Research And Application Of Recommendation Algorithms Based On Formal Concept Analysis

As an effective mean for dealing with information overload,the recommendation system has been extensive ly researched and developed in recent years.Recommender system is constantly emerging in the field of cases of the successful applications,but is still facing many problems to be solved.The concept lattices is the core data structure of formal concept analysis(FCA),and is the effective tool for data analysis and rule extraction.The extension and intension are the component of the concept,which makes the formal concept show the characteristics of clustering.The partial order relations among concepts also reveal the essence of generalization and specialization.With the deepening of the research,formal concept analysis has been gradually applied in the field of data mining,information retrieval and so on.Collaborative filtering is one of the most widely used recommendation strategy.The classic collaborative filtering algorithms based on neighborhood are usually only consider the similar relations between users or items,while ignoring the inherent relations between objects.In addition,more and more researchers have found that the recommendation system is often faced with the implicit data which cannot directly reflect the user preference.And with various kinds of products increase,the implicit data generated b y users and items will become extremely sparse.Therefore,due to the lack of informations in the sparse data environment,collaborative filtering algorithms can not get enough neighborhood informations,which directly affects the final recommendation.In view of the above problems,this thesis proposes a conceptual neighborhood-based collaborative filtering algorithm(CNCF),which is mainly based on the concept neighborhood for the implicit data.In this algorithm,the concept lattice is used as the data carrier to solve the Top-N recommendation problem.First of all,we construct concept lattice based on the formal context converted from the relational data between users and items.Users and products are respectively in the forms of objects and attributes,and gathered in the concept.Then based on the concept lattices,the initial concept index is generated and futher improve the efficiency of the location of the initial concept.Afterwards,with the help of the partial order relations among concepts,the initial concepts of objects(users)are used as the starting points to explore the neighbor concepts and obtain the candidate set.Finally,the proposed global preference and neighborhood preference are used to select the final list of recommendation that the users may be interested in.Through the realization of the CNCF algorithm,and carrying out experiments on two real data sets,compared with the traditional collaborative filtering algorithms based on neighborhood,CNCF algorithm is more suitable for the recommendation under the sparse data environment while maintaining better recommendation effects.