Several debates were held among presidential and vice presidential candidates to convey their ideas for the 2024 presidential general election (PEMILU). This research analyzes the topics discussed in the debates using Latent Dirichlet Allocation (LDA), K-means Clustering, and word tagging methods for each candidate pair. The K-Means Clustering method yielded more diverse and evenly distributed topics for each candidate pair, while LDA produced fewer topics but was more effective in identifying topics for candidate A. The K- Means Clustering method yielded more diverse and evenly distributed topics for each candidate pair, while LDA produced fewer topics but was more effective in identifying topics for candidate. These are somewhat consistent with previos works. A. In dataset 1 using the LDA model, candidate pairs A have a probability of 60%, B have a probability of 25%, and C have a probability of 0%. In dataset 2 using the K-Means model, candidate pairs A have a probability of 37.04%, B have a probability of 25%, and C have a probability of 17.24%. In dataset 2 using the LDA model, candidate pairs A have a probability of 100%, B have a probability of 40%, and C have a probability of 0%. In dataset 2 using the K-Means model, candidate pairs A have a probability of 35.71%, B have a probability of 14.29%, and C have a probability of 28.57%.