The Design And Implementation Of Long-Tail Query Classification System Of E-Commerce Search Engine

Under the search environment of e-commerce, the interaction between user and system is finished by the query.As a result, it is important to judge the user query intention.Because of the Matthew effect, the top query has enough search click data, and the corresponding category prediction is very accurate. But the tail query cannot get enough search click data, which could not reach the requirements of application. Consequently, it is necessary to optimize the category prediction algorithm for the long-tail query.This thesis takes a survey on the technology of Chinese query analysis, especially about part of speech tagging. Then this thesis provides a complete solution for long tail query classification. The solution transforms the low-frequency tail query to high-frequency pattern, and make category judgement for the pattern.At last, the query is provided by retransforming the pattern.The thesis also introduces the algorithm implementation based on the Hadoop platform, including training offline tagging model, ananlysing the query patterns and predicting the category intention of query. The experiment also shows the optimzed long tail predicting algorithm has a great improvement on accuracy and recall rate,which reachs the basic online requirement.