Research Of Some Key Issues About Information Core Optimization Based On Coevolutionary Algorithms In Recommender Systems

The popularization of the Internet makes data information paly an increasingly important role in our lives.The data brings convenience to our lives while making us have to face the difficulties of information overload.In order to solve the more and more serious problem of information overload and improve the ability of information retrieval,the recommendation system is presented.As one of the most outstanding algorithms in the field of recommendation algorithms,collaborative filtering algorithm gradually exposes its inherent defects under the bombing of massive information,among which the scalability problem severely inhibits the development of collaborative filtering algorithm.The scalability problem refers that the time consumption and space consumption of the collaborative filtering algorithm increase at an exponential growth trend when the data size increases.This growth trend leads to a sharp decline in the performance of collaborative filtering algorithm.This thesis proposes an information core optimization method based on coevolutionary algorithm to improve the accuracy of the recommendation,reduce the online recommendation time consumption and deal with the scalability problem.The behavior of finding effective information core using coevolutionary algorithm is called the information core coevolutionary optimization.The use of an optimized information core for collaborative filtering recommendation can guarantee the accuracy of the recommendation while alleviating the scalability of collaborative filtering.The research contents are as follows in detail:1.The recommendation algorithm based on information core can reduce the online recommendation time consumption.However,the existing methods just construct information core based on some heuristic strategy.The information cores constructed by these methods are easy to get into the local optimal maxima because they are built by subjective criteria or experience.For this situation,in this thesis,an information core optimization based on coevolutionary algorithm is proposed to overcome the drawback that current information extraction methods cannot fully exploit the advantages of information core.Compared with those methods based on heuristic strategy,the information core optimization based on coevolutionary algorithm has better global search performance.And the optimized information core has better recommendation quality and makes full use of the advantages of information core.The experimental results show that the information core optimization based on coevolutionary algorithm can find the better information core and thus greatly improve the accuracy of recommender systems.2.The above information core optimization method based on coevolutionary algorithm can give full play to the advantage of information core.But it may lead to the insufficient population diversity and then premature convergence because it strongly emphasizes the role of the elites in the evolution process.In order to further improve the diversity of the coevolutionary algorithm in the evolution process,the above algorithm is improved and a captain difference based selection mechanism is proposed.Firstly the size of elite sub-population is increased to improve the convergence speed.And then the captain difference based selection mechanism is used to maintain the diversity of the population during later evolution,which can avoid the premature convergence.In addition,we often meet a situation when there are many items whose prediction rating is same,we don't know which item to select to join the recommendation list.For this problem,the rule of priority recommendation based on item popularity is proposed in this thesis.According to the experimental result,the captain difference based selection mechanism can effectively alleviate the phenomenon of the premature convergence.And the rule of priority recommendation based on item popularity can improve the recommendation quality based on information core.3.The information core optimization problem that needs to consider multiple selection criteria simultaneously is modeled as a multi-objective optimization.And an information core optimization based on multi-objective coevolutionary algorithm is proposed.The method aims at providing different information core combinational schemes for customers' different needs,thus achieving the balance between the accuracy and the diversity.The results of experiments show that the information core optimization based on multi-objective coevolutionary algorithm can get different information core combinational schemes to satisfy the different needs of customers.In addition,the information core optimization based on multi-objective coevolutionary algorithm shows good performance in convergence.