抄録
In order to visualize keyword trends embedded in newspaper articles, this paper proposes FACT-Graph (Frequency and Co-occurrence-based Trend Graph). Firstly, we introduce a trend analytical method that works by keywords class. We identify four classes of keywords by TF (Term Frequency) values and DF (Document Frequency) values in an analytical period, and then some keywords are classified into different classes by period. We pay attention to such class transition between periods and make it a clue of trend analysis. Next, we apply a method of identifying relationship between multiple words by their co-occurrence and their transition to resolve problems which occurs in prior class transition analysis. Finally, we output FACT-Graph by extending the traditional simple co-occurrence graph, which visualizes trend analysis and simultaneously examine keyword class and keyword co-occurrence relationship. FACT-Graph is based four classes of keyword, keywords co-occurrence and their transition between time periods. While each class is characterized by shapes of nodes and keywords co-occurrences relationships are represented by the types of lines, the trend transition patterns are colored. Applying the proposed FACT-Graph to the dataset of 220,000 newspaper articles, this paper exemplifies some example result and validates effective in visualizing keyword trends embedded in volumes of text.
