I have reviewed several tools to analyze sentiment: their methods varied from manual sentiment setting to textual analysis, emoticons and symbols. Today we are looking at a semantic sentiment analyzer: Opinion Crawl which uses SenseBot (we have mentioned previously) as the semantic analysis engine.
Opinion crawl has two large section we’ll review separately: the sentiment tracker (for hot topics news broken into categories) and sentiment search.
Opinion Crawl Sentiment Tracker
The main page displays key topics grouped into categories:
- Current events (Oil spill, Afghan war),
- Political figures (Barack Obama),
- Entertainment (Lady Gaga, The Last Airbender),
- Companies (Goldman Sachs),
- Economy (Unemployment, Chinese economy),
- Products (iPad), etc.
Clicking on a topic takes you to the blog which is automatically generated every day. The blog tracks the sentiment trend on a large number of Web publications and provides a daily/weekly/monthly view. For example, clicking on “The Twilight Saga” (it’s not that I am a fan, it’s just that I found that example most interesting) takes us to this page that visualizes:
1. Daily Sentiment:
2. Sentiment Trend
(Trending of positive/negative/neutral and overall mentions over a period of time):
3. Positive-to-Negative Ratio
(Relation of positive mentions to the negative ones)
Opinion Crawl Search
There are 2 search buttons on the main page: Sentiment in the news and Sentiment on Twitter. They allow you to get an ad-hoc reading of the sentiment on a small number of recent news items or tweets. (Note that the ad-hoc analysis presumes that the topic you are searching for is in the news or is being actively tweeted about.)
The result is displayed as a snapshot of the current sentiment as a pie-chart; latest news and images on the topic.
One of the most useful features I loved was a semantic cloud consisting of key concepts extracted from the news or tweets. The concepts allow you to see which issues may be driving the sentiment in a positive or negative direction.
Analysis of Twitter consists of searching the tweets on a topic and analyzing their text. Many tweets do not have enough textual content – they may just contain a URL pointing somewhere and a couple of words. So the engine needs to extract a decent volume of tweets and parse them for sentiment expressions, and for key concepts that are being discussed (semantics).
Twitter sentiment often swings widely throughout the day, whereas the news sentiment is more stable.
Compare the Twitter sentiment for “Old Spice”, especially the sematic cloud with the above screenshots – amazing!
A pleasant surprise was that the tool was quite reliable at identifying the current sentiment, you may want to give it a try and let me know your thoughts!