Learning features weights from user behavior in Content-Based ImageRetrieval

Bibtex entry :

@inproceedings { VG:MMS2000e,
    author = { Henning M{\"u}ller and Wolfgang M{\"u}ller and David McG. Squire and St{\'e}phane Marchand-Maillet and Thierry Pun },
    title = { Learning features weights from user behavior in Content-Based ImageRetrieval },
    booktitle = { ACM SIGKDD International Conference on Knowledge Discovery and DataMining (Workshop on Multimedia Data Mining MDM/KDD2000) },
    year = { 2000 },
    editor = { S.J. Simoff and O.R. Zaiane },
    address = { Boston, MA, USA },
    month = { aug 20-23 },
    url = { http://vision.unige.ch/publications/postscript/2000/MuellerHMuellerWSquireMarchandPun_mdmkdd2000.ps.gz },
    abstract = { This article describes an algorithm for obtaining knowledge aboutthe importance of features from analyzing user log files of a content-basedimage retrieval system (CBIRS). The user log files from the usageof the \emph{Viper} web demonstration system a re analyzed over aperiod of four months. Within this period about 3500 accesses tothe system were made w ith almost 800 multiple image queries. Allthe actions of the users were logged in a file. The analysis onlyincludes multiple image queries of the system with positive and/ornegative input images, because only multiple image q ueries containenough information for the method described. Features frequentlypresent in images marked together positively in the same que ry stepget a higher weighting, whereas features present in one image markedpositively and an other image marked negatively in the same stepget a lower weighting. The \emph{Viper} system offers a very largenumber of simple features. This allows the creation of flexible featureweightings with high values for importan t and low values for lessimportant features. These weightings for features can of course differbetween collections and as well between users. The results are evaluatedwith an experiment using the relevance judgments of re al users ona database containing 2500 images. The results of the system withlearned weights are compared to the system withou t the learned featureweights. },
    owner = { steph },
    timestamp = { 2008.05.04 },
    url1 = { http://vision.unige.ch/publications/postscript/2000/MuellerHMuellerWSquireMarchandPun_mdmkdd2000.pdf },
    vgclass = { refpap },
    vgproject = { viper },

Keywords: machine learning, information geometry, data mining, Big Data, affective information retrieval (recherche d'information), information visualisation, content-based image and video retrieval (CBIR, CBR, CBVR, CBMR, CBMIR), information mining, classification, multimedia and multimodal information management, semantic web, knowledge base (RDF, OWL, XML, metadata, auto-annotation, description), multimodal information fusion