本研究以三個階段建立中文情緒詞資料庫，階段一情緒詞彙收集，以中文分詞庫及詞典分析、大學生自由聯想等多個來源，獲得218個情緒描述詞與395個情緒誘發詞。階段二以353位大學生針對情緒描述詞進行正負向性、激發水準、持續時間、自主程度、出現頻率及典型性等六個向度的九點評量。1,673位大學生針對情緒誘發詞進行正負向性、持續時間、自主程度及出現頻率等四個向度的九點評量。評量結果以全體、男性與女性的平均評量值與標準差呈現。階段三以相關及群落分析進行二類情緒詞之分析，主要發現為（一）情緒描述詞中，正負向性與自主程度、出現頻率的相關最高，激發程度與典型性有顯著正相關，但與持續時間與自主程度皆有小效果量的負相關，出現頻率除了與激發水準無關外，與其他向度皆有中到高的正相關。（二）情緒誘發詞的四個向度間皆有中等以上的正相關，其中正負向與自主程度的相關最高。（三）以六個基本情緒詞的向度平均為中心，進行群落分析，結果分類後與中心點差距大於2的詞占21.1%、21.8%，若以分類加疊代方式則有1.8%、1.5%個詞大於2，且原六個基本情緒詞有多個被歸於同一類別。此可能顯示六種基本情緒詞之獨立性不足，無法涵蓋完整的情緒詞範疇，華人文化下之基本情緒類型尚待更深入研究。

This study attempts to construct a Chinese emotion word database in three stages. It started first with a collection of emotion words. The words were collected mainly from four sources: (1) an analysis of the Sinica Balanced Corpus; (2) emotion words used in theses for master degrees; (3) analyses of Chinese dictionaries by emotion types defined across dimensions of Chinese culture such as Confucianism, Buddhism, and Traditional Chinese Medicine; and (4) free associations of the six basic emotion words from 120 college students. As an emotion word can either describe an emotional state or induce it, all collected words were categorized into emotion-describing and emotion-inducing words. Our analyses yielded a total of 218 emotion-describing words and 395 emotion-inducing words. In the second stage, on a 9-point Likert scale, 353 participants rated each of the emotion-describing words for valence, arousal, continuance, dominance, frequency, and typicality; 1,673 participants rated each of the emotion-inducing words for valence, continuance, dominance and frequency. As significant gender differences were noted in the subjective feeling for certain emotion words, the means and standard deviations of ratings in the database are presented respectively for the entire sample and separately for both genders. The third stage of the study analyzed the two types of words on their clustering in various dimensions. The main findings were: (1) for emotion-describing words, valence had the highest correlations with dominance and frequency, arousal had higher correlations with typicality, but had a small negative correlation with continuance and dominance. Frequency had medium to high positive correlation with all of the dimensions except for arousal, with which no significant correlation was found. (2) Each of all four dimensions rated for emotion-inducing words had medium to high positive correlation with the other three, in which valence and dominance had the strongest correlation. (3) Two methods of K-means cluster analysis (classify-only and classify with-iteration) were used to categorize the emotion words. (a) The first method (classify-only) with the six basic emotion words as centroids revealed that 21.1%, 21.8% of the words had a distance greater than 2 from the centroids. (b) When classified with iteration, 1.8%, 1.5% of the words had a distance greater than 2 from the final centroids. Further, after iteration the six basic emotion words were classified into three to four groups. A possible implication is that we need more research for basic emotional types in Chinese culture.