Research On Recommendation Algorithm Based On Variable Precision Fuzzy Concept Lattice

Recommendation system can find the most valuable information from many data,and it is an effective means to deal with information overload.Formal Concept Analysis(FCA)is a theoretical system put forward by Wille based on order theory.It models binary relational data as formal context,strongly related object attribute pairs as formal concepts,and organizes all concepts into concept lattice.In recent years,to solve the problem of data sparsity,as an excellent tool for data analysis and rule extraction,formal concept analysis has been introduced into recommendation.However,the traditional concept lattice structure still has many shortcomings when it is applied to recommendation,such as low evaluation indexes such as accuracy and recall,rough concept lattice structure and excessive time complexity.Aiming at these problems,this thesis has done research and improvement,and the main work includes the following aspects:Improve the structure of concept lattice,the core data structure of formal concept analysis,to make it more suitable for recommendation.Firstly,the classical formal context is transformed into fuzzy formal context by using the scoring information.On this basis,the Crisp-crisp variable precision concept lattice structure constructed with fuzzy context is modified,and the global variable precision threshold K is transformed into a threshold set K,and each element in the set K corresponds to a user,which is used as the individual variable precision threshold of this user.Finally,a new Crisp-crisp variable precision concept lattice structure is generated,and it is named as object threshold variable precision concept lattice.A series of experiments on classic movie data set MovieLens show the effectiveness of this structure.Processing the movie type data in the data set,combining the properties of pessimistic concept lattice and optimistic concept lattice,transforming the movie type data into movie type formal context,generating movie type concept lattice,and finally calculating the similarity with object threshold concept lattice and movie type concept lattice to make recommend for users.The experiment proves that this recommendation method is better than that of using object threshold variable precision concept lattice alone.Aiming at the low efficiency of concept lattice construction when there is a large amount of data,this thesis applies concept lattice with variable precision of object threshold to heuristic recommendation,and proposes a heuristic recommendation method based on the concept of variable precision of object threshold.Firstly,the fuzzy formal context is transformed into the variable precision formal context with variable object threshold,and the heuristic information of strong concept construction is defined on the basis of this formal context.Then,using connotation constraints,on the basis of ensuring the similarity of groups,the concepts with the largest current weighted area are constructed,and these concepts are formed into a strong concept set.Finally,the recommendation method is used to make recommend for all users in this concept set.A series of experimental results on MovieLens series data sets and FilmTrust data sets prove that this optimization method is effective and can greatly improve the accuracy and performance of recommendation.