A Proposed Upgrade to the Keyword Based Searches

918 Words2 Pages

INTRODUCTION: KEYWORD-BASED search has been the most popular search in today’s searching world. The result of Keyword based search is better than Google .On Google search engine user or searcher did not find relevant image result. This is because of two reasons. Queries are in general short and non-specific. Number of users may have different intentions for the same query . Searching for apple by a farmer has a different meaning from searching by a technical person .There is one solution to solve these problems is personalized search where user specific information is considered to distinguish between exact intentions of user queries and reranked the images. Figure.1: (top) non-personalized and (bottom) personalized search results for the query “Samsung Laptop”. Fig. 1 shows the example for non-personalized and personalized image search results from the search engines. The non-personalized search returned results only based on the user query relevance and displays Samsung laptop images as well as it can displays the Samsung charger battery on the above image in figure1. While personalized search results consider as both user query relevance and user preference, so the personalized results from a laptop lover rank the laptop images on the top. Increasingly developed social networking websites, like Flicker and YouTube allow users to create, share, upload, and annotate images. Flicker database is used to demonstrate the effectiveness of proposed system. The proposed system has two components 1) Ranking Based Multicorrelation Tensor Factorization model (RMTF) is used to calculate user’s annotation prediction which provides user preferences to assigning tag on image. RMTF avoids common noi... ... middle of paper ... ...person can find the personalized image list easily. 4. User obtained highly ranked images. CONCLUSION: Metadata created by users through their everyday activities on social networking site is used to obtain highly relevant images. Ranking Based Multicorrelation tensor factorization is introduced to eliminate the severe sparsity problems appeared in existing system. To find users topic, LDA (Latent Dirichlet Allocation) algorithm. The system introduces two main components to obtained personalized images. First is to calculate user’s preferences to assign a tag to the image and second is selection of single keyword query for relevant image searching. Users Sensitive Topics are generated to predict the user’s profile. The query mapping or query relevance and topic sensitive user preferences(TSUP) are integrated into final ranked result of relevant images .

Open Document