Online shopping through e-commerce sites is becoming more prevalent with the expansion towards a more digital age in our society together with recent factors such that of the COVID-19 Pandemic. Through Machine Learning and the concept of Sentiment Analysis, algorithms would be able to identify the sentiment of reviewers by processing the words used in the sentence. The research aims to determine the reliability of star ratings compared to sentiment analysis and which classification algorithm suits best for text classification by Filipino customer reviews in Shopee for Medical Personal Protective Equipment or PPEs. It also aims to identify the best classifier model to use in terms of its performance. The study was divided into two models: star ratings and sentiment analysis. Both data sets performed different preprocessing techniques and tested for Naive Bayes and Support Vector Machine classification models, and their performance measures were obtained. The findings of the study show that star ratings and annotated reviews present high similarity in terms of the sentiment and polarity classified per review. In terms of the best performing model, Support Vector Machine achieved the best scores for the performance measures among the tests.
Coastal areas are subject to dynamic threats due to nature, e.g., wind, waves, coastal storms, sea-level rise, and human activity. The consequences may interfere with human activities such as aquaculture, tourism, and infrastructure. The shoreline along the Krachai subdistrict of Chantaburi Province in Thailand was monitored using satellite-based images from 2014-2022. The aim of this research is to apply CoastSat and Google Earth Engine to monitor shorelines. Using satellite imagery, we can observe and analyze coastlines over time with actual, reliable images in context. To save time, the methodology applied Python software programming based on Google Earth Engine to automatically search satellite datasets and study areas to avoid downloading a whole image. Accuracy was verified using visual interpretation and distance differences were calculated showing an average of 0.25 meters. The shoreline detected from CoastSat was clear and gave positive results, compensating for operational time, starting from satellite images received for shoreline detection. The CoastSat approach can be applied anywhere detectable by satellite. Slight changes cause difficulty showing these as printed material; therefore, a web-based solution was developed to allow users to select areas of interest.
Smart Home technology is becoming an integral tool for senior citizens. The Interface of smart home applications should be usable and compatible for senior citizens in order to provide an effective experience. For assessment of any application usability is the key attribute to measure its usefulness. Designing information and technology products for the home has drawn attention to the narrowness of traditional conceptions of usability; therefore evaluating usability is a vital task. Design methods and guidelines were developed in work contexts and so are mainly concerned with ease-of-learning, low level ease-of-use and task fit. With the rapid advancement of smart home technology, usability of smart home technology apps gains attention of modern researchers. The focus of this paper is to provide the usability evaluation framework to evaluate the UI of smart home technology (SHT) applications(apps). Existing interface design guidelines of SHT are reviewed in this paper. To ensure the effectiveness and reliability of the framework, it was validated by applying the proposed metrics and evaluation instruments in a usability study conducted on two SHT apps for senior citizens. Results gathered from usability testing proved that the framework is applicable for evaluation of SHT apps for senior citizens.
The information people receive from social media contains a variety of topics, and these massive amounts of information are often disorganized. It is difficult for people to get valuable information directly from it. The advantage of our approach is two-fold: to accommodate users of various interests, and Human-in-the-Loop can be responsible for guiding AI system learning. And machine learning is one of the standard methods that is widely used for data classification tasks. The development of machine learning processes often requires efforts of data labeling and programming, which is often time-consuming and labor-intensive. To help non-experts better understand the model training process, we design a no-code data classification platform. It uses messages crawled from social media as data sources, filters data, and incorporates active learning methods to reduce the cost of creating datasets. Our platform uses an automated process to help users complete model training through simple web operations without programming. The ultimate goal is that users can intervene in the process of model training, quickly perform classification tasks and obtain information according to their needs.
In view of the abnormal behavior of online shopping consumers, this paper integrates machine learning algorithms, takes online shopping consumers' purchase data as the research object, mines effective information, uses PCA algorithm to extract features by dimensionality reduction for high latitude and strong correlation in click data, and uses One Class SVM algorithm to solve the problem of imbalance between normal and abnormal transaction behaviors. A classification of consumption behavior based on XGBoost algorithm in the integration algorithm is proposed to improve the anti-noise ability as well as generalization ability of machine learning algorithm through the combination of algorithms. The model is verified by K-fold cross-validation algorithm, and the performance of abnormal behavior detection model of online shopping consumers is described by F1-score, which verifies the effectiveness of this method in abnormal behavior detection.
