Page 164 - AJWEP-22-5

Li and Wu

Figure 1. Trends in the issuance of support policies for emission and carbon reduction in China (2016 – 2023)

national transportation authority, the ministry overseeing
agriculture and rural development, the national economic
planning authority, and the commerce administration
department all reached their historical highs.
From the end of 2019, owing to the influence of the
general environment, the Chinese government began
prioritizing the emission and carbon reduction sector,
and various types of emission- and carbon reduction-
related policies issued from 2020 onwards have guided
the development of the emission reduction and carbon
reduction industry across multiple levels. Thus, starting
from 2021, the nation’s focus on emission and carbon
reduction intensified, which naturally increased the
regulation of emission and carbon reduction in macro
policy – marking the peak period of China’s emission Figure 2. Line graph of theme perplexity
and carbon reduction support policy issuance. 31,32
Unstructured policy texts form the core data of this observed when the number of topics is 14, indicating the
research, making data preprocessing a crucial initial best clustering performance. Therefore, the number of
step. This involves removing invalid data – such as clusters for the LDA model is set to 14.
numbers, English characters, garbled text, intonation LDA topic modeling was performed on the text
marks, punctuation, and irrelevant content like names of corpus, and the outcomes are displayed in Figures 3-5.
people, companies, and document descriptions. These In the topic clustering visualization, each circle on the
unwanted elements can be processed using regular left side corresponds to an individual topic. The greater
expressions to replace symbols or by adding a stopword the spatial separation between topics, the more distinct
dictionary to filter out unnecessary words. their content differences, indicating better classification.
On the right, the distribution of words for each topic is
4.2. Data analysis shown, with words positioned further forward having
4.2.1. LDA analysis higher frequencies within that topic.
The number of clusters is selected based on the perplexity As shown in Figures 3-5, based on the spatial layout
score of the topic model. Perplexity reflects the of each category, it is observed that most categories are
effectiveness of classification under varying topic counts distributed at considerable distances from one another,
– lower perplexity values generally indicate better topic with only two showing slight spatial overlap. Thus, the
separation and coherence. The optimal number of topics theme division across the three phases demonstrates
in the LDA model is determined based on this perplexity relatively good performance. From the 14 thematic
score. As shown in Figure 2, the lowest perplexity is clusters identified through LDA, the most prominent

Volume 22 Issue 5 (2025) 158 doi: 10.36922/AJWEP025160117