amozo commited on
Commit
eeddd27
·
verified ·
1 Parent(s): 0ef025e

Upload folder using huggingface_hub

Browse files
Files changed (5) hide show
  1. .gitattributes +2 -0
  2. README.md +100 -1
  3. characters.csv +3 -0
  4. slop_fingerprints.json +0 -0
  5. stories.csv +3 -0
.gitattributes CHANGED
@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ characters.csv filter=lfs diff=lfs merge=lfs -text
62
+ stories.csv filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,102 @@
1
  ---
2
- license: mit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: cc-by-4.0
3
+ task_categories:
4
+ - text-generation
5
+ language:
6
+ - en
7
+ - es
8
+ - fr
9
+ - de
10
+ size_categories:
11
+ - 10K<n<100K
12
+ tags:
13
+ - llm-generated-text
14
+ - creative-writing
15
+ - diversity
16
+ - mode-collapse
17
+ pretty_name: Character-Name Concentration in LLM Fiction (44,782 stories, 20 models)
18
  ---
19
+
20
+ # Dataset: LLM-generated stories and character extractions
21
+
22
+ 44,782 short stories generated by 20 LLMs under 8 mitigation strategies
23
+ (plus supplementary runs), with 88,919 extracted characters. Stories were
24
+ generated between 2026-05 and 2026-07 via OpenRouter and a local Ollama
25
+ endpoint, from a fixed stimulus set of five pictograms per story and a
26
+ four-language split (English 40%, Spanish, French, German 20% each).
27
+
28
+ ## Files
29
+
30
+ | File | Rows | Description |
31
+ |---|---|---|
32
+ | `stories.csv` | 44,782 | One row per generated story |
33
+ | `characters.csv` | 88,919 | One row per extracted character |
34
+ | `slop_fingerprints.json` | 20 models | Per-model baseline name fingerprints |
35
+
36
+ Join: `characters.story_id -> stories.story_id`. Stories with no extracted
37
+ character have no rows in `characters.csv`.
38
+
39
+ ## `stories.csv` columns
40
+
41
+ | Column | Description |
42
+ |---|---|
43
+ | `story_id` | UUID |
44
+ | `model` | Model identity, including modality and reasoning suffixes: `(chat)`, `(vision)`, `Thinking` |
45
+ | `strategy` | Prompt/generation strategy: `baseline`, `avoid_slop_1`, `ban_top_20_names`, `multiphase`, `multiagent`, `speculative_l00`, `speculative_l10`, `speculative_hard`, `without_stimulus`, `without_stimulus_antislop`, plus supplementary variants (`speculative_l05`, `speculative_l20`, `*_min_p_*`) |
46
+ | `language` | Story language: `en`, `es`, `fr`, `de` |
47
+ | `generation_type` | Internal generation mode label |
48
+ | `stimulus_tags` | The five stimulus concepts shown to the model (empty for no-stimulus runs) |
49
+ | `narrative_voice`, `tense` | Extracted narrative features |
50
+ | `input_tokens`, `output_tokens`, `execution_time_s` | Generation cost |
51
+ | `created_at` | Generation timestamp |
52
+ | `text` | Final story text (names substituted) |
53
+
54
+ ## `characters.csv` columns
55
+
56
+ | Column | Description |
57
+ |---|---|
58
+ | `story_id` | UUID, joins to `stories.csv` |
59
+ | `character_name` | Extracted name (may be empty for unnamed characters) |
60
+ | `name_type` | `PROPER_NAME`, `NO_NAME` (unnamed/role-based), or `STIMULUS_AS_NAME` (a stimulus concept used as a name). Slop metrics use `PROPER_NAME` only |
61
+ | `character_title`, `origin_word`, `gender`, `race_species`, `role`, `propp_function`, `importance`, `alignment`, `traits` | Character attributes from the extraction |
62
+ | `from_selection` | True if the name came from the candidate-selection stage |
63
+
64
+ ## `slop_fingerprints.json`
65
+
66
+ For each model, the name distribution of its baseline cell:
67
+ `total_chars`, `unique_names`, `freqs` (normalized name -> frequency), and
68
+ precomputed known-slop sets (`pct_2`: names with frequency >= 2%, `top_20`,
69
+ `knee`). These fingerprints drive the known/emergent decomposition of the
70
+ Slop Ratio in fingerprint mode.
71
+
72
+ ## Hyperparameters
73
+
74
+ K = 5 candidate name-sets per story (verbalized sampling); speculative
75
+ penalty lambda in {0, 10, HARD}; multiagent threshold of three uses per
76
+ name per model; banlist of 20 names.
77
+
78
+ ## License and provenance
79
+
80
+ All story text is machine-generated; no human-authored text is included.
81
+ Released under CC BY 4.0. Model outputs were collected through the OpenRouter
82
+ API and local Ollama endpoints; generation is not deterministically
83
+ reproducible for the online models, so this dataset is the definitive
84
+ artifact of the study.
85
+
86
+ ## Intended use
87
+
88
+ Reproducing the paper's metrics (Slop Ratio with known/emergent decomposition,
89
+ naturalness attestation), studying name concentration in LLM fiction, and
90
+ evaluating diversity-mitigation methods. Analysis code:
91
+ `GITHUB_URL_PLACEHOLDER`.
92
+
93
+ ## Citation
94
+
95
+ ```bibtex
96
+ @article{mozoqueesada2026charactername,
97
+ title={Character-Name Concentration in LLM Fiction: Measuring Displacement and Mitigating Repetition},
98
+ author={Mozo Quesada, Alejandro and G{\'a}lvez Rojas, Sergio and L{\'o}pez Valverde, Francisco},
99
+ journal={arXiv preprint},
100
+ year={2026}
101
+ }
102
+ ```
characters.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:444ab05a6bcc53c3d445e27004bb45778540845643e1f9ee9158acf7080f9a1d
3
+ size 14366513
slop_fingerprints.json ADDED
The diff for this file is too large to render. See raw diff
 
stories.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:32183ed020fb1d8ca3ee1865ed19331e48efebf7701d8e65adb63422d8be72ec
3
+ size 47824257