| The library is an important place to learn knowledge, which has a large number ofbooks with kinds of categories and variable quality. It is urgent for readers to borrowhigh-quality books. Plenty of readers’ registry info, books’ introduction and circulationlogs are stored in the existed library systems, but they are used as a simple statisticalprocessing. Managers are not able to obtain useful information from this kind of analysis.Since that readers demand for personalized books, the service mode of traditional libraryis forced to change from passive to active recommendation. The proposed serviceconcept is a novel way to develop library service in the network era. The ratio of libraryand service level will be promoted by this proposed service. So, it has widespreadconcern.To accomplish this proposed service, data mining techniques are used to analyze thedata which are got from library system. According to existed circulation logs, readerborrowing behavior will be predicted. The k-means algorithm has been used to readersbehavior cluster in the traditional library application, but the algorithm theories are notresearched deeply. Due to the shortage of the k-means algorithm, the clustering result isineffective. Harmony search fuzzy clustering algorithm (IHFCM) is proposed in thedissertation, which provides a good way to solve parameter initialing set up and globallyconvergent. Then it has been used into a personalized recommendation in a certainlibrary appropriately. The following is the main work of the dissertation.1. The background, research significance and development of personalizedrecommendation on library are introduced firstly, such as the target of personalizedrecommendation, content and the existing problems. The paper structure is descried inthis section. Then some kinds of clustering algorithms used in library are described indetail.2. The improved harmony search algorithm is proposed in this dissertation.Harmony search with fixed parameter setting is easy to fall into local minima is found.The reason is that the way of producing new harmony affects convergence speed andglobal convergence ability. Dynamic parameter settings and changing the way ofupdating harmony accelerate global convergence ability. Finally, the improved harmonysearch algorithm and traditional algorithm were compared by some classical optimization benchmark in the experiments. The results show that the improved algorithm with strongrobustness is better effect in accuracy, convergence speed, global search and localadjusting than others.3. The harmony search fuzzy clustering is studied as following. The principle andcomplete process of IHFCM has been introduced. The original evaluation function onlyfocuses on the similarity of inter-cluster. IHFCM improve the evaluation function, alsofocus on the similarity of intra-cluster and inter-cluster. The convergence of IHFCMalgorithm is analyzed by Markov chain theory. The difference dimension of dataset isweighted by a certain method. Classical UCI datasets is used to validate the performanceof IHFCM algorithm in the experiment. The proposed IHFCM has better effect in thetime and clustering result than others. Then the IHFCM is applied to a personalizedrecommendation system.4. The design thinking and system framework of the personalized library isdescribed in the final section of the dissertation. The target and task of each functionmodule is analyzed and implementation process is introduced in details. Some exampledataset from a certain library are used to cluster reader behavior and get borrowingpattern of each cluster to evaluate the recommendation system cluster. Next, thepreference degree of books is calculated according to reader behavior info. Finally, thetarget of books recommended is completed and the reference suggestion is advised forthe development of library. |
PDF Full Text Request