Research and Projects
Mental Disorder Detection
This study relates to the detection and analysis of mental disorders in the realm of social media, in particular Twitter. We are interested in extracting linguistic, emotional and behavioral styles for the task of mental disorder classification. We leverage crowd-sourcing concepts and propose a novel data collection mechanism for obtaining reliable online patient datasets.
(EmoViz) - Emotion Analysis / Visualization for Intention Identification
This work focuses on utilizing emotional knowledge to further investigate user behavior, in particular user intention or user interest identification. Our understanding is that emotions have a tremendous role in how we perceive and interact with the world around us. To that end, we adopt emotion detection algorithms to understand how users are feeling about a product for a given period of time. Through such emotional analysis, we are able to identify trending products as shown in the demonstration provided below.
Multilingual Emotion Detection / Classification
Emotion classification algorithms deal with the problem of predicting the emotion portrayed by a microblog post. Our in-house algorithms are able to achieve remarkable performances, even when applied to different domains and languages. Currently, we support emotion classification for English, Spanish and French. In future work, we are seeking to extend to other complex languages such as Indonesian, Japanese and Chinese.
This field of study is related to the detection of trending events on social media platforms. Platforms such as Twitter contain many events happening around the world but it’s important to recommend the more relevant and important ones. Adopting graph-based algorithms, we are able to identify events and obtain significant results as compared to techniques that rely on popular algorithms such as LDA.
Depression Support System
By combining studies related to interest identification, mental disorders, event detection and emotion detection, we are able to design a complex but efficient depression support system. Such system acts as a recommendation platform for depressed social network users. By understanding user’s emotions, we can obtain their current interests and then suggest content through events or concept extraction. More to come…
These set of studies aim at building dynamic algorithms that are able to extract online users’ interests. Currently, we rely on graph-based contextual approaches for interest identification, but are looking into other areas related to behavior analysis such as emotion analysis. By combining these two kind of approaches, we are now able to build more reliable and efficient algorithms for interest identification from different platforms.
Reviews Helpfulness through Opinion Mining
This study relates to the use of state-of-the-art Opinion Mining techniques to deduce how helpful reviews can be. Our approach focuses on the emotional styles used by online users, particularily those users talking about different features of products and services. Once the emotional styles are obtained, we employ clustering and classification techniques to further understand the consumers’ position through the content found in their reviews. Our approaches also works well with reviews submitted to different domains (e.g., movies and books).
Secure URL Shortner
A url shortner is a service that basically renders a shorter url for the one being provided (usually long). In this project, we implemented a service in a secure way that allows users to auto-generate a short URL that is secure and shareable among users with proper permission. Additionally, authors and users with the proper permission will be able to access a dashboard and obtain insights into the traffic information pertaining to the URL provided. Our hope is to provide a safe service using state-of-the-art security tools. Code