Racial Filtering Classification Model Through Data Analysis of Racial Contents in Twitter

Stop Asian Hate or Stop Asian American Pacific Islanders (AAPI) Hate refers to the national movement against racially-motivated attacks on Asians. This protest was initiated in line with the Black Lives Matter (BLM) movement, to dismantle the ongoing hate and targeted crimes against Asians, and to educate people of such threats. Hate crimes targeting Asians have been occurring steadily across the U.S, but with the effect of COVID-19, these crimes started increasing in number. For the Stop Asian Hate movement, the matter was exacerbated with people accusing certain Asian countries as the source for COVID-19. In 2021, Asian Americans reported a single biggest increase in serious incidents of online hate and harassment with racist and xenophobic slurs blaming people of Asian descent for the spread of COVID-19. To specifically assess the impacts and measures of each movement, research was conducted to examine the racial slurs used towards Asians on social media, specifically Twitter. For analysis of the data on social media, Python programming was used to collect and analyze the ratio of racial slurs and Anti-Asian hate. In doing so, the data set was modeled through data labeling, which classified each social media tweet into one of three sub-categories. Data were classified into two types: type 1 that contains racial contents or information against Asians and type 0 that has non-racial contents. The data collection was done through Twint, a Python scraping tool for Twitter, gathering a total of over 2,000 recent tweets for keywords relevant to the movement. Then, a preprocessing step was taken through Python, involving the process of decapitalizing, lemmatizing, and tokenizing. These data were then represented by graphs and word clouds, displaying some of the most commonly used terms targeting Asians on social media. Lastly, the data went through a design of a binary classification model for filtering tweets with racial content. We compared the accuracy of classification models with three different algorithms: logistic regression, random forest, and SVM. The model created would be able to safeguard users from exposures to racist terms vastly pervaded on the internet.