Research On The Discovery And Analysis Of Internet Financial Advertisements

Internet financial advertisements with the characteristics of fast spread,wide coverage,and difficulty in tracing the source,has become the focus of attention of relevant regulatory departments.It is urgent to establish a long-term governance mechanism for financial advertising.In this paper,the search term database construction of advertisements acquisition,financial advertisements recognition and financial advertisements analysis are studied.And a prototype system of Internet financial advertisements discovery and analysis is designed and realized,providing data and technical support for the governance of financial advertisements.Firstly,the paper proposes an unsupervised,graph-based algorithm to expand the financial search terms.The algorithm simultaneously incorporates the position of words and their frequency in texts to choose the words with financial semanteme characteristics,and involves three essential steps:(1)the graph construction at words level;(2)the clustering based on the graph;and(3)the selection of clusters based on semantic information.Through the expansion of financial terms such as loans,achieve the construction of search term database.The experiments show that the algorithm is excellent in terms of time performance,and the obtained extended words play a positive role in the acquisition of financial advertisements.Secondly,the paper proposes a financial advertisements recognition algorithm based on multi-model fusion,which improves the fitting ability of the recognition algorithm by considering the difference in expression semantics between the various text representation methods.The algorithm involves four steps:(1)pre-processing of advertisements text,(2)the expression of text semantic information in multiple text representations,(3)prediction of the expression of each text representation,and(4)fusion of multiple prediction results through integrated learning.The experiments show that the multi-model fusion recognition algorithm has stronger fitting ability than the single prediction model.Thirdly,financial advertisements are analyzed from three dimensions: source tracking,data statistical analysis and keyword extraction.Data statistical analysis includes the statistics of the geographical distribution and the statistics of advertising trends;keyword extraction includes the discovery of illegal words and the extraction of characteristic keywords.Realize the source tracing based on ICP query;realize the discovery of illegal words by finite automata machine and through experimental comparison,TextRank was selected for feature keyword extraction.Finally,a prototype system of Internet financial advertisements discovery and analysis is designed and implemented based on the above research.The prototype system mainly consists of four modules.The advertising data acquisition module carries out the Internet advertisements acquisition,including the advertising data provided by the third party and the advertising data crawled from the search engine channel.The financial advertisements recognition module filters non-financial advertisements to achieve precise identification of financial advertisements.The financial advertisements analysis module analyzes the financial advertisement from three dimensions.The financial advertisements data output module displays the original data and analysis data of the financial advertisements.