The Research And Implementation Of Synonyms Mining Method Based On The Search Log And Click Log

With the development of technology of internet, the e-commerce come to be mature gradually. Regarding searching engine in internal e-commerce site, low efficiency is a common problem. This paper is accomplished on the basis of comprehending the intention of end users. The paper is based on the research and realization of end user's log-searching and near-synonyms mining of log clicking. The near-synonyms chart is abstracted for the automatic recognition of near-synonyms, aiming at the current feature of internet domain that the number of new words, typos, and near-synonyms is increasing sharply.In this paper, at first, one candidates set is achieved by two ways, namely mode of cutting commodity titles and concentrating searching based on SimRank idea. Meanwhile, one initial chart is abstracted on searching and clicking log with the method of Chinese- English and Chinese-Chinese minings, then feature is abstracted. This paper concentrates on demonstrating literal feature, topic feature, searching feature, clicking feature, and at last The value of every feature of the word matches in the initial words is calculated for the training the machine learning model. Finally, screening near-synonyms in the candidates set to generate one near-synonyms chart.This near-synonyms chart is applied for the company and a good feedback of debug and function is sent by the company. SVM and GBDT are used to train vector machines model, and then achieve the decided threshold value of near-synonyms. GBDT is better than SVM. The GBDT result of experiment is that, accuracy rate is 56.52%, recall rate is 27.37%.