This research collects abstracts from the International Conference on Environmental Education Academia and Practices held by the Chinese Society for Environmental Education (CSEE) between 2013-2018. Through the automatic topic classification techniques, it explores the vocabulary characteristics of classified articles in the field of environmental education and the consistency of classification. Techniques applied include natural language processing, two-step cluster analysis, word cloud, co-word analysis and association rules analysis. In this study, the research abstracts from the conference papers have been imported into the natural language processing algorithm of the CKIP Chinese Lexical Knowledge Base of Academia Sinica for word segmentation. The opinions of environmental education experts have been applied for auxiliary word segmentation, and corpora of abstracts from conference papers have been made into quantitative Term Frequency-Inverse Document Frequency (TF-IDF) weights. Afterwards, two-step cluster analysis technology has been performed to automatically classify articles clusters; the techniques of word cloud, co-word analysis and association rule analysis have been used to show the vocabulary characteristics of distinct clustered articles and the consistency of the classified articles. Based on the results of 561 abstracts of conference papers from 2013 to 2018, the number of original keywords after word segmentation is 4,980. The top 500 (10%) words account for the 74.1% of the cumulative word frequency. The selection of professional vocabularies can match the Pareto principle. The two-step cluster analysis classifies the number of K-means clusters into six categories, namely (1) environmental policy and regulation; (2) sustainable development; (3) environmental ethics and sustainable use of energy and resources; (4) disaster prevention and response, sustainable use of energy and resources; (5) climate change; (6) environmental ethics. This study applies the word cloud to enlist the dominant words with high TF-IDF weights, word frequency and proportions for distinct clusters; utilizes the cross-check method to assess the consistency of topic classification and enlists the top three article titles and keywords with the smallest distance in each category. In addition, the web map is drawn in accordance with classification results, and dominant keywords and their association rules are screened, and then dominant keywords of different themes have been suggested. For natural language word segmentation process and automatic document classification in topic modeling, the assistance of domain experts for environmental education plays a crucial role in providing correctness and consistence in aforementioned academic tasks.