The sentiment shown for each tweet should be used as an indication only. Tweet sentiment in Social Bearing is not always accurate and as with any natural language detection, there will always be false positives. Concepts such as sarcasm and irony are particularly hard to detect algorithmically.
The sentiment scores given on Social Bearing can often be useful to give an overall sense of sentiment for a particular search term, or to find the passionate tweets in any set of search results. The best results are usually around more emotive subjects.
Social Bearing gives a sentiment score based on the ‘bag of words’ technique, where each word in each tweet is matched against a set of positive and negative words. The sentiment score for each word is then added together to create an overall sentiment score for the tweet.
I used the ‘bag of words’ technique as a basis for sentiment scoring over more complex methods of language algorithms such as ‘Latent semantic analysis’ because:
I have plans to further refine the sentiment analysis by matching tweets against common emoji 😉 and other language patterns.
In the future, I may look to 3rd party API’s for sentiment analysis. Meta Mind’s deep learning platform for example looks incredibly promising.