Study of an Improved Text Filter Algorithm Based on Trie Tree

摘要

There are some atypical texts hidden in the Telecom's customer complaint text. These atypical complaints can be divided into several classes. Atypical items have high confidence with intergroup neighbor, but low support in full complaint set. After filtering out the high-frequency items, we can use k-means method to clustering the complaint texts. However, the clustering result is affected by the random choosing of the original K centers and it is not accurate to extract the atypical complaint classes. This paper will propose a genetic algorithm optimized k-means method, and design a fitness function. It helps to choose the global optimum K centers for k-means method, and make the result most accurate. The improved model is more suitable for small memory systems, and it has better performance in security and dynamic adaptation. This improved model has good application value.