The Study On Personalized Search Based On Semi-Supervised Clustering

With the rapid development of e-commerce in the twenty-first century, onlineshopping on the Internet increase more and more frequently. It is necessary to improveour services to meet customers’ requirements. In e-commerce system, the customersneed to search for related products through search engine. Traditional search function isbased on keyword or some certain rules, which finally return query results to thecustomers. In order to provide customers with personalized query service, thepersonalized search engine was born out. With the help of personalized service, thecustomers can get better results; at the same time, it can help the customers to increasetrading opportunities, identify potential customers and so on. Facing to the lack of thecurrent search engines and the search function demand for current users, based onresearch of search engine and its technology, this paper explores two solutions ofrealizing the personalized search and implements a personalized search engine systemwith integrating semi-supervised clustering. The work is as follows:1) As an automatic process to collect information, the web crawler can not onlycollect network information for the search engine, but also as a directional informationcollection device to obtain some certain information about sites. This paper implementsa web crawler system by utilizing java language based on the regular expression and thestructure analysis of web pages. We crawler the details on Jingdong Mall, and then storeall the data into database after preprocess.2) This paper studies on personalized goods search based on feature orderpreferences. Semi-supervised clustering algorithms guide clustering by integrating asmall amount of priori knowledge, and improve clustering quality and affectivity. Thispaper puts the commodity attribute information of users concerned as priori knowledgeto guide clustering and assist the products division, in order to implement thepersonalized search. This paper analyzes the clustering results based on objective andsubjective criteria, so it can confirm the validity of the products division. 3) In order to more accurately reflect the requirements of users, this paperintegrates the instance-level knowledge in the form of pair-wise constraints intosemi-supervised clustering with attribute order preferences to assist partition ofgoods,then implements the personalized search. This paper introduces the acquisitionand representation of the instance-level knowledge in the form of pair-wise constraints,and how to implement the similarity division of products by integrating instance-leveland attribute-level knowledge. Finally, we analyzes the clustering results based onobjective and subjective criteria.4) Combing above studies, based on the web crawler, facing to the twopersonalized search engine, we design a personalized search engine system.Through the above problems, this paper can be used as a solution for personalizedsearch engine in e-commerce. This work can be applied to e-commerce and providepersonalized recommendation technology and marketing strategy with technical supportand theoretical basis. At the same time, the methods of this paper can provide otherrelated application with research ideas.