Research On Recommendation Algorithm Based On Label And Trust Relationship

Under the circumstance of big data,the problem of information overload is becoming increasingly serious than before.This poses challenges to accurately provide users with corresponding information and services they required.Recommender systems and personalized customization services,as an important technology to overcome information overload,have been widely used by users in various fields of e-commerce.Personalized recommendation stems from collaborative filtering algorithms.Many scholars and scientific research institutions have played a key role in the development of this field,and various algorithms have emerged one after another.Current recommendation algorithm is not prefect and has a lot of limitations.Many recommendation algorithms tend to ignore the impact of mu lti-source information such as labels,time and trust relationship information and so on.Even if some state-of-the-art researches are inclined to consider the effects of a single factor mentioned above.This caused a decline in the quality of recommendations and failed to provide users with satisfactory recommendations.Focusing on the above mentioned issues,aiming at tacking three tasks based on the matrix factorization and integrates labels,time and trust relationship information into this paper: the first point is building a new recommendation model on the algorithm which fusing trust relationship information based on the relevant label-based algorithm to provide target items for users;the second point is proposing a new rating prediction model which fusing time information on the basis of probability matrix factorization algorithm to predict the user's preference for the items;the third point is fusing labels,trust relationships and time into the new rating prediction model to further improve the effectiveness of recommendations.The research questions and solution in this paper can be summarized as follows:Item-based recommendation for the fusion of trust relationship information: The existing label-based recommendation algorithm takes into account the influence of time factors on the recommendation results,but does not consider the impact of trust relationship information on the recommendation results.For this reason,this paper proposes a recommendation model for the fusion of tag labels,time,and trust relationship information.The algorithm uses the matrix factorization model for item-based recommendation.The addition of trust relationship can relieve the data sparsity to a certain extent.The experimental results on the open datasets show that the model has a good recommendation effect.Rating prediction based on probability matrix factorization: On account of traditional collaborative filtering algorithms are susceptible to data sparse,a new rating prediction model TPMF is proposed based on Probabilistic Matrix Factorization.The rating matrix is factorized into two non-negative matrices,the rating results are normalized to show probabilistic semantics,then variational inference is used to compute the distribution of the real posterior distribution of this model,the user-item rating matrix is complemented by the rating prediction values and then get a dense rating matrix,finally the time weights are integrated into the rating matrix to build a user-item-time three-dimensional model to generate a recommendation result.The rating prediction model is combined with the collaborative filtering algorithm and the time window technology is also integrated into this model.The experimental results show that the rating prediction model can effectively improve the effectiveness of the recommendation.Multi-source information recommendation: To synthesize multi-source information,a new recommendation algorithm fusing labels,trust relationships and time information is proposed.By label clustering,the semantics fuzziness and redundancy in labels are eliminated to a certain extent.In this algorithm,the weight of trust relationships can make the recommendation results more targeted to a certain extent and time weight can better describe the change of user's interest.To overcome the data sparsity,the algorithm also uses the new rating prediction model TPMF to predict the user's rating.It is proved by experiments that this series of optimization measures help to improve the recommendation result to a certain degree.