Service Recommendation Based On The User Similarity Under The O2O E-commerce Perspective

O2O e-commerce has an unique mode of combination of online and offline.This made it not only has the advantages of traditional e-commerce,but also allows consumers to experience the fun of offline consumption.How to maximize the advantages of these two areas is the problem of the O2O e-commerce facing now.The application of service recommendation model in traditional e-commerce is relatively mature.It makes it easier for consumers to find the products they like when shopping online and shortens the shopping period.It not only improved customer satisfaction,but also enhanced the competitiveness of enterprises.O2O e-commerce's section of buying online is similar to the traditional e-commerce.Therefore,giving the O2O e-commerce an applicable service recommendation model will also get good results.Based on the above considerations,this thesis determine the target user's neighbor user by measuring user's history consumption position similarity,the similarity of user's interest to the project categories and the similarity of user ratings.Then generate service recommendation.When measuring the similarity of user's history consumption position,cluster all consumer position by K-means clustering algorithm firstly and getting a number of consumption domains.Then get the history consumption location similarity by taking advantage of the times user accessing every consumer domains.In the thesis,the traditional collaborative filtering algorithm is improved.The new algorithm give the method of how to measure the similarity of user's interest to the project categories and the similarity of user ratings.When measuring the similarity of user's interest to the project categories,we should have the project categories.Using the times of user ratings for each category,calculate the users degree of interest.Whether have the same degree of interest in the same category determines the similarity between users.The similarity of user ratings is measured by the user's ratings for various purposes.We can get the neighbor user by three times similarity measurement.By using the neighbor user's score to each project,we can speculate the target user's score for ungraded items.Lastly select the projects which have a higher score as the target user's recommended items.In this thesis,the service recommendation model is constructed by using of the above recommendation algorithm.In order to solve the "cold start" problem occurs when a new user into the system,this model will make a distinction between the new user and old user.This model is built for O2O e-commerce to make service recommendation.Compared to the direct use of traditional e-commerce service recommendation model,it not only takes into account the impact of user's offline consumption position may generated to the user's purchase intention,but also recognizes the drawbacks of the method that getting the service recommendation by the direct using of all users scores.When this model is used in O2O e-commerce environment,the recommendation will have a higher efficiency and the result will be more accurate.The last part of this thesis is empirical research by using some consumer records of a buy site.Firstly,make a data preprocessing.Secondly,select any user as the target user.Then,get the final neighbor users by measuring user's history consumption position similarity,the similarity of user's interest to the project categories and the similarity of user ratings.Lastly,according to the neighbor users' ratings for each restaurants,produce the restaurants that will recommend to the target user.In the end,through the analysis of the results,prove the service recommendation model constructed in this thesis may be more accurate for users of the service recommended.