För den otålige:
- här är ordlistorna som jag tar fram i den här bloggposten.
- här är notebooken med koden i bloggposten.
- här är Christopher Kullenbergs fråga på Facebook som detta är ett svar på.
För den otålige:
Just like words have different meanings depending on context, the positive or negative charge of an expression depends on the belief system of the audience. The Swedish company Gavagai and also Recorded Future that was recently funded by Google and CIA, have a simple and elegant solution to the problem of word meaning and different languages. Judging the sentiment of a text has the same problem to solve. When I started out as a media analyst (once upon a time) one of the things we where trained with was to define and learn to look from the customers eyes – their most important stakeholders eyes to be more exact. A famous example is news about downsizing at a company – positive news from the eyes of the majority of stock holders, but negative news from the eyes of the majority of the employees. When doing the analysis manually you normally need to choose only one of the interpretations for clarity and cost efficiency. When storing large amounts of data is no problem and you´re using computers for the text analytics process it´s possible to get a fuller picture.
The thing is that beauty, of course, lies in the eyes of the beholder – what makes one person thrilled makes another one angry. The underlaying mechanism is the belief system. So, in order to be more flexible in reporting sentiment across different audiences you need to start with analyzing the belief systems of different audiences. With that data available, sentiment analysis would come closer to being useful for people interested in measuring, evaluating and predicting the spread of ideas. Which is kinda what everyone from well-funded counterterrorism intelligence analysts to individual power bloggers are really interested in, in this medialized society.