Overview / Usage

2018 Kim Jong-un was high over the New Year message "The US mainland across this and in our nuclear strike within range is always placed on the nuclear button on my office desk." But just two months passed since the Pyeongchang Olympics, North Korea is to switch to a policy of appeasement The South-North summit and the North American summit were concluded.
It is expected that another round of six-party talks will be held when diplomatic actions are taken against the North after the summit. In the case of the US, Japan, and China, their attitudes toward North Korea are relatively clear on the route, and they may be able to see their news in Korea. However, in the case of Russia, after the election of Trump in the United States, it was difficult to see the public opinion of Russia because the attitude toward North Korea was not clear by choosing the pro-American route. .
The survey was largely divided into two groups: the reaction of the Russian media and the reactions of the Russian public. In the case of the Russian media, articles on North Korea were gathered from Russia's representative conservative media 'aif' and the progressive media company 'echo Moscow', based on data on Russia's North Korea in the GDELT project. Russia's public response was analyzed based on Twitter, a social network service, and comments from each press article. The former is an objective and public medium as compared with the latter, so it is easy to investigate the attitude of the Russian government to North Korea. On the other hand, the latter is not representative, but it can be expected that if a large amount of data is available, it can be reestablished and it is the latter that affects the actual politics.
All the data and codes used in this paper are published in Github. If you need a code for more data collection and investigation for a new poll on North Korea, visit "https://github.com/dongwook412/NorthKoreaReactionInRussia".

Methodology / Approach

In the methodology applied to the model, machine learning and word embedding were used. Logistic regression was used for machine learning and the words were vectorized using one-hot encoding. The word embedding uses Fasttext and Doc2vec (Document to Vector). Because the data with the label was very biased (translated by a Russian expert, the majority of the neutral data), we applied the methodologies by grouping the two labels together. We have tried several other methods such as Random Forest, RNN, and CNN, but the methods mentioned above are the best ones.

Technologies Used

We used Python's BeautifulSoup library to collect articles and comment data. Also, Twitter data was collected directly, and Twitter was collected using the api.
The Logistic Regression method uses the LogisticRegression library for sklearn.linear_model, Fasttext uses FastText for Python, and Doc2vec uses the Doc2Vec library of gensim.models.

Repository

https://github.com/dongwook412/NorthKoreaReactionInRussia