FIELD: physics.
SUBSTANCE: invention relates to a personalized search method based on product image features. Method comprises extracting, using a neural network model, a vector of abstract semantic features of an image by category, calculating the average value and dispersion of the abstract semantic features vector for each measurement, respectively, and performing the normalization process in each measurement on the abstract semantic features vector, calculating the user's behaviour weight when viewing, at that, adding the normalized vectors of abstract semantic features, extracted by category from all images viewed by the user, with obtaining weight vector of user interest for each category, finding a scalar product on image feature vectors not viewed by the user for the category to obtain a score of each of the images not viewed by the user, ranging the image according to the obtained score, selecting a predetermined number of images with the highest score for storage, personalized search based on the ranging result of the ranking step.
EFFECT: technical result is to increase the relevance of personalized search.
10 cl, 2 dwg
