(1) Human-Automation Trust: All-Encompassing or Category-Based?
Ryu, H. & Lee, K. "Human-Automation Trust: All-Encompassing or Category-Based?", European Academy of Management 2018, June 2018.
- Key Research Question: Is trust for automated recommendations identical for all recommender system categories or not, and what is human-automation trust composed of?
- Methods Used: Scraping (Selenium, BeautifulSoup), TF-IDF, Decision Tree and Gaussian Naive Bayes classifiers, Euclidean distance, Sentiment VADER, PCA), surveys, Gephi
Agents within teams are not limited to humans subsequent to the fourth industrial revolution. To utilize social capital theory in the industry 4.0, trust in social capital theories needed to be defined in a more conspicuous manner than the status quo. The previous literatures dealing with social capital theories and its usage of trust as a social capital defined trust as an interpersonal trust that was generally understood between humans. With this obscure definition of trust, application of such social capital theory to human-automation agent teams were not effective. To define human-automation trust for its usage in the application of social capital theory to the industry 4.0, four major steps were taken. The first step was clustering trust and the second step was clustering recommender systems. Subsequently, the recommender system clusters were analyzed in aspects such as part-of-speech, sentiments and distribution. Lastly, the recommender system clusters were paired with the trust clusters. The results were that all the recommender system clusters were in accordance with trust regarding validity, transparency, and previous experience.
(2) Prognostic Impact of Social Events' Past Categorization
Ryu, H., Jo, J. & Lee, K. "Prognostic Impact of Social Events' Past Categorization", Cybercommunication Academic Society 2019. Jeju Island, Korea, May 2019
- Key Research Question: How can we investigate the social area membership change of the same event within years to observe the change of social attitudes towards the event?
- Methods Used: Scraping (Selenium, Beautiful Soup), noun extraction with Komoran morpheme analyzer, t-SNE, TF-IDF, Visualization(matplotlib), Multinomial Naive-Bayes and Support Vector Machine classifiers
Another research I have conducted was on devising a new methodology for examining the events’ impact of policies on social areas e.g. categorical change of impact from 'societal issues' to 'political issues'. After scraping 447 reports and 53,042 reports from National Knowledge Information System and Korea Policy News respectively, I used Komoran morpheme analyzer to extract the nouns for which I calculated the importance using TF-IDF and visualized the top hundred words using t-SNE. In order to examine the change in impact for events on social areas, I made Multinomial Naive-Bayes classifier and Support Vector Machine classifier to calculate the probability of each report to each of the four relevant ministries with the training data being reports from 2008 to 2009 and the testing data being those from 2010 to 2012. By visualizing the four-dimensional results into two-dimensional plots using matplotlib, I examined the change of membership of the articles.
(3) GoPizza Brand Analysis
- Key Research Question: How is the GoPizza brand perceived differently amongst its branches and how does it differ from its five competitors?
- Methods Used: Scraping (Selenium, BeautifulSoup), kkma morpheme analyzer, t-SNE, TF-IDF, Latent Dirichlet Allocation, pyLDAvis, Visualization (matplotlib, seaboard)
Brand analysis of GoPizza was conducted by scraping the reviews for each of its branches from Yogiyo, a popular delivery and review site in Korea. After preprocessing the scraped reviews, the important nouns for each branch were extracted using kkma morpheme analyzer and TF-IDF. The results were examined through visualization using matplotlib and seaborn for each of the branches. Subsequently, the topic clusters within the reviews were investigated through topic modeling with Latent Dirichlet Allocation and visualizing it with pyLDAvis. The same process was repeated for the reviews of GoPizza's five main competitors and the results were compared to scrutinize GoPizza's brand position amongst its competitors.
(4) Are We There Yet? Analyzing Scientific Research related to COVID-19 Drug Repurposing
Park, N., Ryu, H., Ding, Y., Yu, Q., Bu, Y., Wang, Q., Yang, J. and Song, M. “Are We There Yet? Analyzing Scientific Research related to COVID-19 Drug Repurposing,” Accepted to: ISSI 2021
Key Research Question: Are there explicit and implicit consensus among COVID-19 drug repurposing publications which may lead to the expedition of COVID-19 cure discovery?
- Visualized the derived conspicuous consensus on which drugs to focus on in COVID-19 drug repurposing using document clustering, a TF-IDF matrix, and co-occurrence of drug entities
- With the five clustered documents, formed drug entity co-occurrence networks for the clustered papers by setting the nodes as entity instances and the edges as the number of co-occurrences between the entity instances, making an entity co-occurrence network
- Calculated the degree, betweenness, and closeness centrality values for each node, and cut the values into terciles to examine low-high centrality combinations for the drug entity to detect inconspicuous consensus among the publications