A GMM-based user model for knowledge recommendation

With the exponential increase of available information, the phenomenon of information overload has receivedextensive research attentions. Knowledge recommender system(KRS) is an efficient way to decrease information overload, andthe user model is very critical for KRS. This paper proposes amethod to establish a user model based on Gaussian MixtureModel (GMM). In detail, we first select the keywords fromknowledge databases, and then represent knowledge items withVector Space Model (VSM). Next, for a certain user, the VSM ofall scanned knowledge items and related scores rated by the userare combined together to be a new matrix, named as Vector SpaceModel with Rating(VSMR,with dimension of m times n), wherethe first n-1 columns represent the VSM of the items, and the finalcolumn lists the scores given by the user. And then the GMMbased user model is trained with VSMR. Finally, the traineduser model is used to predict the user's ratings on the knowledgeitems and the items with the higher score are considered as user'sinterest, which will be recommended to the user. The proposedmethod is validated by two experiments, which indicate that themethod works well.