自然语言处理 - Natural Language Processing

课程介绍 Course Introduction

学分：3 | 先修课：机器学习导论、深度学习 | 学期：第6学期

自然语言处理研究计算机理解、生成与处理人类语言的理论与方法。课程涵盖文本预处理、词向量表示（Word2Vec、GloVe）、语言模型、循环与注意力网络、Transformer与预训练模型（BERT、GPT）、文本分类、机器翻译、问答系统与生成式任务。学生将掌握从规则方法到深度学习的演进，理解大语言模型原理，能应用于实际NLP工程。

Natural Language Processing studies theories and methods for computers to understand, generate, and process human language. Topics include text preprocessing, word embeddings (Word2Vec, GloVe), language models, recurrent and attention networks, Transformers and pre-trained models (BERT, GPT), text classification, machine translation, question answering, and generation. Students trace the evolution from rule-based to deep learning approaches and master large language model principles for real NLP engineering.

大作业 Final Project

作业标题：影评情感分类系统

学生需基于IMDb或豆瓣影评数据构建情感分类系统。要求实现基于Transformer的文本分类模型，对比传统方法（TF-IDF+SVM）与预训练模型微调（BERT）的效果，完成数据清洗、tokenization、训练与评估。需提交代码、模型对比报告及可视化混淆矩阵，F1分数作为核心指标。

Students build a sentiment classification system on IMDb or Douban movie reviews. The project requires implementing a Transformer-based text classifier, comparing traditional methods (TF-IDF + SVM) with fine-tuned pre-trained models (BERT), and completing data cleaning, tokenization, training, and evaluation. Deliverables include code, a model comparison report, and a confusion matrix visualization, with F1 score as the key metric.