台灣1950年至2010年出版社演變情況之資料視覺化方法

Abstract

多重對應分析早期應用於應用於語言學研究，主要是利用該方法將獲得資料進行降維分析，讓資料可以呈現在二維或三維空間，透過視覺就能對資料進行解讀。本文探討的是多重對應分析是否真的能夠將資料原始型態在降維後呈現，實驗中的檢定方式會利用到特徵值去觀察，我們所使用的資料是台灣1950至2010年出版社資料，將討論幾種視覺化分析方法所分析出來的內容差異性，經實驗觀察多重對應分析在此資料的研究中出現解釋率不足的問題，因此再比較多元尺度分析及潛在語義索引兩種方法，另外兩種方法在解釋率上都有明顯的提升，而潛在語義索引是三種方法中結果表現最為突出的。

The early application of multiple correspondence analysis in linguistic research was mainly to use this method to perform dimensionality reduction analysis on the obtained data, so that the data can be presented in two-dimensional or three-dimensional space, and the data can be interpreted through vision. This article is discussing whether multiple correspondence analysis can really present the original form of the data after dimensionality reduction. The verification method in the experiment will use eigenvalues ​​to observe. The data we use are Taiwanese publishing houses from 1950 to 2010. The content differences analyzed by several visual analysis methods will be discussed. After experimental observation, multiple correspondence analysis has the problem of insufficient interpretation rate in the research of this data. Therefore, we will compare the two methods of multi-scale analysis and latent semantic indexing. Both methods have a significant improvement in interpretation rate, and latent semantic indexing is the most prominent result among the three methods.