Samurai007AK commited on
Commit
c550325
·
0 Parent(s):

ChapterEnd: human and generated textbook questions, corpus and scoring scripts

Browse files
Files changed (7) hide show
  1. .gitattributes +60 -0
  2. README.md +174 -0
  3. bm25_baseline.py +29 -0
  4. corpus.jsonl +0 -0
  5. generated.jsonl +0 -0
  6. human.jsonl +0 -0
  7. score.py +21 -0
.gitattributes ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.avro filter=lfs diff=lfs merge=lfs -text
4
+ *.bin filter=lfs diff=lfs merge=lfs -text
5
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
6
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
7
+ *.ftz filter=lfs diff=lfs merge=lfs -text
8
+ *.gz filter=lfs diff=lfs merge=lfs -text
9
+ *.h5 filter=lfs diff=lfs merge=lfs -text
10
+ *.joblib filter=lfs diff=lfs merge=lfs -text
11
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
12
+ *.lz4 filter=lfs diff=lfs merge=lfs -text
13
+ *.mds filter=lfs diff=lfs merge=lfs -text
14
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
15
+ *.model filter=lfs diff=lfs merge=lfs -text
16
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
17
+ *.npy filter=lfs diff=lfs merge=lfs -text
18
+ *.npz filter=lfs diff=lfs merge=lfs -text
19
+ *.onnx filter=lfs diff=lfs merge=lfs -text
20
+ *.ot filter=lfs diff=lfs merge=lfs -text
21
+ *.parquet filter=lfs diff=lfs merge=lfs -text
22
+ *.pb filter=lfs diff=lfs merge=lfs -text
23
+ *.pickle filter=lfs diff=lfs merge=lfs -text
24
+ *.pkl filter=lfs diff=lfs merge=lfs -text
25
+ *.pt filter=lfs diff=lfs merge=lfs -text
26
+ *.pth filter=lfs diff=lfs merge=lfs -text
27
+ *.rar filter=lfs diff=lfs merge=lfs -text
28
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
29
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
30
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
31
+ *.tar filter=lfs diff=lfs merge=lfs -text
32
+ *.tflite filter=lfs diff=lfs merge=lfs -text
33
+ *.tgz filter=lfs diff=lfs merge=lfs -text
34
+ *.wasm filter=lfs diff=lfs merge=lfs -text
35
+ *.xz filter=lfs diff=lfs merge=lfs -text
36
+ *.zip filter=lfs diff=lfs merge=lfs -text
37
+ *.zst filter=lfs diff=lfs merge=lfs -text
38
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
39
+ # Audio files - uncompressed
40
+ *.pcm filter=lfs diff=lfs merge=lfs -text
41
+ *.sam filter=lfs diff=lfs merge=lfs -text
42
+ *.raw filter=lfs diff=lfs merge=lfs -text
43
+ # Audio files - compressed
44
+ *.aac filter=lfs diff=lfs merge=lfs -text
45
+ *.flac filter=lfs diff=lfs merge=lfs -text
46
+ *.mp3 filter=lfs diff=lfs merge=lfs -text
47
+ *.ogg filter=lfs diff=lfs merge=lfs -text
48
+ *.wav filter=lfs diff=lfs merge=lfs -text
49
+ # Image files - uncompressed
50
+ *.bmp filter=lfs diff=lfs merge=lfs -text
51
+ *.gif filter=lfs diff=lfs merge=lfs -text
52
+ *.png filter=lfs diff=lfs merge=lfs -text
53
+ *.tiff filter=lfs diff=lfs merge=lfs -text
54
+ # Image files - compressed
55
+ *.jpg filter=lfs diff=lfs merge=lfs -text
56
+ *.jpeg filter=lfs diff=lfs merge=lfs -text
57
+ *.webp 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
README.md ADDED
@@ -0,0 +1,174 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-sa-4.0
3
+ language:
4
+ - en
5
+ task_categories:
6
+ - text-retrieval
7
+ - question-answering
8
+ pretty_name: ChapterEnd
9
+ size_categories:
10
+ - 1K<n<10K
11
+ tags:
12
+ - retrieval
13
+ - rag
14
+ - benchmark
15
+ - evaluation
16
+ - textbooks
17
+ - openstax
18
+ - synthetic-vs-human
19
+ configs:
20
+ - config_name: human
21
+ default: true
22
+ data_files:
23
+ - split: test
24
+ path: human.jsonl
25
+ - config_name: generated
26
+ data_files:
27
+ - split: test
28
+ path: generated.jsonl
29
+ - config_name: corpus
30
+ data_files:
31
+ - split: corpus
32
+ path: corpus.jsonl
33
+ ---
34
+
35
+ # ChapterEnd
36
+
37
+ Real end-of-chapter questions from six open textbooks, side by side with LLM-written questions about the same sections.
