Research On A Personalized Recommendation Algorithm Based On Web Mining

With the development of IT technology, web2.0technology and cloud computing technology,on one hand, the problem of information overload make it difficult the people search information tosatisfiy their needs; on the other hand, the lack of personality of Information Retrival reduces theuser experience and leads to the user churn. The thesis introduces the personalized recommendationtechnology and explains its effect and method to solve the above problems; elaborates the theory ofcollaborative filtering and analyzes its shortages; presents a personalized recommendation algoritmbased on Web mining to improve the performance and quality of recommendation system.To solve the problems in traditional collaborative filtering, including data sparsity,“cold start”and high user engagement and so on, Web usage mining technology is introduced to excavate theweb log and know about the user behavior patterns, interests and preference; the matrix of“users-items” interest degree is built and the collection method and representation of data intraditional collaborative filtering are changed from explicit rating data to implicit interest measuredata. To solve the problems of similarity measure in traditional collaborative filtering, includingfalse judgment of neighbors and helplessness for new users or items and so on, singularity impactmechanism is introduced to compute the impact of a single component of user vector on similaritymeasure from the whole matrix of “users-items” interest degree and produce weighted influence onsimilarity measure. To optimize the neighbor set generated by traditional collaborative filtering,recommendation importance mechanism is introduced to compute the recommendation importanceof a user for an item and filter the result of similarity measure.To verify the personalized recommendation algorithm based on Web mining, the thesis collectsand mines the web log data from NJUPT servers by Web usage mining and builds the matrix of“users-items” interest degree as the experimental data source. Four evaluation indexes of MAE,coverage, precision and recall are used to compare it with the traditional collaborative filteringbased on PC. The experimental results show that the personalized recommendation algorithm basedon Web mining has more excellent performance on recommendation quality.In summary, the personalized recommendation algorithm based on Web mining improvestraditional collaborative filtering through the analysis of the shortages existing in it from threeaspects of data source getting, similarity measure and neighbor set generating; introduces Webmining technology, singularity impact and recommendation importance mechanism to ensure the reliability and accuracy of the recommendation results; acquires experimental verification.