realmaud commited on
Commit
fda7763
·
verified ·
1 Parent(s): 75972b9

Add leaders, number_one_songs and prices configs; update card

Browse files
README.md CHANGED
@@ -2,9 +2,9 @@
2
  license: cc-by-sa-4.0
3
  language:
4
  - en
5
- pretty_name: "slashyear: dated historical events with source revision ids"
6
  size_categories:
7
- - 10K<n<100K
8
  task_categories:
9
  - question-answering
10
  - text-retrieval
@@ -16,7 +16,10 @@ tags:
16
  - timeline
17
  - wikipedia
18
  - wikidata
19
- - knowledge-base
 
 
 
20
  - rag
21
  - citations
22
  - temporal-reasoning
@@ -24,28 +27,29 @@ tags:
24
  configs:
25
  - config_name: events
26
  data_files: data/events/*.parquet
 
27
  - config_name: years
28
  data_files: data/years/*.parquet
29
  - config_name: subjects
30
  data_files: data/subjects/*.parquet
31
  - config_name: subject_events
32
  data_files: data/subject_events/*.parquet
 
 
 
 
 
 
33
  ---
34
 
35
- # slashyear — the dated historical record, with the source revision on every row
36
 
37
- Every dated historical event we could extract from English Wikipedia's year, decade
38
- and century articles, spanning from roughly 3000 BCE to the present.
39
 
40
- **The point of this dataset is the last column.** Every row carries `source_revid`,
41
- the numeric id of the exact Wikipedia revision the sentence was quoted from. A
42
- Wikipedia article is a moving target — a quotation you take today may not be there in
43
- six months, and a reader who follows your link cannot see what you actually read. A
44
- revision id names one immutable version that will read the same way in ten years. So a
45
- model that answers from this data can hand the user a link that stays true, which is
46
- not something you get by quoting an encyclopedia article.
47
 
48
- Live API, full-text search and MCP server: **https://slashyear.com/data**
 
49
 
50
  ## Configs
51
 
@@ -54,54 +58,51 @@ Live API, full-text search and MCP server: **https://slashyear.com/data**
54
  | `events` | one dated event, quoted verbatim, with its source revision |
55
  | `years` | one year, with its lead summary |
56
  | `subjects` | one subject (person, place, institution), most with a Wikidata QID |
57
- | `subject_events` | the join: one (subject, event) pair, so "every dated line mentioning Rome" is a filter |
 
 
 
58
 
59
  ```python
60
  from datasets import load_dataset
61
 
62
- events = load_dataset("<repo-id>", "events", split="train")
63
  events.filter(lambda r: r["year"] == 1969)
64
 
65
- rome = load_dataset("<repo-id>", "subject_events", split="train")
66
- rome.filter(lambda r: r["slug"] == "rome")
67
  ```
68
 
69
  ## Fields
70
 
71
- `events`: `year` (astronomical numbering — `-43` is 44 BCE), `year_label`, `date`,
72
- `topic`, `text`, `source_title`, `source_revid`, `source_section`, `source_url`, `page`.
73
 
74
- `subjects`: `slug`, `label`, `qid`, `description`, `entries`, `years`, `first_year`,
75
- `last_year`, `wikipedia`, `wikidata`, `page`. The rows carrying a `qid` are joinable
76
- against any other database keyed on Wikidata, with no name matching.
77
 
78
- `subject_events`: `slug`, `label`, `qid`, `year`, `year_label`, `date`, `topic`, `text`,
79
- `source_revid`, `source_url`.
 
 
 
 
 
80
 
81
  ## How it was built
82
 
83
- Deterministic extraction from the wikitext of English Wikipedia's year, decade and
84
- century articles at a pinned revision, then mechanical cleaning. **No language model
85
- touches the wording.** A model does exactly one thing: choose which of twelve topic
86
- labels a sentence files under. That decision cannot make a row say anything untrue,
87
- because it never alters the row.
88
 
89
- The subject associations are not inferred either — a subject is attached to an event
90
- because a Wikipedia editor wrote a `[[wikilink]]` to it inside that dated line.
91
 
92
- ## Limitations, stated plainly
93
 
94
- This dataset inherits Wikipedia's errors. If a date is wrong there it is wrong here.
95
- What it does not inherit is Wikipedia's instability, and it adds no errors of its own,
96
- because no step in the pipeline is permitted to generate text. Coverage follows what
97
- Wikipedia's year articles happen to cover, which is uneven across periods and heavily
98
- weighted toward Europe and the last three centuries. Entries before roughly 1000 BCE
99
- are sparse and often approximate. English only.
100
 
101
- ## Licence and attribution
102
 
103
- CC BY-SA 4.0, inherited from Wikipedia. Free for any use including commercial use and
104
- including training. Every row carries the article title and revision id needed to
105
- attribute automatically:
106
 
107
  > {source_title}, English Wikipedia, revision {source_revid}. {source_url}
 
2
  license: cc-by-sa-4.0
3
  language:
4
  - en
5
+ pretty_name: "slashyear: dated history with the source revision on every row"
6
  size_categories:
7
+ - 100K<n<1M
8
  task_categories:
9
  - question-answering
10
  - text-retrieval
 
16
  - timeline
17
  - wikipedia
18
  - wikidata
19
+ - world-leaders
20
+ - prices
21
+ - inflation
22
+ - billboard
23
  - rag
24
  - citations
25
  - temporal-reasoning
 
27
  configs:
28
  - config_name: events
29
  data_files: data/events/*.parquet
30
+ default: true
31
  - config_name: years
32
  data_files: data/years/*.parquet
33
  - config_name: subjects
34
  data_files: data/subjects/*.parquet
35
  - config_name: subject_events
36
  data_files: data/subject_events/*.parquet
37
+ - config_name: leaders
38
+ data_files: data/leaders/*.parquet
39
+ - config_name: number_one_songs
40
+ data_files: data/number_one_songs/*.parquet
41
+ - config_name: prices
42
+ data_files: data/prices/*.parquet
43
  ---
44
 
45
+ # slashyear: dated history with the source revision on every row
46
 
47
+ Dated historical events from English Wikipedia's year, decade and century articles, from about 3000 BCE to 2025, plus world leaders by year, Billboard No. 1 singles and US prices.
 
