Automatic Image Analysis in Social Media

In social media, photos are at least as important as text. On Facebook alone, members upload an average of 350 million images every day. Until now, image content has remained largely closed to market research, but many images convey important information about consumers and brands. NIM has developed the PictureScan tool, which automatically extracts marketing-relevant knowledge from user-generated photos.

The raucous party with friends, the vacation on the beach, the newly acquired smartphone – all this is captured with the camera and shared via social networks. The flood of images is correspondingly large: the photo-sharing platform Flickr, for example, already has 8 billion images. Around 3.5 million are added every day. Facebook's 1.15 billion users upload an average of 350 million photos a day. Facebook's total photo database already numbers 250 billion photos.

These snapshots not only provide insight into users' lives, they also reflect their attitudes and experiences toward brands and products. And they influence a potentially large circle of viewers. In this context, the impact of images is often greater than that of text, because they are perceived more subtly and influence viewers' emotions more strongly. User-generated photos are also characterized by a high level of credibility compared to professional photographs. Social media images therefore represent a rich source of data for market research – which, however, has hardly been usable so far because existing tools for social media analysis only look at word posts.

Due to the large number of images in social networks, manual analysis is only possible to a limited extent. Therefore, automated methods for image analysis are necessary. The goal of the NIM is to develop a tool that extracts marketing-relevant knowledge from user-generated photos. To do this, the image content must first be recognized using methods from the field of computer vision. Through further analyses, awareness, popularity and usage of brands and products can be determined. These key figures are evaluated in comparison to the competition and over a period of time. In this way, trends can be uncovered, and opportunities and risks for image and sales can be estimated.

In addition to an analysis of the image content, it is also important to look at the people who posted the images online. From this, interest profiles can be formed that are useful, for example, for addressing people in direct marketing, placing personalized advertising or acquiring brand ambassadors in word-of-mouth marketing. The project is also investigating how brand- and product-related images on the social web influence people's preferences. This information provides valuable clues, for example, for the design of advertisements and web presence as well as for the control of word-of-mouth advertising.     

• Innovation Award 2016 of the Professional Association of German Market and Social Researchers (BVM)