1. Research background

Topic modeling is an unsupervised learning method used in text mining to mine corpus topics, and LDA models are favored for their simplicity and wide application. T ö rnberg uses LDA model to study social media discourse and believes that it is suitable for processing large-scale data and can structurally summarize data. Jo explored the application of topic modeling in discourse analysis, emphasizing its ability to reveal high-frequency vocabulary and inter word relationship networks in discourse, understand discourse dynamics, and track historical changes. Jacobs U0026 Tsch ö tschel pointed out that topic modeling can compensate for the subjectivity, lack of systematicity, and operability of discourse analysis. Aranda et al. proposed that the combination of critical discourse analysis and structural thematic modeling can expand traditional methods. However, corpus based critical discourse analysis rarely uses topic modeling and often employs topic word methods. This study aims to fill this gap by analyzing the English coverage of the Beijing Winter Olympics through LDA thematic modeling, and combining it with the extended meaning unit model to reveal the construction of discourse meaning from both macro and micro levels.

2 Implicit Dirichlet Distribution (LDA) Theme Model

The LDA topic model was proposed by Blei et al. in 2003 and developed based on the probabilistic latent semantic indexing model, used for generating probabilistic models of text corpora. The LDA model solves the posterior distribution of topics by combining the Dirichlet prior distribution of topics and words with observed data. The model assumes that a document is composed of multiple topics, each of which is a probability distribution of a word. The process of generating words in a document involves randomly generating a probability distribution of topics, and then selecting topics and words based on this distribution. The task of the LDA model is to infer implicit topic structures based on observable words. In practice, LDA topic modeling is usually implemented by third-party open-source toolkits such as Python's Gensim and scikit learn, as well as R's Mallet package. The setting of the number of topics has a significant impact on the analysis results and needs to be balanced based on the research objectives and characteristics of the corpus.

Research on the Discourse of the Beijing Winter Olympics

Introduced the steps of corpus collection and analysis for discourse research on the Beijing Winter Olympics. The research used the Factiva database to collect overseas English news reports from July 31, 2015 to September 1, 2021 through specific keyword searches, forming the Beijing Winter Olympics corpus. The research steps include cleaning the corpus, using LDA model for topic mining, determining the number of topics and keywords, using AntConc software to analyze high-frequency collocations and semantic rhymes, and combining collocations, semantic rhymes, and word indexes to analyze discourse construction and media attitudes.

4 Research Results and Discussion

The LDA thematic modeling of the Beijing Winter Olympics corpus reveals three main reporting themes: Winter Olympics hosting rights, event preparation work, and government measures. Keywords such as Beijing, host, bid, Games, winter, etc. are related to hosting rights, while snow, event, ski, venue, etc. are related to event preparation, and sport, mill, bill, etc. are related to government funding support. Further analysis shows that when 'host' is used as a verb, reports related to the 2022 Winter Olympics mostly focus on Beijing's acquisition of the hosting rights, and there is a semantic rhyme praising Beijing as the 'dual Olympic city'. When snow is used as a noun, its high-frequency collocations such as natural, artificial, man made, real, fake, etc. focus on the way snow is made, expressing criticism of Beijing's lack of natural snow and questioning of artificial snow making. When sports is used as a plural noun, its collocations are divided into spiritual and cultural, economic activity, and promotion and development categories. Among them, spiritual and cultural collocations often express criticism of Beijing's lack of winter sports traditions, while economic activity and promotion and development collocations demonstrate the positive efforts made by the Chinese government to develop ice and snow sports.

5 Conclusion

This study analyzed the theme and discourse meaning construction of foreign media's English coverage of the Beijing Winter Olympics through LDA topic modeling. Research has found that reports mainly focus on hosting rights, event preparation, and government measures. Foreign media hold a positive attitude towards Beijing's status as the "Double Olympic City" and the Chinese government's efforts to promote ice and snow sports, but question Beijing's climate conditions and cultural heritage of ice and snow sports. The study also shows that the LDA model has objectivity and reliability in critical discourse analysis, which helps to deeply analyze the meaning of discourse construction from both macro and micro levels. Future research can combine more methods to analyze media discourse.