by Thomas Bosshard
Advances in the field of opinion mining and sentiment analysis has opened up unprecedented access to insights into what people think. On Facebook alone, the most widely used social media platform, 510,000 comments are posted, 293,000 statuses are updated, and 136,000 photos are uploaded every minute. Due to the sheer quantity of data that an advanced, AI-based analytics tool can mine, insights gained from social media analytics are considered “more accurate” than the results of traditional avenues of market or political research and are being used to drive business decisions and shape government policies.
But what about all the “noise” and trashy information on social media? Or even worse: the recent onslaught of fake news? There is no question that misinformation is on the rise, and it may also be having an effect on the quality of data gained from social media conversations. Furthermore, not everyone is honest on social media, and people have become used to not believing everything they read or see. Therefore, can an algorithm be just as skeptical and, for example, able to recognize “spin” when it sees it?
As a result of the negative role Facebook played in spreading intentionally misinformation during last year’s US presidential election, Facebook was forced to acknowledge that it had a fake news problem. It responded by announcing partnerships with third-party fact-checking platforms such as Factcheck.org as well as with independent journalists from Associated Press. Their task is to verify the truth of a story on Facebook and to flag those news items with a "disputed" tag that they deemed fake. However, according to a recent article The Guardian, one year later these journalists are now raising concerns that the collaboration with Facebook is not having the intended effect. Furthermore, a lot of news still goes unchecked, or it takes days until a fake news story gets flagged, and by then it can have already gone viral.
Considering the trillions of posts on social media, manual fact checking will hardly be ever able to solve the problem. Some believe the answer to this issue is Artificial Intelligence. Fact-checking bots, i.e. AI-based application devised to detect words or patterns that indicate questionable information, are already used on Facebook. However, whether pattern recognition will be able to curb the proliferation of fake news has yet to be proven. Purely AI-based approaches still struggle with the nuances of human conversation, and social media is full of sarcasm, innuendo and humor. Thus, a real understanding of social media conversations will remain a very tall order indeed for an algorithm.
For this reason, many opinion-mining approaches add a layer for “human analysis” which should enable classification of “unstructured” data such as social media posts and conversations – a requirement for the ability of an algorithm to accurately analyze it. Therefore, it is still important to realize that the more data available to algorithms to mine, the more opportunity there is for bad information to seep into the results. It does not matter whether this data is fake, distorted, biased or simply not accurate or whether the information was intentionally planted on social media or not. As long as there is no solution to separate the huge amounts of wheat from the huge amounts of chaff, insights gathered from opinion mining on social media may never be very sophisticated or even be compromised.