Recent Weighted Frequent Itemset Mining For Data Streams And Applications

With the rapid development of computer and big data collection technology,data flow has become one of the most important resources for enterprises and organizations,but it also poses serious challenges to effective knowledge discovery.Pattern mining is an important subfield of data mining,which can discover implicit correlation information in data streams.However,traditional frequent itemset mining methods only consider the frequency of items in transaction databases,while ignoring other implicit factors such as project or itemset weights,interests,risks,or profits,which have shown significant limitations in practical applications.Compared to traditional frequent itemset mining,weighted frequent itemsets are more in line with user needs.However,over time,a decrease in user interest may result in outdated weighted frequent itemsets losing their relevance and ability to guide existing users.In addition,real-world applications involve multiple data streams that constantly generate objects,and some interesting observations may appear on the same object in multiple data streams.In response to the above issues,this article conducted in-depth research on frequent itemset mining for data streams,mainly including:(1)In response to the problem of traditional frequent itemset mining only considering information loss caused by item frequency and the knowledge contained in streaming data that will change over time,this paper introduces a time decay factor and adaptively adjusts the decay degree of data streams in different time domains by setting similarity thresholds.A single data stream recent weighted frequent itemset mining RWFIM-Neg(Recent Weighted Frequent itemset NegNodesets)algorithm is proposed.The RWFIM-Negalgorithm uses a more efficient data structure,NegNodeset,in the mining process to avoid complex tree building processes and tedious support calculations,improving connection efficiency.At the same time,it adopts superset equivalence and parent-child equivalence pruning strategies,effectively reducing the search space for the nearest weighted frequent patterns.(2)Due to the fact that the RWFIM-Negalgorithm is only suitable for mining recently weighted frequent itemsets in single data streams and cannot effectively solve the problem of mining recently weighted frequent itemsets in multiple data streams,this paper proposes the Collaborative Recent Weighted Frequent itemset(CRWFIM)algorithm for collaborative recently weighted frequent itemset mining in multiple data streams.The CRWFIM algorithm uses a frequent itemset mining algorithm based on bit combinations to convert data into binary and gradually increase element combinations to mine frequent itemsets and calculate support.At the same time,it records the number of times each frequent item in the result set exists in a continuous superset,and quickly prunes non frequent candidates to further improve mining efficiency.(3)We have designed a product recommendation system based on the RWFIM-Negalgorithm,which can help managers more intuitively understand user purchasing behavior and trends in product flow,thereby providing more meaningful recommendation results.