38
+
39
+ I built it to answer one question: do retrieval scores measured on generated questions hold up on questions that people wrote? In my tests they don't. The same BM25 gets 91.7% Recall@1 on the generated questions here and 54.6% on the human ones.
40
+
41
+ ## Why this exists
42
+
43
+ I was reproducing [STAIR](https://arxiv.org/abs/2609.03874), a paper from IBM Research that finds the right section of a book by reading its table of contents. Its benchmark, [SearchTome](https://huggingface.co/datasets/meghanadh/SearchTome), is made of questions written by an LLM, which is how most document benchmarks are built now. I wanted to see what happens with questions written by humans and couldn't find a test set for it.
44
+
45
+ OpenStax textbooks turned out to be a good source. Every chapter ends with review questions written by the book's authors, grouped by the section they belong to, so the correct answer comes with the question.
46
+
47
+ ## What the two kinds look like
48
+
49
+ Three sections, with one question of each kind:
50
+
51
+ | Section | Written by the textbook authors | Written by an LLM |
52
+ |---|---|---|
53
+ | Sleep and Why We Sleep | How could researchers determine if given areas of the brain are involved in the regulation of sleep? | What does comparative research indicate about the relationship between predatory risk and the amount of deep sleep in different species? |
54
+ | Overview of Photosynthesis | What two products result from photosynthesis? | What process provides the majority of energy for living organisms on Earth? |
55
+ | How To Organize Economies | What are the three ways that societies can organize themselves economically? | How does the scarcity of resources affect the media's decision to send reporters to cover an event? |
56
+
57
+ The LLM was shown one paragraph at a time, so its questions follow that paragraph closely. The authors ask about the section as a whole, often in words the text doesn't use.
58
+
59
+ ## The numbers
60
+
61
+ | | Human | Generated |
62
+ |---|---|---|
63
+ | Questions | 3,568 | 2,604 |
64
+ | Sections they point to | 492 (shared) | 492 (shared) |
65
+ | BM25 Recall@1 | 54.6% | 91.7% |
66
+ | BM25 Recall@5 | 82.7% | 98.6% |
67
+
68
+ ## Load it
69
+
70
+ ```python
71
+ from datasets import load_dataset
72
+
73
+ human = load_dataset("Samurai007AK/ChapterEnd", "human", split="test")
74
+ generated = load_dataset("Samurai007AK/ChapterEnd", "generated", split="test")
75
+ corpus = load_dataset("Samurai007AK/ChapterEnd", "corpus", split="corpus")
76
+ ```
77
+
78
+ ## The task
79
+
80
+ You get a question and the book it comes from. Rank that book's sections so the one that answers the question is first. A section is an OpenStax lesson module, for example "2.1 Describe the Income Statement, Statement of Owner's Equity, Balance Sheet, and Statement of Cash Flows". Books have between 32 and 142 sections, so guessing at random gets 1 to 3% Recall@1.
81
+
82
+ The books are different from SearchTome's, so scores here can't be compared with scores there.
83
+
84
+ ## What's inside
85
+
86
+ | File | Rows | |
87
+ |---|---|---|
88
+ | `corpus.jsonl` | 492 | one section per row |
89
+ | `human.jsonl` | 3,568 | questions written by the textbook authors |
90
+ | `generated.jsonl` | 2,604 | questions written by Mistral-7B-Instruct-v0.2 |
91
+ | `score.py` | | Recall@1 and Recall@5 for a run file |
92
+ | `bm25_baseline.py` | | the BM25 baseline, also a template for a run file |
93
+
94
+ **corpus.jsonl**: `module_id` (the OpenStax page id), `book`, `path` (chapter and section title) and `paragraphs` (the section text as a list).
95
+
96
+ **human.jsonl**: `qid`, `book`, `question` (as printed), `question_no_lo` and `lo_tagged` (explained below), `page_type` (which end-of-chapter page it came from, such as "Review Questions") and `gold` (the `module_id` of the correct section).
97
+
98
+ **generated.jsonl**: `qid`, `book`, `question`, `gold`, and `gold_paragraph` (which paragraph of that section the question was written from).
