Research On Association Rules Algorithm For Personalized Recommendation

As the rapid development of internet technology, data explosion has gradually become a serious problem, and it's very hard for consumers to find prober products among vast goods in commercial field. As an important part of the personalized recommendation, the association rules have a remarkable effect on "data explosion". There are lots of research in the area of association rules at present, but most of the studies are focused on the work of improving algorithm efficiency. However, the algorithm of association rules can produce a mass of rules as its outcome, which make this method inconvenient to use, so some scholars proposed a series of method to simplified the outcomes of association rules, but these studies barely consider the effect of domain knowledge, it's an important factor in the work of extracting rules. Based on this point, this paper carried on a series of research as follows.First of all, this thesis reviews the research of personalized recommendation, and points out the difficulties of the research. At the same time, briefly introduced the content-based recommendation, collaborative filtering recommendation, knowledge-based recommendation and the association rules, analysis the merit and demerit of relevant algorithm, this thesis propose that mix different kind of algorithm can bring a better recommendation results.Second, the thesis analysis the difficulties of the current research, and indicate that low efficiency of association rules algorithm is inevitable, and this defect can be reduced only by actual situation. At the same time, this paper analysis the problem of the large quantity, low value, low storage and disorderly of the rules, the current research of improve the efficiency of the algorithm only from internal angle, but in fact domain knowledge play an important role for the use of rules, while this knowledge can be overlooked regularly. Therefore, this thesis proposed a new way to process the outcomes of association rules.This thesis put forward a new way to cluster the rules, and select the valuable rules through comprehensive evaluation index, this index contains domain knowledge, and the thesis prove that this method is effective. At last, this paper proposed a new way to solve the problem of rules extracting work, this method (ARCUR) integrate the working principle of collaborative filtering, and based on the user's ratings, the thesis conduct a secondary mining on the results of association rules, which works out the problem of extracting rules only by the index of support and confidence.In the end, as there are many parameters in the algorithm of the paper, the values of these parameters should consider the practical experience, then the algorithm can have a good effect. Association rules has a lots of important applications in real world, this thesis proves that the mining rules combined domain knowledge can improve the recommendation effect.