/NTU_ADL_2023_Fall

Primary LanguagePython

NTU_ADL_2023_Fall

  • This repository contains the homework assignments and final project implementations in NTU 2023 Fall Applied Deep Learning
  • The corresponding README file of each assignment is placed under the assignment folder

Final Project (Team)

Books Wrapped - Explore, understand, and grow with your reading tastes

  • Abstract 
    In an era where music streaming services like Spotify and YouTube Music offer personalized insights into users’ musical preferences, we
introduce ’Books Wrapped,’ an innovative application that extends this concept to reading. ’Books Wrapped’ empowers users to gain a
deeper understanding of their personalities through their reading choices. Our system comprises three integral components: the Abstract
Web Scraper, which collects book abstracts from the internet; the Key Phrase Extractor, which identifies genres and significant words from
these abstracts; and the Personality Trait Generator, which derives insights about users’ personalities based on these key phrases. Users
simply input their reading catalog to receive a visually engaging tag cloud of keywords and an analysis of their personality traits, offering
a unique perspective on how their reading choices reflect and shape their identity.
  • System Diagram system_diagram
  • Link to the project page: https://github.com/TzuMinYang-NTU-lecture-homework/1121_Applied-Deep-Learning/tree/main/final%20project
  • Details: https://github.com/Mike-YANG-11/NTU_ADL_2023_Fall/blob/main/final/adl_final.pdf

HW1

Chinese Extractive Question Answering (QA)

  • Fine-tune bert-base-chinese model
    1. Paragraph Selection: Determine which paragraph is relevant
      • Implemented in Multiple Choice framework
    2. Span selection: Determine the start and end position of the answer span
      • Implemented in Question Answering framework

HW2

Chinese News Summarization (Title Generation)

  • Fine-tune pre-trained small-multilingual-T5 model
    1. Input: news content
    2. Output: news title

HW3

Instruction Tuning (Classical Chinese)

  • Fine-tune Taiwan-LLaMa
  • Utilize QLoRA technique