99
+
100
+ | Book | Sections | Human | Generated |
101
+ |---|---|---|---|
102
+ | Principles of Economics 3e | 142 | 1,060 | 356 |
103
+ | Psychology 2e | 88 | 571 | 404 |
104
+ | Principles of Accounting, Volume 1 | 84 | 777 | 417 |
105
+ | Concepts of Biology | 82 | 351 | 517 |
106
+ | Principles of Accounting, Volume 2 | 64 | 665 | 440 |
107
+ | Business Law I Essentials 2e | 32 | 144 | 470 |
108
+
109
+ ## Scoring your own method
110
+
111
+ Write a JSONL file with one line per question, best section first. Only rank sections from the question's own book.
112
+
113
+ ```
114
+ {"qid": "h0000", "ranking": ["046014ef-296a-4ef2-ae9d-8e58aec02bf3", "..."]}
115
+ ```
116
+
117
+ ```
118
+ python score.py human.jsonl my_run.jsonl
119
+ python score.py generated.jsonl my_run.jsonl
120
+ ```
121
+
122
+ To reproduce the BM25 numbers, run `pip install rank_bm25 numpy` and then `python bm25_baseline.py`.
123
+
124
+ ## Two things that fooled me
125
+
126
+ My first results on the human questions were wrong twice, for reasons worth knowing if you build something similar.
127
+
128
+ The first time, every method scored far lower than made sense. The end-of-chapter pages themselves ("Review Questions", "Key Terms", "Summary") were in my corpus, and they contain the questions word for word, so every retriever found the page that quotes the question. BM25 did this for 442 of 450 questions I sampled. The corpus here contains lesson sections only.
129
+
130
+ The second time, one method looked suspiciously good on accounting. 1,441 accounting questions start with a learning-objective tag like "LO 2.1", and that number is the number of the correct section 99.3% of the time. A model that reads the table of contents can match the digits and skip the question. `question_no_lo` has the tag removed and `lo_tagged` marks those questions, and `score.py` reports tagged and untagged questions separately.
131
+
132
+ ## Baselines
133
+
134
+ Recall@1 on the human questions. BM25 and the dense retriever score paragraphs, and a section gets the score of its best paragraph. The re-rankers re-order the retrieved candidates.
135
+
136
+ | Method | All | LO-tagged | Untagged |
137
+ |---|---|---|---|
138
+ | Dense retrieval, bge-base-en-v1.5 | 62.2 | 52.2 | 69.0 |
139
+ | bge-reranker-v2-m3, zero-shot | 58.2 | 44.9 | 67.2 |
140
+ | BM25 | 54.6 | 45.6 | 60.7 |
141
+ | bge-reranker-base, zero-shot | 53.0 | 40.6 | 61.4 |
142
+ | bge-reranker-base, fine-tuned on generated questions | 44.3 | 25.8 | 56.8 |
143
+
144
+ The last row surprised me most. Fine-tuning the re-ranker on generated questions improved it on generated questions and made it worse on human ones, from 53.0 to 44.3.
145
+
146
+ ## How it was built
147
+
148
+ Text and human questions come from the OpenStax archive API. On each end-of-chapter page the questions are grouped by lesson section, in order, so a question's correct section is the matching section of its chapter. Where the number of groups didn't match the number of sections I skipped the page (149 of 450). I dropped questions shorter than 25 characters and exact duplicates within a book.
149
+
150
+ For the generated set, Mistral-7B-Instruct-v0.2 wrote three questions per paragraph for a random sample of paragraphs from these books. The correct section is the one containing the paragraph. This file is the held-out test part of that set.
151
+
152
+ ## What it can't tell you
153
+
154
+ - It covers six books in English. I tried twelve more OpenStax books and left them out because their end-of-chapter pages didn't line up with their sections.
155
+ - The correct section is the one the book files the question under. Sometimes another section answers it too, so Recall@1 is a little harsh on every method.
156
+ - Some human questions aren't search queries. A few point at a figure or table and some ask for an opinion. I kept them as printed.
157
+ - The generated questions come from one model and one prompt. Another question writer may give a smaller or larger gap.
158
+
159
+ ## License and credit
160
+
161
+ All text and questions come from [OpenStax](https://openstax.org) textbooks (Rice University), licensed [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/). This dataset carries the same license: credit OpenStax, no commercial use, share changes under the same terms. The two scripts are MIT. OpenStax did not create or endorse this dataset.
