Knowing what will happen in the future would be of great value to every person and to every business. In the business case, anticipating future events could allow for changes in strategy and planning to maximise the possible benefits or minimise the probable damage of the future event.

While a perfect knowledge of what even tomorrow will bring is famously impossible to be certain of, companies have always sought to optimise their business activities according to the best knowledge possible of what will probably happen in the future. In this age of big data driven by the scale of Web and social media data, combined with the latest AI technologies (machine learning, deep learning), global expectations of future events can be collected and analyzed. Predictive analytics seeks to maximise the accuracy of predicting the future and generally works only within  tightly controlled domains.

At the School of Data Science, we are researching how different data inputs – e.g. past topics of discussion in social media or knowledge of past events – can be processed by different AI models to improve the predictive accuracy in open domain settings such as what will be the next trending topic in social media or what will be the most significant event on a future date. Perfect knowledge of the future may be unreachable but our research shows how knowledge extracted from big data sources can help guide organizations to make better decisions based on the most probable future.

The field of computer vision deals with making computers capable of “seeing” what is in an image. Recent advances in computational understanding of multimedia driven by AI have enabled significant progress in computer systems that can accurately identify concepts in visual content and label images (or video frames) according to emotional characteristics, objects and events. Deep learning – the use of neural networks that emulate the neurons in our brains with multiple processing layers – has enabled the most significant advances and while “generic” deep learning networks seek to identify the primary objects visible in an image, these networks can also be trained for specific use cases.

At the School of Data Science, we are researching how deep learning-based visual classification of images can be applied to the field of destination image measurement, meaning extracting a model of how a tourist destination is presented visually. Here, our network needs to be trained specifically on images typical to tourist activity at a destination and to classify them according to commonly accepted attributes of tourist destination image models. This is the first example to our knowledge of a visual classifier system specifically for tourist destination image. With >90% accuracy we can model how different tourist destinations are presented visually in social networks like Instagram and compare the tourist’s destination image with that of the official marketing organizations. As a result, destination marketing organisations can better adapt their online marketing to promote those aspects tourists are most interested in.