Research On Multimedia Advertising

With the rapid growth of Internet bandwidth and the development of new Internet services, the explosively growing online multimedia data and the booming social network have brought new challenges to online advertising. The traditional text-based contextual advertising reveals a number of shortcomings on multimedia advertising. Efficient storage and management of online multimedia data for mining the relevance of multimedia data and advertisement are becoming the key issues of multimedia advertising. On the other hand, the rise of social network is making people’s daily network behavior changing from the content-based Web browsing to interaction-based social browsing. By using the huge backbone and user interaction in the social network to perceive the advertising trends has become an important direction in multimedia advertising research.In this thesis, we first discuss the background of multimedia advertising; then we investigate the key problems and point out the shortcomings of existing research. On multimedia advertising, we conduct a deep research in three paradigms:data management, content analysis and advertising optimization, and propose the Near-Lossless Video Summarization, the improved Near-Lossless Summarization and social-network-based image adverting system. This thesis conducts a deep research on multimedia advertising and obtains the following achievements:1. The daunting yet increasing volumes of videos on the Internet bring the challenges of storage and indexing to existing online video services. Current techniques like video compression and summarization are still struggling to achieve the two often conflicting goals of low storage and high visual and semantic fidelity. In this thesis, we develop a new system for video summarization, called "Near-Lossless Video Summarization"(NLVS), which is able to summarize a video stream with least information loss by using an extremely small piece of metadata. The summary consists of a set of synthesized mosaics and representative keyframes, a compressed audio stream, as well as the metadata about video structure and motion. Although at a very low compression ratio, the summary still can be used to reconstruct the original video nearly without semantic information loss. We show that NLVS is a powerful tool for significantly reducing video storage through both objective and subjective comparisons with state-of-the-art video compression and summarization techniques.2. Although achieving a very low compression ratio while preserving the semantic lossless, the proposed NLVS has not reached its limit. In NLVS, we leave the mosaic uncompressed and only compressed the audio track without content analysis. Moreover, we need to perform NLVS on a series of video analysis task to evaluate its performance in video content analysis. We presents an improved approach to video summarization, called "Near-Lossless Summarization"(NLS), which is able to summarize a video stream with the least information loss by using an extremely small piece of metadata, which is even smaller than NLVS. The proposed approach, can be applied to many video-based applications, such as visualization, indexing, presentation, duplicate detection and concept detection. We evaluated NLS on TRECVID and other video collections, showing that, NLS is a powerful tool for significantly reducing storage consumption through comparisons with our previous work NLVS and state-of-the-art video compression techniques, as well as a reliable tool for keeping semantic fidelity through video content analysis tasks in video management system.3. The conventional content-based contextual image advertising system dedicates to mining the contextual relevance of user’s browsing content for advertising. The social network has brought huge backbone and massive content organized on it. which makes the user influence an important role in information propagation. Unlike content-based contextual advertising which only considers contextual relevance, we proposes a social-network-based image advertising system, integrating the user influence, contextual relevance and local relevance, and simultaneously optimizing the dissemination of advertising and contextual relevance, as well as the insertion point of advertisement. We formulate the advertising problem as a non-linear0-1integer programming problem, and proposed a heuristic search scheme to reduce the computation consumption. Experiments confirmed that on the most popular image sharing site Flickr. the system achieved good relevance and encouraging user satisfaction.