## Heatmap: Classification accuracies
### Overview
The image is a heatmap comparing classification accuracies across four methods (TTPD, LR, CCS, MM) for 12 categories. Values are presented as percentages with standard deviation (±) uncertainty. Colors range from purple (low accuracy) to yellow (high accuracy).
### Components/Axes
- **X-axis (Methods)**: TTPD, LR, CCS, MM
- **Y-axis (Categories)**:
1. cities_de
2. neg_cities_de
3. sp_en_trans_de
4. neg_sp_en_trans_de
5. inventors_de
6. neg_inventors_de
7. animal_class_de
8. neg_animal_class_de
9. element_symb_de
10. neg_element_symb_de
11. facts_de
12. neg_facts_de
- **Color Legend**: Vertical bar on the right (0.0 = purple, 1.0 = yellow)
### Detailed Analysis
#### TTPD Column
- cities_de: 92 ± 1 (yellow)
- neg_cities_de: 100 ± 0 (bright yellow)
- sp_en_trans_de: 93 ± 2 (yellow)
- neg_sp_en_trans_de: 96 ± 1 (bright yellow)
- inventors_de: 86 ± 1 (orange-yellow)
- neg_inventors_de: 77 ± 2 (orange)
- animal_class_de: 79 ± 1 (orange)
- neg_animal_class_de: 85 ± 1 (orange-yellow)
- element_symb_de: 64 ± 3 (orange-red)
- neg_element_symb_de: 82 ± 3 (orange-yellow)
- facts_de: 71 ± 2 (orange)
- neg_facts_de: 64 ± 3 (orange-red)
#### LR Column
- cities_de: 98 ± 2 (bright yellow)
- neg_cities_de: 99 ± 1 (bright yellow)
- sp_en_trans_de: 91 ± 4 (yellow)
- neg_sp_en_trans_de: 94 ± 3 (yellow)
- inventors_de: 87 ± 5 (orange-yellow)
- neg_inventors_de: 91 ± 6 (yellow)
- animal_class_de: 81 ± 4 (orange)
- neg_animal_class_de: 82 ± 2 (orange-yellow)
- element_symb_de: 86 ± 3 (orange-yellow)
- neg_element_symb_de: 75 ± 9 (orange)
- facts_de: 74 ± 5 (orange)
- neg_facts_de: 68 ± 5 (orange-red)
#### CCS Column
- cities_de: 80 ± 19 (orange)
- neg_cities_de: 84 ± 19 (orange-yellow)
- sp_en_trans_de: 73 ± 21 (orange)
- neg_sp_en_trans_de: 70 ± 20 (orange)
- inventors_de: 70 ± 24 (orange)
- neg_inventors_de: 73 ± 20 (orange)
- animal_class_de: 67 ± 15 (orange-red)
- neg_animal_class_de: 75 ± 16 (orange)
- element_symb_de: 63 ± 15 (orange-red)
- neg_element_symb_de: 56 ± 6 (orange-red)
- facts_de: 63 ± 9 (orange)
- neg_facts_de: 60 ± 8 (orange-red)
#### MM Column
- cities_de: 87 ± 3 (orange-yellow)
- neg_cities_de: 99 ± 2 (bright yellow)
- sp_en_trans_de: 96 ± 2 (bright yellow)
- neg_sp_en_trans_de: 80 ± 2 (orange)
- inventors_de: 85 ± 2 (orange-yellow)
- neg_inventors_de: 93 ± 1 (bright yellow)
- animal_class_de: 75 ± 2 (orange)
- neg_animal_class_de: 84 ± 1 (orange-yellow)
- element_symb_de: 54 ± 1 (orange-red)
- neg_element_symb_de: 66 ± 3 (orange)
- facts_de: 70 ± 2 (orange)
- neg_facts_de: 50 ± 4 (orange-red)
### Key Observations
1. **TTPD and LR dominate**: Both methods achieve >90% accuracy in 6/12 categories, with neg_cities_de reaching 100% in TTPD.
2. **CCS variability**: High standard deviations (e.g., 80 ±19 in cities_de) suggest inconsistent performance.
3. **MM underperformance**: Struggles in neg_facts_de (50 ±4) and element_symb_de (54 ±1), with lower overall accuracy than TTPD/LR.
4. **Negated categories**: Generally perform worse across all methods (e.g., neg_facts_de vs. facts_de).
### Interpretation
The data demonstrates that **TTPD and LR** are the most reliable classifiers, particularly for non-negated categories like cities_de and sp_en_trans_de. **CCS** shows high variability, possibly due to sensitivity to input noise or data distribution shifts. **MM** underperforms in specialized categories (e.g., element_symb_de), suggesting limitations in handling symbolic or negated data. The stark contrast between negated and non-negated categories (e.g., neg_cities_de at 100% vs. neg_facts_de at 50%) implies that negation introduces significant classification challenges across all methods. The heatmap highlights the need for method-specific optimizations for negated or symbolic data types.
