Abstract: Text classification technology is an important basis of information retrieval and text mining, and its main task is to mark category according to a given category set.Text classification has a wide range of applications in
natural language processing and understanding, information organization and management, information filtering and other areas. At present，text classification can be mainly divided into three groups; based on statistical methods,
based on connection method and the method based on rules. The basic idea of the traditional association text classification algorithm associative rule-based classifier by
category(ARC-BC) is to use the association rule mining algorithm Apriori which generates frequent items that appear frequently feature items or itemsets, and then use these frequent items as rule antecedent and category is used
as rule consequent to form the rule set and then make these rules constitute a classifier. During classifying the test samples, if the test sample matches the rule antecedent, put the rule that belongs to the class counterm to the
cumulative confidence. if the confidence of the category counter is the maximum, then determine the test sample belongs to that catcgory.However，ARC-BC algorithm has two main drawbacks;(1)During the structure classifier, it only considers the
existence of feature words and ignores the weight of text features for mining frequent itemsets and generated association rules may affect the classification results; (2) In the class prediction stage, it gives too much emphasis
on rule confidence. In the mining process, there will be ruck that have the same antecedent but different consequent, and if only considering the rules’confidence in predicting the impact of text classification, without
considering the correlation between rules antecedent and consequdent, it will also affect the classification accuracy. In order to solve the two problems，in this paper, a new algorithm itemset significance-based a ssociation rule-based
categorizer(ISARC) is proposed on the basis of the ARC-BC. The new algorithm ISARC makes use of features weight to define the significance of k-itemset, mining the
significant itemset to generate association rules.The main idea of SARC is given below: Firstly, preprocess the text，each document corresponds to a transaction, each unique keyword is assigned an item. Secondly, weight the
features according to the to frequency-inverse document frequcncy(TF-lDF)，and calculate each transaction weight as WT according to the defined formula, and then to be 1-important itemsets as Ll.The function of Ucne-CSigitem
is to generate candidate itemsets and wcightsig function is to calculate for the importance of each candidate set，that meeting the minimum set of items is as an important item set. Then, in the class stage，it considers the lift towards
the impact of the text classification. Finally, the experimental results show that mining the significant itemset of ISARC algorithm can improve the accuracy of the association classification.