Efficient Algorithms for web recommendation systems

Abstract

With the explosive and diverse growth of web contents, web recommendations is a newlinecritical aspect of the search engine. Different kind of web recommendations like query, newlineimage, webpage, movie, music and book etc. are used every day. Web recommendations helps the users to locate their required information more precisely for a given newlineinput query. It also helps the search engine to return appropriate answers and meetusers needs. Web recommendations mainly involve, (i) Query Recommendations thatrecommend related queries to users input query from user s search log, (ii) Webpage Recommendations that guide users to visit the Web pages during their activity on the web and (iii) Image Recommendations that retrieve relevant images to meet the user s requirement. newlineKeyword based search is extensively used to discover knowledge on the web. Generally, web users are not able to arrange and define input queries relevant to their search because of inadequate knowledge about domain. A method is proposed to recommend queries by combining two graphs generated from the user search log: newline1) query click graph that uses the knowledge of link between user input query and clicked URLs and 2) query text similarity graph that finds the similarity between two queries using Jaccard similarity. It provides literally as well as semantically relevant newlinequeries for the users need. Experimental results show that the QRGQR algorithm newlineoutperforms heat diffusion method[1] by providing more number of relevant queries. newlineUsers search log provides information needs from the users click behaviour. If a certain retrieved result is clicked by the user, it can not be concluded that the clicked result is completely relevant to the user query since the user has not seen the full document. But the brief description of the document i.e., snippet would have be already read by the user if he has decided to click on that document. It can be considered that snippet reflects users information need. Related Search Recommendation (RSR)framework discovers keywords pres