Web Search Keyword Selection Research:Keyword Concentration And Keyword Correlation Analysis

Web search for information analysis and prediction is an information analysis and prediction activity, based on the internet user’s web search behavior to explain the law of things, and to estimate the unknown trend of things. Its origin and development is inseparable from the development of the internet, especially the emergence of web search engines. Web search engines provide a huge users and vast mounts of data to information analysis and prediction. Early in this centry, foreign researchers began the empirical research of this field. Ginsberg’s research results about influenza outbreak trend is an important representative in this field.In this paper, keyword selection, which is the key process of web search for information analysis and prediction, is the research object. This paper attempts to explore the thoertical basis of web search for information analysis and prediction, to analysis keyword selection process, to select the source of keyword data. This paper tries to verify the selection effection by using keyword concentration and keyword correlation.In summary, the content of this paper is divided into seven chapters:(1) Theoretical basis of the research of web search for information analysis and predictionThis chapter describes the web search for information analysis and prediction concerpt, origin and development, effections, characteristics and five research perspectives, the internal meachnism and influencing factors. In the case of Ginsberg’s keywords table, some limitations of information analysis and prediction are founded.Finally,this chapter describes the basic process of web search for information analysis and prediction.(2) Basic process of keyword selectionThis chapter describes the basic process of keyword selection process. Keywords, datasets and analytical prediction model are the three main factors of web search for information analysis and prediction. This process includes six key parts:pre-analysis, keyword selection, correlation analysis and prediction model, verifying effect, prediction implement and dynamic correction. The keyword selection process includes seven major processes:task preparation, keyword data source selection, initial keyword selection, candidate keyword collection, observation keyword selection, final keyword table and feedback. The keyword selection process is a process from the initial keyword, the candidate keyword to the observation keyword. We found that the observation keywords are observed with typical Matthew effect and Long tail effect. These keywords discrete distribution is a Zipf distribution.(3) Keyword concentration analysisThis chapter introduces the keyword concentration, keyword moving average concentration, keyword concentration change ratio and keword moving average concentration change ratio. In the case of Ginsberg’s45keywords table, we found that keyword concentration change ratio has more effective results than keyword concentration for determing the keyword’s "stability-mutation" characteristics.(4) Keyword correlation analysisThis chapter describes the simple correlation analysis method and timing correlation analysis method. These methods include pearson, spearman, lead-lag timing correlation and peak comparsion. In the case of keyword selection about Vanke shares, we found that the pearson and spearman correlation method can effectively determine the keyword correlation degree, and the timing correlation method can effectively identify the keyword’s "lead-lag" characteristics.(5) The experimental study for keyword selection about H7N9avian fluThis chapter describes an experimental study of keyword selection about the outbreak of H7N9avian flu during the first half of2013. This experiment used two analysis methods, including keyword concentration and keyword correlation, to comparing keywords about H7N9avian flu. This experiment made three observation keyword table for different research fields.(6) SummaryThis chapter summarizes the entire research, described the expectation of the future development of web search for information analysis and prediction, and proposed three potential research questions.