Natural Language Representation Of Association Rules Based On Meta Graph

Data mining is widely used in many fields.The association rules mining is an important task in data mining.However,most representation methods of the association rules require users to have professional knowledge of data mining,which is not conducive to ordinary users to understand and apply the association rules.Based on the application of the traditional Meta graph,a new method for representing association rules with Meta graph is proposed.Moreover,Meta graph is further transformed into the natural language text based on domain knowledge base.Therefore,ordinary users can understand its meaning which meet the needs of different types of users.In this essays,the main work and conclusions are as follows:1)A representation method of association rules with Meta graph based on keyword attribute matching is proposed.The Meta graph is improved according to the characteristics of the association rules on the traditional method by expressing the relationship between attributes,the support and confidence.Firstly,keywords are extracted from the prerequisite and consequent of association rules.Then based on the attribute database of terms,part of speech and the concept attributes of the key words in association rules are acquired.Finally,the node locations in Meta graph is determined by part of speech and the relationship among the objects in Meta graph are built by restrict relations of concept attribute of key words.And the support and confidence are expressed by size and gray value of the connection point between the prerequisite and consequent of association rules.2)A method of converting association rules into natural language texts is proposed.First of all,domain knowledge base is established to make enlargement of vocabularies,syntactic definitions and sentence combination.Secondly,based on the domain knowledge base,association rules expressed by Meta graph are converted into a tree structure of the text.Furthermore,the choice of lexical and sentence elements are completed by microplanning and surface level chiefly.Finally,the natural language sentences are the final output text after grammatical modification.3)A prototype system for natural language representation of association rules is designed and implemented based on population database.It transforms the association rules which mined from the population database into natural language texts.The operation results show that the proposed method can convert association rules into natural language texts with smoothly expression for easily understanding by ordinary users.