48
 
49
+ Every row that comes from Wikipedia carries `source_revid`, the id of the exact revision it was quoted from. Articles change all the time, so a quote taken from the live page today might not be there next year. A revision id always shows the same text, so anyone can open the link and check the row.
 
 
 
 
 
 
50
 
51
+ Website, JSON API and MCP server: https://slashyear.com/data
52
+ Code: https://github.com/ctrl-maud/slashyear
53
 
54
  ## Configs
55
 
 
58
  | `events` | one dated event, quoted verbatim, with its source revision |
59
  | `years` | one year, with its lead summary |
60
  | `subjects` | one subject (person, place, institution), most with a Wikidata QID |
61
+ | `subject_events` | one (subject, event) pair, so "every dated line mentioning Rome" is a filter |
62
+ | `leaders` | one head of state or government holding office in a given year, 1800 to 2025 |
63
+ | `number_one_songs` | one week's Billboard Hot 100 No. 1, from August 1958 |
64
+ | `prices` | one yearly average of a US price series (CPI, gas, groceries, electricity, new house price, household income, minimum wage) |
65
 
66
  ```python
67
  from datasets import load_dataset
68
 
69
+ events = load_dataset("realmaud/slashyear", "events", split="train")
70
  events.filter(lambda r: r["year"] == 1969)
71
 
72
+ leaders = load_dataset("realmaud/slashyear", "leaders", split="train")
73
+ leaders.filter(lambda r: r["year"] == 1985 and r["country"] == "France")
74
  ```
75
 
76
  ## Fields
77
 
78
+ `events`: `year` (astronomical numbering, so `-43` is 44 BCE), `year_label`, `date`, `topic`, `text`, `source_title`, `source_revid`, `source_section`, `source_url`, `page`.
 
79
 
80
+ `subjects`: `slug`, `label`, `qid`, `description`, `entries`, `years`, `first_year`, `last_year`, `wikipedia`, `wikidata`, `page`. Rows with a `qid` join to anything else keyed on Wikidata.
 
 
81
 
82
+ `subject_events`: `slug`, `label`, `qid`, `year`, `year_label`, `date`, `topic`, `text`, `source_revid`, `source_url`.
83
+
84
+ `leaders`: `year`, `state`, `country`, `region`, `name`, `title`, `qid`, `quote`, `source_title`, `source_revid`, `source_url`.
85
+
86
+ `number_one_songs`: `chart_date`, `title`, `artist`, `source_title`, `source_revid`, `source_url`.
87
+
88
+ `prices`: `year`, `item`, `label`, `unit`, `value`, `series`, `publisher`, `source_url`. Values are the average of that year's monthly or quarterly figures, in dollars of the time.
89
 
90
  ## How it was built
91
 
92
+ The event text is extracted from the wikitext of each article at a pinned revision and cleaned mechanically. No language model writes or edits any of it. A model is used for one thing only: picking which of twelve topic labels a line goes under, which never changes the line itself.
93
+
94
+ Subjects are attached to an event only when a Wikipedia editor linked them inside that dated line.
95
+
96
+ Leaders are quoted from Wikipedia's "List of state leaders in ..." articles. A holder is kept for a year only when the person's own Wikidata record has the same post over the same term (within a year). For the US, UK, Canada, France and Germany the result is also checked against the official lists of officeholders, and the build fails on any disagreement. Lines that fail a check are left out, so coverage is thinner before 1900.
97
 
98
+ Prices are FRED copies of BLS, Census and Department of Labor series (public domain). No. 1 songs are quoted from Wikipedia's yearly Billboard lists, each with its revision.
 
99
 
100
+ ## Limitations
101
 
102
+ Anything wrong on Wikipedia is wrong here too, unless a check above caught it. Coverage follows what the year articles cover, which leans heavily toward Europe and the last three centuries. Entries before about 1000 BCE are sparse and often approximate. English only.
 
 
 
 
 
103
 
104
+ ## License and attribution
105
 
106
+ CC BY-SA 4.0, inherited from Wikipedia. Free for any use, including commercial use and training. The `prices` rows are US government data (public domain). Each Wikipedia row has what you need to attribute it:
 
 
107
 
108
  > {source_title}, English Wikipedia, revision {source_revid}. {source_url}
data/leaders/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b12d56f4d1413ba00fbface919893336049616aaa88a1fad028639a56b8039cd
3
+ size 502084
data/number_one_songs/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74683768dba36db0598c4d0f85c806fd5ae78b092ea0618121b440144f29aa78
3
+ size 41560
data/prices/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fafcd98f358bda74825babbaecbaaadf45d4b94fa7611a25ff1034336c11314
3
+ size 8050