"He loves me...He loves me not?"
People have relationships with brands.
This is a very complex concept because, after all, how can you have a relationship with an idea?
Social media has helped us all understand this concept much better by publicly exposing the emotions people portray when speaking about brands on their social networks. And it doesn't stop at emotion towards brands... We have sentiment towards products, concepts, events and many other sorts of entities.
The fact that these sentiments are now portrayed in the public domain offers a wide set of opportunities. Measuring and analyzing these sentiments gives us a much clearer picture of human perceptions, needs and aspirations.
Measuring sentiment effectively is a challenge which is dealt with in numerous ways, by different people for a vast array of applications. In order to measure sentiment we must go through 2 stages:
- Sentiment extraction (identification)
- Sentiment scoring
The sentiment extraction challenge presents the difficulty of recognizing the existence of sentiment in textual conversation with reference to a specific entity. In this paper we concentrate on the second stage - the scoring of sentiment.
"How deep is your love?"
Sentiment scoring must give us a full understanding of the displayed emotion and therefore should include all the necessary explaining factors. The critical factors needed for good analysis of sentiment are:
In many cases, the analyzed entity is mentioned within a conversation with no recognizable sentiment. Such is the case with informative conversations, for instance, the relative amount of incidents within a corpus of conversations in which sentiment towards the analyzed entity was recognized. In Buzzilla, we refer to "sentimentality" as an explaining parameter which is listed as an addition to the main sentiment index (the polarity index).
The most basic question that arises when it comes to sentiment is of course: "is it good or bad?" This is the question of sentiment polarity which represents the most central issue and therefore takes the role of the main parameter that is measured in terms of sentiment. Buzzilla's "Main Sentiment Index" represents the measurement of positive conversations vs. negative conversations. The index also takes into account the level of balanced conversation.
This index is represented on a scale of (-10 : 10), where "10" is 100% positive, "0" represents a balanced conversation and "-10" represents 100% negative.
And then there's the question of passion, or in other words, the intensity of the displayed sentiment. Do people just like the brand or do they really love it? Is it just a little resentment or are we witnessing fully fledged hate? Traditionally, sentiment tagging is done on a simple 3 point scale (Positive/Neutral/Negative) - but this system does not enable the measurement of sentiment intensity. Buzzilla uses a 5 point ordinal scale (Very Good/Good/Balanced/Bad/Very Bad) which allows us to measure the variance of the tagged sentiment, in order to understand if the displayed sentiment is of high or low intensity, or in our terms, whether it is passionate.
Passion is the second explaining parameter of the main sentiment index.download PDF