Study On Context-integrated Personalized Recommendation Method In O2O Environment

With the development of mobile Internet and social marketing,as well as the continuous integration of online and offline resources,O2O has become an indispensable consumption form of our daily life.However,information overload increasingly becomes an urgent problem in 020 environment along with the sharp increase in the number of goods and services.At the same time,in the 020 environment,the interaction between online and offline is more contextual,the user's personalized demand has context-sensitivity,and its demand is changing with the change of the context.Under this circumstance,how to effectively mine the user's preferences in different contexts,find the resources required by the user from the rapid growth of goods and services,and solve the information overload problem,has become an important topic which needs to be further researched in the field of context-integrated personalized recommendation.Context-integrated personalized recommendation needs to integrate contextual information into recommendation,analyze contextual preferences of users,and recommend resources which meet the needs of the current context.Therefore,the key problems to realize effective personalized recommendation are to determine which context information to be integrate,how to integrate,and how to effectively mine the user's contextual preferences.This paper focuses on personalized recommendation in 020 environment,according to the weakness of post-filtering recommendation algorithm considering context correlation probability,using TF-IDF algorithm to mine user's contextual preferences,and proposing the contextual post-filtering recommendation algorithm which based on TF-IDF algorithm.Besides,considering the situation that the contextual post-filtering recommendation paradigm ignores the association between context and users/items,which will destroy data integrity,this paper proposes a heuristic recommendation algorithm based on ant colony clustering.In this paper,the main research work is as follows:1.Research on the problem of context-aware computation in 020 environment.Based on the analysis of the contextual information of consumption in 020 environment,obtaining the corresponding contextual information by explicit or implicit method.At the same time,taking the advantage of information entropy theory to identify the effective contextual information in different 020 environments to reduce the interference of invalid contexts.And according to the high frequency consumer behavior,the AVG(average value)method is used to calculate the aggregation in different dimensions.Then,using dimension modeling to establish a personalized recommendation model of the contextual data in 020 environment.2.Research on the problem of mining the user's context preference in 020 environment.When establishing the contextual preference prediction model,the existing researches only consider the association probability between the user and the project under the current context,but ignore the impact on the current context when user chooses the project in other contexts.In this paper,we adopt the idea of TF-IDF algorithm,combining the association probability between the user and the project under the current context with universal context importance to establish context preference prediction model.It will improve the accuracy of mining the user's context preference.3.Research on the contextual post-filtering recommendation paradigm in 020 environment.On the basis of traditional initial prediction score,this study adopts the contextual post-filtering recommendation paradigm,proposes a contextual post-filtering recommendation algorithm which based on TF-IDF model,and uses the TF-IDF contextual preference prediction model to adjust and generate the final recommendation.During the initial score prediction,aiming at the shortcomings of traditional similarity computing,searching the nearest neighbor by weighted combination of item category preference similarity and project score similarity,which improves the accuracy of the initial prediction score.During the post context filtering,using TF-IDF contextual preference prediction model to adjust the initial score.Since two methods of direct filtration and score correction have its own advantages and disadvantages,this paper will combine the two methods in order to improve the accuracy of the prediction score.4.Research on context modeling recommended paradigm in 020 environment.The contextual post-filtering recommendation paradigm isolates the association between the context and users/items,resulting in incomplete data set.To solve this problem,our research adopts a heuristic recommendation method,based on constructing data sets similar to current situational preferences through ant colony clustering,using multidimensional distance calculation to predict the user score for each item in current context.In this way,recommendation quality as well as user satisfaction will be improved.5.Research on experiment and evaluation of the context-integrated personalized recommendation in 020 environment.This chapter tests algorithms developed in this paper.Through the calculation of contextual information entropy,five contextual factors have been proved to have greater impact on user's purchase decision:region,daily schedule,weather,mood,and companion.Meanwhile,experiments show that the contextual post-filtering recommendation algorithm which based on TF-IDF model has significant advantages compared with the traditional recommendation algorithm.The contextual post-filtering recommendation algorithm which based on TF-IDF model is much more efficient than the post-filtering recommendation algorithm considering context correlation probability which just considers the current context preference.Therefore,Heuristic recommendation algorithm based on Ant Colony Clustering is much better compared with the contextual post-filtering recommendation.algorithm based on TF-IDF model.