Final Project

This is the final project for the Natural Language Processing course. In this project, you are free to pick any theme you choose, as long as it is related to the course and based on current scientific literature or industry practices. You can use any technique you prefer, and you can choose any problem you prefer to deal with.

Choosing a track for the project

You can pick any of these tracks:

  1. Builder: you want to build something. You want to see a system running, allowing actual people to use it. Your project is useful because it uses NLP to help people achieve a (possibly non-NLP) goal.

  2. Researcher: you want to discover something. You want to run an experiment that solves a novel, relevant scientific question. Your project is useful because it brings forward a particular result related to the development of NLP itself.

  3. Educator: you want to teach something. You want to show people a different perspective (maybe your own) on how to apply a particular technique or on how to approach a particular problem. Your project is useful because it allows a particular audience (or: a niche) to understand an NLP-related concept more in depth.

Defining your objective

Your project must have a clear objective related to a measurable metric (in the sense of: "I will be happy with this project when it reaches point X, as measured by getting Y score in the Z measure"). The goal, and the metric, must be drawn from previous work in the field (which you will research). Examples of metrics are (these are examples, do not use them!)

  • for builder: the system allows reducing the time for completing some task in 20%.
  • for researcher: the project highlights that system A takes 50% more computational effort to train than system B for the same accuracy.
  • for educator: the presented material brings a different perspective to learning how to interpret BERT embeddings.

A good reference for objectives is the related work, that is, other projects available online that share the same type of problem as you.

Deliverables

DISCLAIMER: We live in an age of AI. From May 2026, it has become clear that AI coding agents are being widely used in industry and by students. Also, AI agents are being increasingly used to write text. This creates a problem: code and text are, now, insufficient proxies for evaluation. The rubrics for this project have been updated to account for this new scenario.

The rubrics and deliverables aim to assess whether the student is able to:

  1. Investigate the current state-of-the-art to find a relevant project theme and goal,
  2. Defend the use of adequate methods to achieve the desired goal,
  3. Relate your choices to the current state-of-the-art
  4. Communicate your choices and findings using adequate language

Report

At the end of the project, each group must present a report formatted using the IEEE Conference Template. The report must have, at most, two (2) pages, including all figures, tables, and references. This formatting requirement is a hard requirement, meaning that reports with more than two pages or using a different template will be automatically disclassified.

The report does not need an abstract. It does require:

  1. An introduction explaining the context of your project and its relation to other work. Your introduction must defend the relevance of your problem and your method.
  2. Sections for methods, datasets, etc., which explain the rationale of your work. For builders, this could be market research and interviews with potential users; for researchers, these are descriptions of materials and methods; for educators, an explanation of the rationales underlying the learning materials and their design choices.
  3. At least one section for results: user evaluations, accuracies, a qualitative assessment of your learning material, and so on.
  4. A concluding section explaining the "so what" of your work: how did this solve the problem you found in the first place (or: did it not solve the problem?)? What are its limitations? What did you learn while doing this work?
  5. References

Material

The actual things that you produced: source code, link to app, videos (or other learning materials), and so on.

Appendix

If you have more results than you can fit into 2 pages, submit them as appendixes. Appendixes will not be graded and cannot be referred to in the main text. However, they can serve as additional reading.

Rubric

  • F (0%), if:
    • your project is off-topic
    • there are missing deliverables (that is, do NOT try to make half of the project)
    • the project was clearly copied from some online source (on top of the grade, this is an ethics violation)
    • your project fails to clearly state its goal and its relevance considering prior work, or goals were set without observing prior work in the field
    • the report formatting did not follow the requested format
  • D (25%) if your project was entirely delivered, but has clear flaws that derive from misunderstanding concepts in courseware (for example: you trained a neural network in a small dataset or with clear data leakage, your tutorial points at wrong directions, you cannot find arguments indicating that the system you built idea is feasible and useful), or it clearly does not achieve the stablished goals without any justification for such (that is, you ignore the goals when discussing the results).
  • C (50%), if your project is technically correct and achieve desired goals, but is somewhat redundant with existing work (for example: you made a tutorial for which there are many other tutorials online, or your project is a simple, well-knokwn project like a sentiment detector (that is, something that can be solved in a few lines of code), or your project consists of simply following a tutorial online).
  • B (75%) if your project is technically correct (including adequate statistical analysis, where necessary), achieves desired goals, and brings some uniqueness to the table - you approached a meaningful pain point, you made an experiment that actually highlights some interesting point, or you made a tutorial on a subject that was lacking.
  • A (100%) if your project is correct and unique, and the presented material simultaneously approaches more than one track (for example: you did a research on a topic which lead to building a useful system, or you built a system which accompanies learning material on NLP). To get this grade, you must include a research on existing art in your deliverable, showing that similar work has not addressed the specific point you have addressed, or that they were somehow insufficient and you covered that gap.

Interview

One or more members of the working group may be requested to defend their work in an in-person interview at the discretion of the instructor. The interview will be recorded. The group member will be asked to defend the rationale underlying design choices, and to explain (in a whiteboard) topics about the project. The interview will limit the project grade (for the whole group) as follows:

  • F: interviewee is unable to explain key points of the project, and can only discuss topics that were clearly seen in class.
  • D: interviewee mainly paraphrases material from the text when asked, and is unable to explain the rationale undelying design choices and chosen methods.
  • C: interviewee can only explain underlying rationales if helped by the interviewer.
  • A: interviewee can fully explain the methods used in the project and the rationale underlying them.

In this interview, the interviewee should expect questions like: "explain me your work in 30 seconds", "how does method X work?", "why is method X applicable in this work", "why did you choose method X and what other methods did you investigate?", "how does method X compare with Y?".