UNDERSTANDING AND ANALYSING THE LANGUAGE OF IMAGES
As the saying goes, “a picture speaks a thousand words”. For this reason, visual monitoring, i.e. the generation of marketing insight from image analyses with AI, are likely to increase in importance. After all, social media offer a huge data pool that is just waiting to be analysed.
Throughout all stages of the development of written communication, images have maintained their place as a central way of sharing knowledge. While the geographic range of our ancestors’ glyphs and murals was rather limited, big data and social media now enable access to a huge arsenal of user-generated content (UGC) from virtually any location. Basically, it is irrelevant who publishes which content, and where. Each of these images – with or without textual reference to a brand, person or subject – speaks its own language, tells its own story or provides valuable insight on demographic or psychographic properties.
The collection of data is the basic pillar of visual monitoring. The sources from where data are collected depend on the type and purpose of the collection. Social media doubtlessly represent an “El Dorado” for marketing experts and social researchers. Every day, huge amounts of image data are generated, shared and liked on platforms such as Facebook, Instagram and Twitter. However, quantity and quality are quite different concepts. Thus, the huge pond of visual information needs to be raked for data that deliver insight on topics, trends, preferences, products, target groups or target measurements – i.e. data that actually cater to the questions to be answered.
To find relevant image data, crawlers dock on to the APIs of social platforms and other web pages, where they collect data. Often, these search algorithms come with an application (web tool, SaaS, standalone software) that offers numerous features for filtering, sorting and editing the collected data. Tools like Google Cloud Vision, Amazon Rekognition or Talkwalker represent an excellent alternative, especially if the overhead for programming custom crawlers and applications or the storage of data exceeds a company’s internal resources.
Image analysis is another important aspect of visual monitoring. The analysis can be performed automatically or manually. Manual analysis is not recommended, as it involves substantial costs and workload. On the other hand, exponentially growing computing performance and increasingly efficient algorithms are making automated image analysis the solution of choice in many application areas. With its deep learning approaches, artificial intelligence offers vital advantages: Depending on the computing power and code, this emerging technology can speedily identify any relevance, patterns or correlations from huge data repositories, never gets tired, is capable of learning and improves its quality with every new set of training data and every update.
However, technical or legal roadblocks such as the following can limit or even prevent the deployment of fully automated, AI-based tools:
- Limited access to interfaces of relevant sources
- Legal restrictions, such as image rights, data protection law, or GTC of platform operators
- Resource-specific obstacles, such as lack of computing performance or weaknesses in the learning algorithm.
Learning algorithms are already capable of identifying people, faces, emotions, animals, scenes, landscapes, landmarks, utilities, trademark logos, handwriting and other things. An interesting application from the field of medicine, which serves as a showcase example of the use of image analysis, is the machine detection of carcinomas in patient images, which is to assist doctors in their diagnosis. The state and its surveillance authorities also increasingly make use of image analysis. For example, a software application for the identification of licence plates can examine a licence plate virtually in real time and query a database to check whether the particular vehicle has been involved in any offence or crime (e.g. theft, hit and run, etc.). Moreover, the highly controversial face recognition with cameras at railway stations, airports and public sites enables the identification of potential offenders with recognition software.
VISUAL MONITORING – MONITORING, ANALYSING AND GAINING INSIGHT
Visual monitoring is already being used in various research areas and industries. A well-trained artificial neural network can provide assistance in gaining valuable insight beyond the scope of what is possible with mere image analysis. For example, based on a pre-selected, qualified image data pool, an AI-based tool can be used identify and analyse correlations, links or patters to persons, emotions, objects or scenes. Despite the rapid development of self-learning algorithms, such tools must still be viewed as assistants to human analysts, who are able to interpret deeper psychographic insight such as life styles, attitudes or preferences from the pre-selected image data. In the following sections, we draw attention to some of the numerous ways in which visual monitoring can be used.
BRAND AND REPUTATION PROTECTION
Brand and reputation protection is a critical task in an organisation or enterprise. Crises, shitstorms, trademark infringements or personal insults are just some of the potential risks associated with Web 2.0. Sometimes, threats are only detected very late, as purely text-based monitoring is usually limited to search queries that target keywords, phrases, authors, channels or hashtags. Social networks often contain posts that are too complex for algorithms to handle. For example, this includes posts that do not contain any textual reference, but whose image content represents the actual danger. Sometimes, a combination of text and images is what adds up to a threat. Purely text-based monitoring may be unable to identify such posts.
The following critical Instagram post shows the Nutella brand logo in a photomontage. The associated statement represents a potential reputation risk for the brand image. Though the algorithm is not yet able to interpret a potential brand image risk as such from the image context alone, the logo recognition in the context of the visual monitoring allows this post to be captured and analysed by a human analyst – despite the photomontage and the lack of text. As the textual reference to the brand is missing, such a post would be invisible to conventional text-based monitoring. Apart from damaging the brand reputation, the post most likely constitutes an infringement of image and trademark rights.
SPONSORING AND INFLUENCER MARKETING
Influencer Marketing is a disputed, but nevertheless highly popular instrument to raise awareness of products or trademarks and increase their popularity via social media. Visual monitoring can help to find suitable brand ambassadors or influencers, to coach paid partnerships and to analyse the results of such a collaboration. Visual monitoring can also deliver accompanying services in the field of sponsoring, e.g. to measure the ROI of a sponsoring campaign with indicators gained in the social web.
Perimeter advertising in stadiums reaches the spectators who watch an event live or via transmission. Moreover, the reach of perimeter advertising can be enhanced significantly e.g. by means of photographs shared in social media. Whether a selfie for the personal pinboard or a professional shot for an online article – a trademark logo in the context of perimeter advertising always achieves a certain reach and impact on the platform on which it is shared, even if it merely appears in the background. Image analysis enables monitoring to make use of image data in which e.g. the trademark logo merely occupies a secondary place and appears without any textual reference. The following example shows a sponsoring campaign of the Bitburger Brewery Group in the background of a post about a motor sports event.
IDENTIFYING THE PREFERENCES AND INTERESTS OF THE TARGET GROUP
To know what existing and potential customers like or what moves them are just two of the various complex questions to which visual monitoring can deliver meaningful answers. One of the most prominent disciplines in the field of marketing is the identification of correlations and patterns in the consumption behaviour. Where does the consumption take place primarily? Do complimentary products appear? In which temporal or social context is a product or service used? To gain insights, subjects, objects, landscapes or image scenes from the pre-segmented image data pool are analysed for patterns or correlations. Which objects and scenes appear frequently in connection with a product? Is the beverage frequently consumed at the beach, in restaurants or in the backyard at home? By enriching the insights gained with additional variables (e.g. demographic data), it is possible e.g. to develop measures to address target groups more effectively.
TRENDING AND PRODUCT DEVELOPMENT
Trend-setting is not the exclusive domain of creative brains in agencies, innovation labs or product development facilities. Web 2.0 has given every single user a voice and the opportunity to set trends or share in determining trends. For example, insights derived from qualified image data can yield product hypes, popular colour combinations, places that are in vogue or tasty mixed drinks. Could the trend drink of next summer perhaps be developed from a user creation? Or should the latest collection be tuned to “lemon yellow”, which is so popular with the community? Of course, a hype does not necessarily mean a trend, and many hypes are quickly forgotten. Nevertheless, it makes sense to analyse such hypes as well as past trends in order to learn or incorporate creative user ideas in the product development and solutions to problems.
Irrespective of the topic, objective or question that an organisation, a public figure or a company is currently busy with, the potential of visual monitoring should not be left untapped. In the coming weeks, the CURE blog will feature more information on how this can be done.