Assessment Criteria

When grading your project report, we will assess it with respect to the grading rubric given below.

Grade requirements

The grading rubric consists of six criteria. For each of these criteria, we will assess whether your project report fulfills them to the level of E = passing, C = good, or A = excellent. The requirements for each level are given below and should be read as cumulative, i.e., to fulfill the requirements for a higher level (C or A) you normally also need to fulfill the criteria for the lower level(s).

Your final grade will be:

• F/Fx (732A81) or U (TDDE16) if there is at least one criterion that is not fulfilled to the level of E.
• E (732A81) or 3 (TDDE16) if all criteria are fulfilled to at least the level of E.
• C (732A81) or 4 (TDDE16) if you additionally meet one of the following thresholds:
  • at least 5× C
  • at least 3× C and 1× A
  • at least 1× C and 2× A
• A (732A81) or 5 (TDDE16) if you additionally meet one of the following thresholds:
  • at least 5× A
  • 4× A and 2× C

If you study under 732A81, you can also receive a grade of D or B if you fall slightly below the threshold for the respective higher grade.

What happens if I fail the project?

In this case, you can resubmit a new version of your project report to any of the other examination opportunities. This can be a revised version of the same project report, and there are no grade penalties for re-submissions.

Grading rubric

Problem & Scientific Context

Pass (E)	The project is about a relevant text mining problem, and all sources (e.g. data, methods, tools) are appropriately cited.
Good (C)	The report states a specific and answerable research question or hypothesis, and the problem is well-motivated with some discussion of related work.
Excellent (A)	The problem is contextualized with an insightful discussion of related work, including recent scientific publications.

Presentation

Pass (E)	The report conforms to the formal requirements, follows a logical structure, and is written clearly.
Good (C)	The writing is precise and concise, with only minor language issues. Figures and tables are chosen to support key points and are clearly styled.
Excellent (A)	The report uses academic language throughout. Important parts of the project are communicated clearly and pedagogically, e.g. by making use of figures or illustrations to convey something that is hard to express entirely in text.

Experimental Setup

Pass (E)	The project represents an appropriate amount of work for the given timeframe (ca. 80~hours), applying easy-to-use methods and datasets.
Good (C)	Either data or methods required some non-trivial effort (e.g., compiling or annotating a small dataset, or a method from outside the course that required understanding someone else’s code).
Excellent (A)	Data and/or methods required significant effort (e.g., annotating a larger dataset, reimplementing or adapting a method from scientific papers).

Soundness

Pass (E)	Data and methods are suitable for the problem and are applied correctly, and the report contains enough detail to verify this.
Good (C)	Relevant design choices (e.g., data splits, hyperparameters) are reported and justified, and the report contains enough detail for the work to be reproducible.
Excellent (A)	Relevant properties of the data and/or the task (e.g., class imbalance, domain differences) are carefully examined and reflected in the experimental setup.

Evaluation & Analysis

Pass (E)	Results are evaluated with suitable evaluation methods.
Good (C)	At least one informative analysis that goes beyond standard quantitative metrics (e.g., per-class results, qualitative examples or errors, comparison of results on different datasets or data splits).
Excellent (A)	Several complementary analyses, or one in-depth analysis of the results (e.g., systematic manual error analysis, ablation study, non-trivial feature analysis).

Interpretation & Conclusions

Pass (E)	Results are interpreted correctly, and conclusions are clearly stated.
Good (C)	Relevant choices for the evaluation (e.g., macro-F1 vs. micro-F1) are reported and justified, and each conclusion is clearly backed by a result.
Excellent (A)	Robustness of results is quantified (e.g., variance across seeds or folds, confidence interval, significance tests), and limitations and validity threats are clearly discussed.