Thursday, September 25, 2014

Understanding a Sentiment Analysis Engine

Microsoft Lystavlen - the Online display board - Understanding the Sentiment Engine in Microsoft Social Listening

Sentiment Analysis

If you want to see how the public perceives your company or product, you can use sentiment analysis, which determines people’s attitudes toward a topic. Sentiment analysis reflects the public perception of a post’s content in relation to the keywords that were used to find the post (a post is a eg a Twitter post or a Facebook comment)

Each post that results from your defined search queries is processed by the sentiment engine in the original language and annotated with a calculated sentiment value. Sentiment values are provided for the following languages:

  1. English
  2. German
  3. French
  4. Spanish
  5. Portuguese
  6. Italian

The sentiment value results in a positive, negative, or neutral sentiment for a post. Occasionally, the algorithm identifies positive and negative parts of a sentence and still rates the post as neutral. This happens because the amount of a post’s text identified as positive or negative cancel each other out. A post is also classified as neutral if there are no positive or negative statements detected in it.

Note that the sentiment algorithm is not a self-learning system, even if you can edit any post’s sentiment value in the post list.

Understanding the Sentiment Engine

Lets look closer at the sentiment engine using the example post below, in the context of the search topic "Windows Phone"

...

image..."

I thought this post a great explanation of Sentiment Analysis, which comes up in my day job now and then. While we don't use the Social Listen or related product, we do use a library that lets us integrate like functionality into our LOB, but e apps, explaining how it works quickly has always been "fun". This post and explanation will come in handy next time...

 

Related Past Post XRef:
Comparing Sentiment Analysis REST API's
10 Professionals, 10 views on the coming trends in text analytics

No comments: