Democratising Big Brother: Web surveillance for the SME | Deeper Insights™

Inferring the national mood from social media about a product, or event, inevitably relies upon sentiment, or more recently emotion analysis of relevant posts. Sentiment is simply the classification at the sentence, document or topic level of social media posts into three categories: negative, positive or neutral. Older techniques use keywords, but this approach ignores context. And on social media, this is a grave error. For example, the Tweet below would be classified as Negative using older techniques. The latest techniques take into consideration context, as well as word relationships within the domain.

Emotion analysis is a finer-grained form of sentiment analysis where an emotion is assigned to negative or positive sentiment. NC State University has supplied a tool through which emotion analysis can be conducted about specific topics. The following image demonstrates the current sentiment about Brexit.

Source

The tool also demonstrates the words associated with each topic within the Brexit domain as well as their emotional value.

Analysis of social media is not limited to emotion detection. Information aggregation, for example, can be used in various tasks such as food price prediction, and tracking of animal disease outbreaks. Information aggregation techniques locate target information and in some cases the location of the social media post. From the aggregated information a trend is inferred.

The final technique that will be covered is the identification of causal relations between words or phrases in social media posts, and events in the real world. The causal effect of words and phrases are aligned with real-world events through statistical techniques such as Granger Causation. Once the relationship has been established future events can be predicted from social media posts. There have been studies with event prediction from social media that have predicted political revolutions.