UPLB ICS Peak One

An ensemble model composed of Decision Trees, AdaBoost and Gradient Boosting was created in order to tackle the issues of using an imbalanced dataset, and using non-English datasets. The ensemble model was trained on both English and Filipino Yelp datasets and was compared to other models. The results showed that the ensemble model achieved an f-1 score of 0.6822 and an AUC score of 0.7600 using the English Yelp dataset. In addition, it also achieved an f-1 score of 0.6859 and an AUC score of 0.7441. The findings suggest that the ensemble model can reliably classify fake reviews.

Facebook groups were used as an alternative platform for an online marketing tool to provide accessibility without lacking time and distance. The Facebook group named “ElBids” was not efficient to handle many bidding items at once and was not safe because of the non-existing authentication. The study was used to fully utilize the modern technology available to create an official ElBids web application that can provide better user experiences and user interfaces while ensuring safer transactions. By using the System Usability Scale (SUS), the web application got a score of 84.28 which exceeds a passing score of 68.00; hence, the web application is considered effective and usable.

Manual Contact Tracing is an inefficient way to mitigate and control the spread of COVID-19. The amount of labor to record and communicate with individuals regarding their COVID-19 status is tedious and is not sustainable in the long run, which creates a need for a digital contact tracing system. ElbiTrace is a mobile contact tracing application utilizing Bluetooth Low Energy, Wifi Technology, and Geolocation data. It was developed to assist and automate contact tracing for COVID-19. ElbiTrace includes a detection service for nearby devices which automates record creation for every close contact encounters and a virus heat map where users can identify areas with a high number of positive close contact records. The Technology Acceptance Model (TAM) and the System Usability Scale (SUS) were used to evaluate the application. Groups of users belonging to the same household were asked to test and answer 21 five-point scale statements. The statements are divided into three sections: 5 statements for perceived ease of use, 6 statements for perceived usefulness, and 10 statements for perceived usability. The result was computed by getting the average score of each section. Each section scored above 80 which indicates an excellent rating for the application.

FasterPaper is a text generation application developed by fine-tuning OpenAI GPT-2. The application aims to aid in paragraph construction to alleviate writer’s block for academic writing. The application fine-tunes a 355-million parameter GPT-2 model using aitextgen on an uploaded source text by the user. The application then accepts an input text from the user, which will be either summarized or used as a prompt to generate text. The application interface is based on the Flask framework and deployed on Amazon EC2. Testing on the application was be done by ten (10) undergraduate-level students which involved generating and summarizing text based on an academic paper. The idea that the program helps resolve writer’s block was generally well received by the respondents. When it comes to text generation from incomplete sentences and summaries, respondents provided higher scores. When it comes to text production from questions and directions, however, the scores were generally lower with more variation.

We report some benchmarking results of the Peak-Two Cloud(P2C) for High-Performance Computing(HPC) using the NAS Parallel Benchmarks(NPB).