【CE314】Computer Science NLP

Deadline: Please follow deadline on FASER

Build a text classifier on the IMDB sentiment classification dataset, you can use any classification method, but you must training your model on the first 40000 instances and testing your model on the last 10000 instances. The IMDB dataset will be uploaded on the moodle page for you to download.

Your code should include:

1: Read the file, incorporate the instances into the training set and testing set.

2: Pre-processing the text, you can choose whether you need stemming, removing stop words, removing non-alphabetical words. (Not all classification models need this step, it is OK if you think your model can perform better without this step, and you can give some justification in the report.)

3: Analysing the feature of the training set, report the linguistic features of the training dataset.

4: Build a text classification model, train your model on the training set and test your model on the test set.

5: Summarize the performance of your model (You can gain additional marks if you have some graph visualization).

6: (Optional) You can speculate how you can improve your works based on your proposed model.

After you build such a model and test on the test set, you should write a report (no longer than three pages in A4, with Arial 11 fonts) to summarize your work.

(You can use the existing algorithms on github or kaggle, but you must not directly copy and paste their code!

However, you are not allowed to use the Naïve Bayes algorithm and VADER classifier, which practiced in Lab 4)

Suggestion: some bonus points:

Have necessary comments on your code

Have proper reference on your report

Have graph visualization on your report

Investigate more evaluation methods, like not only show the P R F score, but also run multiple times and show the standard derivation on P R F (I am sure you can find more evaluation methods.)

Write your report like a mini-conference paper (you can learn from this paper:

  • Zichao Yang, Diyi Yang, Chris Dyer, Xiaodong He, Alex Smola, and Eduard Hovy. 2016. Hierarchical Attention Networks for Document Classification. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , pages 1480--1489, San Diego, California. Association for Computational Linguistics.
相关推荐
百度Geek说14 分钟前
从代码生成到需求交付:DarkFactory 的工程实践
人工智能
Baihai_IDP19 分钟前
使用场景不同,模型“推理速度慢”的原因也不同
人工智能·面试·llm
挖掘狂人20 分钟前
AI 三分钟写完的代码,你凭什么敢直接上线?
人工智能·程序员·ai编程
墨林陌35 分钟前
AI 热点日报(2026-11-20):苹果自研 AI 智能体曝光,Anthropic 上市首日大涨,智能体支付落地
人工智能
AAA修煤气灶刘哥39 分钟前
32 GB 内存跑起 1800 亿参数模型,代价是什么?
人工智能·后端·架构
玩美数据1 小时前
海外市场数据分析驱动下的产业趋势演变与竞争格局深度解读
大数据·人工智能
oil欧哟1 小时前
Oil Studio 的产品思路与日常使用方式:在 Mac 上录口播演示,并接入 AI Agent 精剪与自动录屏
人工智能·macos
意念回复2 小时前
从 RNN 到 GQA:自注意力与多头注意力机制全解析
人工智能·rnn·深度学习
Flynn Liang2 小时前
AI Agent 分层记忆体系实战:用 Redis 五类结构构建可淘汰、可追溯的长短期记忆
数据库·人工智能·redis
CyrusYiVPP2 小时前
新用户第二周还剩几个?pandas做留存率cohort分析,数据分析实战第6期
人工智能·python