News/Events

Online Seminar title: Aspect-based Fake News Detection

Date and time: 7 March 2024, Thursday 2pm in Melbourne time.

Presenter’s Name: Ziwei Hou

Abstract: The detection of misinformation as “fake news” is vital for a well-informed and highly functioning society. Most of the recent works on the identification of fake news make use of deep learning and large language models to achieve high levels of performance. However, traditional fake news detection methods may lack a nuanced “understanding” of content, including ignoring important information in the form of potential aspects in documents or relying on external knowledge sources to identify such aspects. This paper focuses on aspect-based fake news detection, which aims to uncover deceptive narratives through fine-grained analysis of news articles. We propose a novel aspect-based fake news detection method based on a lower, paragraph-level attention mechanism that identifies different aspects within a news-related document. The proposed approach utilizes aspects to provide concise yet meaningful representations of long news articles without reliance on any external reference knowledge. We investigate the impact of learning aspects from documents on the effectiveness of fake news detection. Our experiments on four benchmark datasets show statistically significant improvements over the results of several baseline models.

Bio: Ziwei currently is a final-year PhD student at Deakin University. Her research interests include text summary, aspect learning, etc.

Online Seminar title: Academic Performance Warning System based on Data Driven for Higher Education

Date and time: 8 Feb. 2024, Thursday 2pm in Melbourne time.

Presenter’s Name: Quoc Huy To

Abstract: Academic probation at universities has become a matter of pressing concern in recent years, as many students face severe consequences of academic probation. We carried out research to find solutions to decrease the situation mentioned above. Our research used the power of massive data sources from the education sector and the modernity of machine learning techniques to build an academic warning system. Our system is based on academic performance that directly reflects students’ academic probation status at the university. Through the research process, we provided a dataset that has been extracted and developed from raw data sources, including a wealth of information about students, subjects, and scores. We build a dataset with many features that are extremely useful in predicting students’ academic warning status via feature generation techniques and feature selection strategies. Remarkably, the dataset contributed is flexible and scalable because we provided detailed calculation formulas that its materials are found in any university or college in Vietnam. That allows any university to reuse or reconstruct another similar dataset based on their raw academic database. Moreover, we variously combined data, unbalanced data handling techniques, model selection techniques, and research to propose suitable machine learning algorithms to build the best possible warning system. As a result, a two-stage academic performance warning system for higher education was proposed, with the F2-score measure of more than 74% at the beginning of the semester using the algorithm Support Vector Machine and more than 92% before the final examination using the algorithm LightGBM.

Bio: Huy currently is a first-year PhD student at Deakin University. His research interests encompass data mining, text classification, and summarization, with a recent focus on text processing within the scientific domain.

Online Seminar title: Strawberry Ripening Stage Classification

Date and time: 11 Jan. 2024, Thursday 2pm in Melbourne time.

Presenter’s Name: Sandya De Alwiz

Abstract: Strawberries are a highly sought-after fruit due to their unique balance of sweetness and sourness. To ensure optimal growth, it is important to harvest the fruit at the correct ripeness stage and accurately estimate its acidity and Brix values. Previous studies have focused on these tasks separately, but we have discovered that effective estimation of acidity and Brix can also improve the identification of ripening stages in strawberries. Can acidity and Brix values be used in strawberry ripening stage classification? Can shadow removal enhance classification accuracy? This talk will demonstrate some research results in strawberry image classification, model explanations, and shadow removal of the images.

Bio: Sandya De Alwiz is a PhD student at School of IT, Deakin University

Online Seminar title: Emerging Scientific Topic Discovery by Finding Infrequent Synonymous Biterms

Date and time: 30 Nov. 2023, Thursday 2pm in Melbourne time.

Presenter’s Name: Junfeng Wu

Abstract: In this talk, we will explore the challenges and solutions in the automatic discovery of emerging scientific topics amidst the rapid growth of research papers. The ability to identify these topics is crucial for a variety of applications, such as resource allocation, technology trend prediction, knowledge gap identification, and personalized research direction recommendation. We will discuss two main obstacles – the scarcity of publications on emerging topics and the linguistic diversity in their descriptions. We will introduce our novel method, Infrequent Synonymous Biterms to discover Emerging Scientific Topics (isBEST), designed to tackle these challenges. Our approach involves reducing linguistic diversity through document-level clustering to identify linguistic variants of each key biterm, and unifying biterms in the same cluster expressing similar meanings to the most common synonymous biterm. To address the issue of rarity, isBEST converts each document into a vector of coefficients on synonymous biterms and clusters them at the corpus level with cosine similarity. We will delve into the logic behind assigning larger coefficients to rarer synonymous biterms, and the potential of a rarer synonymous biterm to denote an emerging topic from two collaborating sub-fields. We will conclude with a demonstration of the accuracy and effectiveness of our isBEST method through experiments on two large scholarly paper datasets.

Bio: Junfeng Wu was recently awarded a Ph.D. degree at Deakin University. His research is primarily focused on scientific text mining, time series analytics, big data analytics, and image and video processing. A significant portion of his work is dedicated to the analysis and prediction of emerging scientific topics. Currently, he is pursuing postdoctoral research at RMIT, where he is delving into the field of social media data mining.

Online Seminar title: Enhancing Document Relevance Classification and Question Generation

Date and time: 2 Nov. 2023, Thursday 2pm in Melbourne time.

Presenter’s Name: Dinesh Nagumothu

Abstract: Passage ranking is essential in Question Answering systems to select potentially answer-containing passages so that the answer-extracting model has a better chance of correctly answering the question. With recent advances in language models, passage re-ranking has become more effective due to improved natural language understanding of the relationship between questions and answer passages. While cross-encoder models achieve state-of-the-art re-ranking performance, can explicit linguistic features such as semantic triples assist in improving its performance further? This talk will introduce research results on the use of semantic triples for improved passage re-ranking by training cross-encoders with an elaborated cross-entropy loss function.

Bio: Dinesh Nagumothu is a final year PhD student at School of IT, Deakin University