Research Of Distributed Recommendation Method Based On Explicit And Implicit Data Of O2O

The online and offline transactions become more frequent in these years and the O2O industry is growing at an unprecedented pace.What come with it is the data that containing massive tremendous opportunities.After effectively processing these data and extracting a large number of related information between user and production,users can be provided with precise recommendations.However,studies find that O2O data is large in amount and variable in type,with fast speed and low density in generation and unbalance in implicit data sample,which bring much trouble to recommendation.Therefore,this paper proposes a scheme of big-data mixed recommendation system for explicit and implicit data of O2O,and designs the corresponding distributed algorithm.The scheme firstly gives a recommendation algorithm combining with the clustering and user collaborative filtering to deal with the O2O explicit data.Through establishing a cold start warehouse,preprocessing of matrix decomposition,training clustering model and optimal clustering,the data sparse and cold start problem has been improved,which narrows the computing range of adjacent user collaborative filtering and improves the whole efficiency.Then the implicit feedback data is recommended based on choice tendency,which is combined with explicit recommendation result,bringing out a new variable weight hybrid strategy.The innovations in this paper are as follows:1)The model-based cluster analysis is assimilated into User CF recommendation algorithm.Therefore,the computing range of adjacent user in collaborative filtering is narrowed much accurately,which heightens efficiency and improves the quality of real-time in recommendation system.2)The cold start warehouse strategy for new users is set before User CF algorithm.It solves the O2O data cold start problem by making TOP-N recommendation to new user cold boot question warehouse through computing the cold boot impact factor.Use ALS matrix factorization to improve the spare data problem in O2O and the Canopy crude clustering improves cluster procedure which makes cluster model more stable.3)A new explicit and implicit data variable weighting hybrid strategy is proposed.The introduction of the mixed recommendation of implicit feedback data and explicit data makes the weights recommendation more individual and targeted.Establishing the public weight in first-time recommendation and then set the user's individual weight,so the weight of explicit and implicit data is controlled by user himself.The visualization of recommending improvement feeds back to the improvement of public weight,which is a variable weight autonomic learning individual hybrid recommending strategy.Finally,the system scheme is tested and evaluated in this paper.When the data quantity is larger than a hundred thousand,the system MAE(Mean Absolute Error)reduces about 7.5% than that of traditional algorithm,while system MAP(Mean Average Precision)increases 4.64% than that of SPCF(Similarity Propagation based Collaborative Filterng)and 3.27% than that of Distributed User CF.The system recall rate is 11.7%,which is tested based on MPR(Mean Percentage Ranking).Thus the enhancing performance of the recommending system in this paper has been verified.