Recommendation Based On Three-way Concept Analysis

The recommendation system can help users quickly find the information they are interested in in the massive information,and the recommendation system has been widely used in many fields.Firstly,The traditional recommendation system divides the recommendation status into "recommended" or "not recommended" according to the predicted score.This is twoway classification,ignores various types of cost in the recommendation.Secondly,the traditional recommendation does not fully mine and utilize the value of the scoring,which makes the recommendation results affected by data sparsity,and it is difficult to learn user preferences;Thirdly,in the traditional recommendation,only the similarity association between items and items is considered,that is,positive association,while the negative association between items and items is ignored.FCA(Formal concept analysis)has unique advantages in data analysis.The recommendation based on FCA can not only establish the correlation between items,but also alleviate the problem of data sparsity,thereby improving the recommendation quality.However,FCA ignores the negative relationship between object set and attribute set.3WCA(Three-way concept analysis)is an extension of FCA,which has the advantage of FCA and can express "commonly not possessed" in the formal context.The recommendation based on the FCA can not only fully mine the positive and negative relationships between items from the scoring data,reduce the impact of data sparsity,but also pay attention to the cost and reduce the recommendation cost by three-way classification.Based on 3WCA,this paper studies the recommendation,focusing on the cost-sensitive recommendation algorithm.First,we study the construction of three-way approximate concept lattices based on the scoring information,and the extraction method of positive and negative association rules in the three-way approximate concept lattices,thereby give an association rule extraction algorithm based on 3WCA,namely 3ARM;On this basis,combining the three-way decision,the recommendation with the least recommendation cost is studied,and a recommendation algorithm based on 3WCA is proposed,namely 3WRE.In this article,the main contents are as follows:(1)According to the value of the scoring,the scoring information is transformed into an incomplete form context,which represents the three attitudes of users to the item in the real situation,like,dislike,and uncertainty;combined with three-way concept lattice construction algorithms Cb O3 C and three-way approximate concept lattice construction algorithm Norris?OE to construct three-way approximate concept lattices.On the three-way approximate concept lattices,the extraction of association rules is studied,and the 3ARM algorithm is proposed.The algorithm uses the characteristics of closed itemset in the threeway approximate concept lattice and the partial order relationship between the three-way approximate concepts,extracts positive and negative association rules from the parent-child relationship and sibling relationship of the three-way approximate concepts.(2)In recommendation based on association rules,the consequence of the rule is generally directly used as the result of the recommendation,or the support and confidence of the rule are used to simply calculate the user's preference for the item.The 3WRE algorithm considers the positive and negative correlations between item sets,and calculates the user's preference for items based on the support and confidence of the relevant positive and negative association rules.The 3WRE algorithm introduces three-way decision,according to the recommendation cost generated in each situation,the division threshold of each decision domain that can minimize the overall recommendation cost is calculated,and the corresponding recommendation behavior is determined by the user's preference for the item and the division threshold: "early recommendation","not recommended" or "pending",and finally complete the recommendation to the target user.Finally,the article uses the classic data set Movie Lens to experiment.First of all,3ARM is verified experimentally and compared with FARM algorithm and FISM algorithm.The experimental results show that 3ARM can simultaneously extract relatively complete and non-redundant positive and negative association rules in less time.Secondly,3WRE is analyzed experimentally,and compared with the collaborative filtering,the content-based recommendation,the recommendation based on FCA,and the recommendation based on three-way decision.The experimental results show that,compared with the above recommendations,3WRE has higher accuracy,recall and coverage,and lower average cost.