Multi-layer Ensemble Framework For Group Recommendation Based On Model Fusion

Recommendations are an ancient method of screening and retrieval.The emperor of the Han Dynasty made filial piety in the Han Dynasty,and imperial examinations were recommended to select talents in the Sui Dynasty,all of which played a prominent role.In modern society,explosive information is everywhere in life.How to obtain valuable information from complicated information becomes more and more important.The emergence of recommendation system provides great help for people who are at a loss.The convenience of the system has also brought generous returns to the companies that introduced the recommendation system.However,most recommendation systems focus on recommendations for individual users and ignore recommendations for groups.In life,they often encounter group dining out,outings,parties,etc.For such scenarios,the decision needs to take into account the opinions of each member of the group to make a decision.At the same time,in the current development of artificial intelligence is not perfect,there are some problems with models,more or less,to compensate for the deficiencies of various models through multi-model fusion,to improve performance,is a commonly used method.For the personalized recommendation system,in order to solve the problem of data sparseness in the traditional matrix decomposition algorithm,the thesis introduces the comment data into the training.Mining comment information through topic analysis,improved the existing matrix decomposition algorithm and topic analysis algorithm fusion model,and proposed Dual-HFT(Dual Hidden Factors as Topic)improved algorithm.At the same time,the effectiveness of the algorithm is proved by theoretical derivation and experimental analysis.In the group recommendation system,based on the social selection theory,the paper chooses Med Rank(Median Rank)algorithm to replace the traditional ranking fusion algorithm,which avoids many problems in the traditional algorithm.At the same time,in view of the problem of ignoring member weights in the existing group selection algorithm based on social selection,the paper proposes a weighted group recommendation fusion strategy WRank(Weighted Med Rank)to make the recommendation results more realistic.Simulation experiments prove that WRank meets the rationality and authenticity of group recommendations.Based on the traditional recommendation system framework,the paper proposes a multi-layer integrated recommendation framework based on model fusion,which has a multi-step fusion process(including personalized recommendation,group recommendation and model fusion).In the recommendation process of different granularities,the paper fused the results of sub-algorithms to make up for their respective shortcomings.Particularly,for the problem that the existing model fusion method cannot be applied to the ranking fusion algorithm,the article uses new measurement indicators and calculation techniques to achieve the fusion of the model.Finally,the experimental results show that the accuracy of the recommendation framework is better than other comparison methods when performing Top-N recommendation.