Social and Content-Based Approach for Visual Recommendation of Web Graphics

To combine social and content-based ltering techniques for image data, we have developed a new approach that uses clustering of images based on both content and social information. In this paper, we propose a visual interface that makes use of this clustering technique to explain the ltering results. The items in the database are laid out on a 2 dimensional map according to content-based similarity. A technique called \dynamic ltering" shows relationship between the items based on social similarity. The social and content-based clusters are visualized as landmarks on the map to enhance comprehensibility of the map. Visualization of both social and content-based spaces on a map combines their own advantages on information browsing: specialization of ltering results in the content-based space and serendipitous information discovery in the social space. We demonstrate an example of visualization applied to our web graphics recommender system. Abstract To combine social and content-based ltering techniques for image data, we have developed a new approach that uses clustering of images based on both content and social information. In this paper , we propose a visual interface that makes use of this clustering technique to explain the lter-ing results. The items in the database are laid out on a 2 dimensional map according to content-based similarity. A technique called \dynamic l-tering" shows relationship between the items based on social similarity. The social and content-based clusters are visualized as landmarks on the map to enhance comprehensibility of the map. Visualiza-tion of both social and content-based spaces on a map combines their own advantages on information browsing: specialization of ltering results in the content-based space and serendipitous information discovery in the social space. We demonstrate an example of visualization applied to our web graphics recommender system.