Research And Implementation Of Hybrid Recommendation Techniques Based On Personalized Modeling In Smart Trading Area

In the era of information overload, merchants and users in trading areaare confronting great challenges of information overflow and redundancy. Asan important part of the Smart Trading Area service platform, building anapplicable recommender system based on the scene of Smart Trading Area isan effective way of solving the above problems. However, nowadaystraditional recommender systems have some limitations:1) With severalyears of development, despite many recommendation techniques have beenproposed, but with practice, it is discovered that no methods can cover all thebases, each method has its disadvantages to some extent;2) Facing with theuser’s multi-dimensioned personalized information, no methods can make fulluse of it.In allusion to the problems mentioned above, the paper focuses on theresearch of personalized modelling techniques and hybrid recommendationtechniques, and implements a hybrid recommender prototype system basedon it. The main work of the paper is as follows:1. As no present personalized model is suitable for the scene of SmartTrading Area, we propose a feature-based, recommendation list-orientedcompatible personalized model, which can derive a recommendation list byfollowing the modeling procedures;2. Aiming at the disadvantages of individual recommendation method,we sum up three categories of hybrid recommendation techniques, andimplement paralleled hybridization and pipelined hybridization based on the effective and simple Weighted Slope One Predictor algorithm. After that weperform experiments and comparison on them;3. After comparison and analysis of the above two hybridizationmethods, we study monolithic hybridization based on modifying originalalgorithms, and propose four different hybrid algorithms of Weighted SlopeOne algorithm mixing with other types of personalized information namedMWSO, including Weighted Slope One algorithm with adding weights ofusers or items, and Weighted Slope One algorithm with adding similarity ofusers or items. Experiments show that the proposed algorithms can improvethe accuracy of Weight Slope One algorithm in most cases, and confirm theadvantages of hybrid recommendation techniques;4. Based on the research and experimental results of personalizedmodelling techniques and hybrid recommendation techniques, we discuss thepossible hybrid recommendation techniques based on the personalized modelin the scene of Smart Trading Area, and implement a prototype recommendersystem based on monolithic hybridization. From the demonstration of thesystem, it is showed that the methods of implementing are feasible and thetechniques researched are effective.