Research And Application Of Information Extraction Technology Oriented To Operator Tariff Knowledge Map

With the rapid development of mobile communication,mobile Internet and other technologies,the demand for communication connection of users is constantly increasing.In order to meet the needs of different users,operators have launched a variety of services,forming a dynamic and complex expenses system.On the one hand,it brings difficulties to telecom operators in operation and management,which makes it difficult for telecom operators to make scientific assessments of their own expenses system;on the other hand,users are unable to choose tariff packages,and it is difficult for users to accurately select the package service that suits their needs.By establishing the tariff knowledge map,we can systematically sort out the knowledge association among the tariffs of packages,help marketers quickly locate the tariff packages required by users,and at the same time make their opponents' tariff strategies and competitive intentions knowledgeable,so that marketers can make quick decisions.Based on the requirement of constructing the operator tariff knowledge map,this paper provides a complete solution from data acquisition,data annotation,data process,tariff information extraction to the final knowledge map construction,and proves the feasibility of the solution through experiments.The constructed tariff knowledge map describes the tariff and knowledge association comprehensively and accurately,and establish the basic knowledge and information support for the intelligent evaluation and prediction of operators' tariff business.The main work and innovations of this paper are as follows:(1)The acquisition and pretreatment of tariff data.The initial data set is obtained by crawling public documents and official website data,and the tariff documents in the data set are extracted by text categorization to prepare for later tariff information extraction.According to the task of classification of tariff documents,this paper introduces artificial features based on the analysis of the characteristics of tariff documents by using chi-square value to obtain feature word vectors.Experiments show that the accuracy rate of tariff documents classification reaches more than 90%.(2)For the information extraction of tariff documents,table data and text data in the documents are extracted in different ways.Tabular data is extracted by rule plus dictionary.Text data extraction is based on BILSTM+CRF basic model.Domain word vectors and word segmentation vectors are introduced,which significantly improves the effect of entity extraction.(3)The top-down and bottom-up approach is adopted to construct operators'tariff knowledge map.By introducing the knowledge of expert,21 tariff entities and their corresponding 63 attribute relationships are defined from top to bottom,covering all tariff types on the market at present.The package knowledge extracted from information is mapped to the defined knowledge base from bottom to top to complete the construction of the tariff knowledge map.