☘️1. Báo cáo viên: TS. Nguyễn Tiến Huy
- Thời gian tổ chức: 10h00 - 11h30, ngày 09/12/2023 (Thứ Bảy)
- Chủ đề:
a. Prompt Engineering for LLMs
Prompt engineering involves the strategic crafting and refinement of input queries to generative models, amplifying their precision, adaptability, and performance. Exploring the nuanced art of fine-tuning prompts, this abstract delves into how this technique empowers users with granular control, enabling customization for specific tasks and domains.
b. EnTube: Exploring Key Video Features for Advancing YouTube Engagement
The ever-growing popularity of video sharing on the internet, such as on YouTube, Twitter, and TikTok, has led to an increased emphasis on viewership ratings and video engagement. Via the engagement analysis, the research demonstrates promising potential in extracting powerful characteristics that enable informed decisions regarding engagement.
Các bạn sinh viên quan tâm có thể đăng ký qua link: https://forms.gle/GgQcvuAfu3TtJtoP7
Thời hạn đăng kí: Từ nay đến 15h00 Thứ Sáu, ngày 08/12/2023
☘️2. Báo cáo viên: TS. Nguyễn Ngọc Thảo
- Thời gian tổ chức: 14h00 - 15h30, ngày 11/12/2023 (Thứ Hai)
- Chủ đề: Graph Convolution Collaborative Filtering with Dense Embeddings
Recommender systems have been a vital part of many ecommerce and online services. These systems support profound user personalization by mining the semantic interactions between users and items in the database, thereby increasing customer satisfaction and revenue. Furthermore, collaborative filtering research has significantly advanced in recent years with the compelling expressiveness of graph-based neural networks. In this seminar, we introduce a novel collaborative filtering approach that represents users and items as graph embeddings and effectively exploits the knowledge from these embeddings. First, the feature vectors for users and items are refined with multiple embedding propagation layers. We then use many dense layers to get extra information as much as possible, main purpose is to support embedding vectors in the propagation process and allow our model to better learning of the useritem interaction, making a noticeable difference in performance from prior works. Finally, we combine them before mapping to a predicted score. The proposed method has been empirically proven superior to the baselines and competitive with modern approaches on public benchmarks.
Các bạn sinh viên quan tâm có thể đăng ký qua link: https://forms.gle/cis8fQdGW7ZBHsNh9
Thời hạn đăng kí: Từ nay đến 15h00 Chủ Nhật, ngày 10/12/2023
- Địa điểm tổ chức: Phòng I.81 - Cơ sở 227 Nguyễn Văn Cừ, P4, Q5
- Số lượng: 15 sinh viên / 1 buổi
Lưu ý: Link đăng ký sẽ đóng sớm hơn khi đã đủ số lượng người tham dự
Thông tin người phụ trách: Nguyễn Trần Thục Uyên (Email: nttuyen@fit.hcmus.edu.vn)
