?This phone is much better than my old phone.? ? Positive
? ?This phone is not superior to my old phone.? ? Negative

The sentiment analysis service is not limited to extracting a single opinion per sentence. It actually detects as many opinions as the sentence contains. For instance in the sentence ?This phone is awesome, but it was way too expensive and the screen is not big enough? three opinions will undoubtedly be extracted: ?phone? + ?awesome?, ?phone? + ?much too expensive? and ?screen? + ?not big enough?.