An Adaptive Recommendation Method Based On Small-world Implicit Trust Network

With the rapid development of information technology, there are more and moreavailable information resources on the network, so it is become more and moredifficult for users to obtain useful message from the information space. In order toresolve the problem of information overload, the recommendation system came intobeing. With the expanding scale of the recommendation system, the number of usersand products has increased dramatically, however, the ratings in the system are verylimited, so that to making the recommendation system under a severe problem nameddata sparseness, which leads to the greatly reduced of the recommendation quality. Inthis paper, on the basis of comprehensive analysis of the current research status athome and abroad, and with the help of the small world features of a trusted network,we try to research in a high level of the sparseness problem in the recommendationsystem.Firstly, a method of constructing the small-world implicit trust network isproposed based on user-item ratings. The users are divided into several groups bythe clustering technology, and connect the users based on the implicit trust betweenthem, which insures the constructed network has a larger clustering coefficient, andthen the edges between correlative groups are optimally added to shorten theaverage path length of the constructed network, thus an implicit trust networkconformed to the structural properties of small-world is constructed.Secondly, an adaptive recommend algorithm is presented based on theconstructed network. We first locate the clustering which the target is belongs to,and an adaptive recommendation strategy is used based on the number of similarusers in this clustering. Then to search the nearest neighbor according to thetopology of the constructed network, which improves the accuracy of selectingneighbors, and the search scope of neighbors is further broadened at the same time,so that the sparsity problem is overcome effectively.Finally, we compare the experimental evaluations and analysis of the algorithms proposed in this paper with the traditional methods on data sets withdifferent sparsity.