Introducing Entity Sentiment

Earlier this year Attivio announced general availability of the Attivio Active Intelligence Engine (AIE) version 2.1. One of the key new features included in that release is a new Entity-Sentiment module.

Sentiment analysis presumably needs little or no introduction. It has been a hot topic for the past few years; no social media strategy is complete without it. Understanding the perceived strengths and weaknesses of your products and services - as well as your competitors' - is extremely valuable, and analyzing user generated content is a great way to gain insight and drive important business metrics, from simple "mentions" to more complex "conversions" measures.

One limitation of common sentiment analysis software, however, is "entanglement". As companies partner, embed and bundle their products together, but retain their own branding, document level analysis may simply conflate them.

For example:

I have seen the future and it is the coolPhone! You won't believe it until you rub your finger across the coolPhone screen and watch pictures and albums and email zip by. The coolPhone is Incredible! And there are endless games you can download and sit and play for hours, if you get bored with that, you can turn on coolMusic and just DJ for yourself all day. I don't use coolPhone to watch video but it is great for music. One problem though is frequently awful and unreliable service from BigNet. Every time I visit NYC I have to bring my old phone because it's hard to get a signal. Maybe I'll look into unlocking it or something but I heard that breaks the warranty.

Is this document positive or negative? On an absolute scale it seems more positive than negative. The question then is one of perspective: coolPhone, using document level sentiment analysis, would correctly count this as a positive mention. BigNet, however, will get a 'false positive' because the review is positive, and they are mentioned. The two companies are entangled from a sentiment perspective.

AIE's Entity-Sentiment module analyzes the sentiment using a patent-applied-for technique. It views the review text as a continuum. Each sentiment bearing word is treated as a signal strength. Other words can effect or reverse that signal. This signal is then smoothed and the entities are analyzed by proximity.

In this way AIE is able to deduce that this review appears to be generally positive in tone, that the entity "coolPhone" is mentioned positively, but the entity "BigNet" is mentioned negatively. This provides overall deeper insight and supports more precise decision making.

Of course coolPhone vs. BigNet is a fairly common sort of entanglement - one between two companies. In many cases the insight will be more subtle - for example a company might want to compare itself to competitors, or alternatives to the product or service they provide.

Beyond entities, the goal might be to study particular concepts, which, like entities, are really noun-phrases. For example it is possible to study the various components of a product and find out how they contribute to the overall value. By studying the strengths and weaknesses of your products and your competitors, you will swiftly identify new opportunities to differentiate, and gain or defend market share.

TrackBack URI for this entry

Comments (0)