IMPROVED TEXT FEATURE SELECTION ALGORITHMS IN CLASSIFICATION SEARCH OF ENVIRONMENTAL PROTECTION INFORMATION

The improvement of classification performance can improve the accuracy of data search engine. Through classification, we can find the information we need from the vast amount of network information to improve the search accuracy. Traditional text categorisation takes text search and selection as its basic algorithm, and its efficiency of search and classification is not high. In this paper, the authors studies are directed to an improved text feature selection algorithms based on decentralisation optimisation and its application in classification search of environmental protection information. This paper focuses on feature selection and discusses text categorisation. Aiming at the phenomenon that the concentration degree in TFFS algorithm tends to cause low frequent feature items to produce high weights and the role of feature items whose dispersion neglects the negative mutual information value in the categorisation, this paper improves the concentration and dispersion degree in TFFS. Finally, an improved feature selection is proposed based on the length information of feature items. The proposed algorithm reduces the weight of low frequent feature items and strengthens the role of negative feature in classification.