Research Of Public Display Interaction Model And Recommendation Strategies Based On Visual Perception

Public displays are replacing static signals in recent years, appears everywhere providing audiences a variety of services such as advertising and navigation. Thus making local audiences interact with these screens possible. Advertising is the main revenue source of public display. When used for advertising, information can be delivered to audience directly and efficiently, which can create a high impact on audience experience. However, due to the difficulty of collecting users’explicit feedback, e.g., user rating or click, the interaction model of users and public display advertisements has always been unsystematic, incomplete and inaccurate, and there is no accurate model of interactive behavior between user and display screen Advertisement recommendation strategies cannot be optimized, personalized advertisement cannot recommend to audiences either. There are great theoretical and practical significance in optimizing recommendation strategy of display advertising and enhancing user experience.This paper focuses on the public display advertisement in malls, office buildings, elevators or some other public areas. We use video camera and computer vision face detection technology to record the audiences’arriving and departure on public display, and her advertisement watching behavior, find the relationship between users and advertisements, then to establish a new interaction model. Based on the above model, we furthermore design interactive recommendation strategies to predict users’ "satisfaction degree".This paper contributes follows:(1) Users’"satisfaction degree" is used as an optimization target, establish a new interactive model of users and public display advertisement by analyzing existing public display interaction and recommendation system.(2) Based on the above model, we study this problem as a group recommendation problem, we build a primary recommendation system framework and further design advertisement strategies handle the case that multiple users arrive and watch the public display at the same time.(3) In the proposed strategies, we first propose to use users’viewing time to infer audiences’preference in the advertisement, and present two methods for inferring individual audiences’preference. Experiment results shows that it is accurate to infer a user’s interest in a public display advertisement from her viewing time and both of the two methods are efficient.(4)On the other hand, in the group strategy, this paper introduces a media similarity metric to characterize the media feature similarity between target advertisement and neighbor advertisement, which reveals the correlation between advertisements and presents how the content of advertisements affect users’ "satisfaction degree". Experiments shows that the proposed strategies can indeed predict the group of users’ "satisfaction degree" efficiently and the media similarity metric has a good performance in explaining the correlation between public display advertisements.