162
+
163
+ Books: [Principles of Accounting, Volume 1](https://openstax.org/details/books/principles-financial-accounting) · [Volume 2](https://openstax.org/details/books/principles-managerial-accounting) · [Business Law I Essentials 2e](https://openstax.org/details/books/business-law-i-essentials-2e) · [Concepts of Biology](https://openstax.org/details/books/concepts-biology) · [Principles of Economics 3e](https://openstax.org/details/books/principles-economics-3e) · [Psychology 2e](https://openstax.org/details/books/psychology-2e)
164
+
165
+ ## Citation
166
+
167
+ ```bibtex
168
+ @misc{konar2026chapterend,
169
+ author = {Arijit Konar},
170
+ title = {ChapterEnd: human-written textbook questions for section retrieval},
171
+ year = {2026},
172
+ url = {https://huggingface.co/datasets/Samurai007AK/ChapterEnd}
173
+ }
174
+ ```
bm25_baseline.py ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # BM25 over paragraphs, a module scores its best paragraph. Writes run files for every question set.
2
+ # pip install rank_bm25 && python bm25_baseline.py
3
+ import json, re
4
+ from collections import defaultdict
5
+ import numpy as np
6
+ from rank_bm25 import BM25Okapi
7
+ from score import load, score
8
+ tok = lambda s: re.findall(r"[a-z0-9]+", s.lower())
9
+ by_book = defaultdict(lambda: {"ids": [], "owner": [], "text": []})
10
+ for m in load("corpus.jsonl"):
11
+ B = by_book[m["book"]]; B["ids"].append(m["module_id"])
12
+ for p in m["paragraphs"]:
13
+ B["owner"].append(len(B["ids"]) - 1); B["text"].append(p)
14
+ for B in by_book.values():
15
+ B["bm25"], B["owner"] = BM25Okapi([tok(t) for t in B["text"]]), np.array(B["owner"])
16
+ def run(qs, field):
17
+ out = []
18
+ for q in qs:
19
+ B = by_book[q["book"]]; s = B["bm25"].get_scores(tok(q[field]))
20
+ best = np.full(len(B["ids"]), -1e9); np.maximum.at(best, B["owner"], s)
21
+ out.append({"qid": q["qid"], "ranking": [B["ids"][i] for i in np.argsort(-best)[:10]]})
22
+ return out
23
+ human, gen = load("human.jsonl"), load("generated.jsonl")
24
+ for name, qs, field in [("human", human, "question"), ("human, LO tags stripped", human, "question_no_lo"),
25
+ ("generated", gen, "question")]:
26
+ r = run(qs, field)
27
+ with open(f"run_bm25_{name.split(',')[0]}{'_nolo' if 'LO' in name else ''}.jsonl", "w") as f:
28
+ f.writelines(json.dumps(x) + "\n" for x in r)
29
+ print(name, json.dumps(score(qs, r)))
corpus.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
generated.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
human.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
score.py ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Score a run file against human.jsonl or generated.jsonl.
2
+ # A run file is JSONL, one line per question: {"qid": "h0000", "ranking": ["<module_id>", ...]} (best first).
3
+ # Rank only modules from the question's own book (corpus.jsonl). Questions missing from the run count as wrong.
4
+ # python score.py human.jsonl run.jsonl
5
+ import json, sys
6
+ def load(p):
7
+ return [json.loads(l) for l in open(p, encoding="utf-8")]
8
+ def score(qs, runs):
9
+ rank = {r["qid"]: r["ranking"] for r in runs}
10
+ hit = lambda q, k: q["gold"] in rank.get(q["qid"], [])[:k]
11
+ out = {}
12
+ groups = {"all": qs}
13
+ if "lo_tagged" in qs[0]:
14
+ groups |= {"LO-tagged": [q for q in qs if q["lo_tagged"]], "untagged": [q for q in qs if not q["lo_tagged"]]}
15
+ for name, g in groups.items():
16
+ out[name] = {"n": len(g), "R@1": round(100 * sum(hit(q, 1) for q in g) / len(g), 1),
17
+ "R@5": round(100 * sum(hit(q, 5) for q in g) / len(g), 1)}
18
+ return out
19
+ if __name__ == "__main__":
20
+ for k, v in score(load(sys.argv[1]), load(sys.argv[2])).items():
21
+ print(f"{k:10} n={v['n']:5} R@1 {v['R@1']:5.1f} R@5 {v['R@5']:5.1f}")