Twitter sentiment analysis
The Australian national science agency CSIRO has developed a tool to mine Twitter data and to analyse the sentiment behind the content. The tool, called We Feel, detects emotions featured in a crowdsourced dictionary of emotions and then goes on to determine the intensity of the emotion.
The tool was used by CSIRO to analyse tweets about the Sydney Siege of December 2014 – you can read this case study here.
Detecting sarcasm
Meanwhile researchers at Carnegie Mellon University are trying to teach computers to recognise sarcasm on Twitter. The challenge for computers is that in order to identify sarcasm you need to do more than analyse the linguistics, but you also have to take into account context. Remember the story of the frustrated would-be traveller who tweeted he wanted to blow up an airport? The researchers have created complex models that take into account the author’s profile, audience, and historical content and they claim that its tool is now 85% accurate.
Computerised small talk
Microsoft has trained chatbot XiaoIce to engage in chatter. Over 40 million users in Asia have surprisingly long conversations with the bot – an average of 23 turns. Harry Shum – head of Microsoft Technology – says the ability to chat and exchange pleasantries is essential in the workplace and developing this ability may help bots integrate more into human lives.
Sources: endgadget; wired; datadrivenjournalism.