joyce-huang-nus Goto Github PK

airport-hub-analysis

In this report, we analyzed the flight network worldwide. Through time series and network analytics, we investigated how the travel networks have been affected by the virus and the restrictions. However, lately as the restrictions are gradually lifted in many countries, we are expecting a recovery in tourism industry. By studying the travel networks in different stages of the regulations, hubs are identified. Coupled with the COVID policies in various countries, we aimed to optimize the airline travel operations to satisfy the needs of the newly revived tourism.

conduct-did-fixed-effect-models-and-propensity-matching-to-analyze-the-impact-of-the-volcker-rule

The Volcker Rule was announced to restrict banks from conducting certain investment activities with their own account in order to decrease risks taken by the bank, and hence to have a greater stability. Due to the restrictions, banks might argue that their profitability has decreased and therefore, assumption was made to analyze the impact of the Volcker Rule. In this project, we used the Difference-in-difference (DiD) models, including the baseline DiD model and the robust DiD models that considered the heterogeneity of different time points and individual bank holding companies by adding the fixed effect technique. In addition, we also conducted a propensity match to compare the post Volcker Rule results with the most similar results before the announcement of the Volcker Rule.

credit-card-customer-churn-prediction

To retain customers, companies are developing sophisticated churn management strategies. These strategies typically ranking customers based on their estimated propensity to churn. The implicit assumption is that this process would maximize firm's profits by targeting customers who are most likely to churn. We propose a predictive model to understand which value customers are likely to churn and should be retained.

google-lens-testing-on-expensive-watches

This experiment aimed to test the accuracy of Google Lens in recognizing rare collectable photos of expensive watches, particularly high-end watch brands. The researchers randomly selected five high-end watch models from the top 10 most popular brands and tested them using more than 30 images for each watch model.

movie-schedule-optimization-using-sentimental-analysis-of-movie-reviews

The majority of cinema income is from screening new release films. A hit film results in high ticket income and high concession sales, and vice versa. Therefore, film arrangement is a significantly essential task for cinemas. Traditionally, theaters would predict a high box office for a movie with famous producers, popular cast and high budget and allocate more screenings to these kinds of movies. This is also how audiences made their decision in the past. However, with the spread of SNS and online critics commentaries, people tend to search for comments on the Internet before buying a ticket. As a result, people’s comments become more and more important to affect the revenue of a movie

music-recommendation-filter-based-on-emotions-extracted-from-facial-expression-and-speech-audio

Music Recommendation Filter based on Emotions Extracted from Facial Expression and Speech Audio

network-analysis-using-networkx

trying to understand a network by calculate the graph-level and node-level metrics.

online-shopper-purchase-behavior-prediction

The shift in consumer’s shopping behavior from retail shopping to e-commerce has resulted in tremendous growth of the e-commerce industry over the past few years. Many e-commerce companies analyze visitor sessions on their platforms, with the aim of increasing purchase conversion rates. Some of these companies are interested in predicting user behavior of each individual visit session on e-commerce websites. The task in the online shopper’s intention dataset is to predict whether a user is going to make a purchase based on data from a user’s current session.

predict-soybean-oil-price-with-xgboost-modeling

stack-overflow-text-classification-using-the-bert-model

For final model I used the BERT model. First, I loaded in the BERT pretrained model configuration as tokenizer. From there, we then encoded the train, validation, and test data. The max_length of the text was set to be 128 and I split the data into input_ids, attention_masks and labels. Second, I loaded in the BERT pretrained model and create data loaders which combine dataset and sampler together. RandomSampler is used for training and SequentialSampler for validation. The batch_size is set to be 1 due to the limited memory in my Google Colab. Third, I created a scheduler with an AdamW optimizer and iterations of 1. I tried to run for 2 to 3 epochs, but due to the limited memory on my laptop, only 1 epoch worked successfully after multiple attempts. Finally, I started with my BERT model training process with free Tesla K80 GPU on Google Colab and saved my model in my device, so I can load my pretrained model and evaluate the performance.

stacking-technique

study-of-network-resilience

Error and attack tolerance of complex networks

time-extraction

tweets-and-news-titles-sentiment-analysis-using-streamlit-app

The task involves bundling a sentiment analysis tool utilizing the FinBERT and RoBERTa models with Streamlit, along with incorporating a DALL-E API for producing images, allowing users to visually comprehend the meaning of the title in a more intuitive manner.

uberizing-airport-hotel-shuttles

With most of the country’s population vaccinated against COVID-19 and a similar movement taking place concurrently across the globe, Singapore has accelerated its shift towards a COVID-endemic strategy. Prompted by the need to jumpstart the economy, measures have been passed to allow the country and various sectors of the economy to re-open gradually. Part and parcel of these reforms is the relaxation of border control measures. As of October 2021, vaccinated travellers from eight (8) countries were allowed to enter the country without the need for self-quarantine. That list has since expanded to include thirteen (13) countries on top of comparatively looser restrictions for jurisdictions not yet part of Singapore’s Vaccinated Travel Lanes (VTL) program.

word-similarity-calculation-using-word2vec-and-gensim-

Word similarity calculation using word2vec and gensim

yahoo-financial-api-playaround

Scrape data from Yahoo Financial API