Abe full period — second administration

2012-12-26 → 2020-09-16 · merged 2012–2014 and 2014–2020 segments · method · about & citation

Abe Activity Graph は、土屋貴裕による研究プロジェクトです。本プロジェクトは、公開された首相動静をもとに、第2期安倍政権期における首相の可視的な接触、移動、会談、会食、外交日程を構造化し、時期別・組織別・カテゴリ別に分析するためのデータ基盤を構築します。現時点では、 組織・カテゴリ・月次推移の比較が最も信頼できる分析単位です。

Method & caveat cards

1. Principal model

安倍晋三 is the principal for all 59,157 events. The principal is the subject of the daily-schedule record and is not counted as a recorded contact.

2. Source layers

2012–2014 public archive segment covers 2012-12-26 → 2014-12-23. 2014–2020 structured activity-log segment covers 2014-12-24 → 2020-09-16. Both are normalized into a common derived schema, but source differences remain.

3. Comparability

現時点では、組織・カテゴリ・月次推移の比較が最も信頼できる分析単位です。人物別の通期比較は、初期セグメントの解決率が低いため限定的です。

4. Observation limit

本データは、首相動静に記録された公開上の接触を対象とする。そのため、官邸内部で日常的に行われる随時協議、電話連絡、非公開調整、秘書官・補佐官を介した事前調整など、公開記録に現れない情報は含まれない。

5. Raw text policy

Raw source text is not displayed. Public artifacts contain structured factual fields, source metadata, and SHA-256 hashes of the parsed segments for chain-of-custody. See /method.

Principal coverage

Every event in the merged corpus is, by definition, about the principal. Principal coverage equals total events; principal is excluded from the recorded-contacts ranking.

Principal

安倍晋三

Principal person_id

prs-abe-shinzo

Full-period events

59,157

2012–2014 segment events

10,373

2014–2020 segment events

48,784

Distinct dates

2,729

Coverage matches expected

true

Source coverage by year

How the 59,157 events split across the two segments, year by year.

Year	2012–2014 public archive	2014–2020 structured activity log	Total
`2012`	86	0	86
`2013`	5,480	0	5,480
`2014`	4,807	187	4,994
`2015`	0	8,659	8,659
`2016`	0	9,128	9,128
`2017`	0	8,301	8,301
`2018`	0	8,353	8,353
`2019`	0	8,639	8,639
`2020`	0	5,517	5,517

Comparability

現時点では、組織・カテゴリ・月次推移の比較が最も信頼できる分析単位です。人物別の通期比較は、初期セグメントの解決率が低いため限定的です。

At this stage, organization-level, category-level, and monthly trend comparisons are the most reliable. Cross-segment person-level comparisons require additional care. See /method §5 for the comparability writeup.

Trend comparison

Organization, category, monthly: reliable

Early-segment person-mention recall

11.7%

Late-segment integrity

Cross-segment ID consistency

Late-vs-early event-rate ratio

1.68×

Persons in both segments

Caveats

Source layers differ in extraction methodology — pm-watcher (newspaper paraphrase) for early, kanteilog (daily-schedule HTML) for late. Event granularity, person-mention coverage, and category vocabulary are not identical.
Principal coverage is complete in both layers; contact rankings exclude the principal by construction.
Early-layer person recall is materially lower than late. Use organization, category, and monthly-event trends for full-period comparison; person rankings are most reliable when split by layer.
Event count per month in late is ~1.68× early — this reflects kanteilog's finer event granularity, not increased real activity.
Raw newspaper / kanteilog text never appears in any full-period artifact; only structured derived fields + SHA-256 hashes for chain-of-custody.

Monthly event trend (merged)

Boundary at 2014-12-24. The visible jump at this date reflects the 2014–2020 segment’s finer event granularity, not a real change in PM activity.

2012-12 → 2020-09 · hover bars for per-month layer split.

Category counts (refined)

Category labels are mapped to a shared vocabulary across layers. The underlying extraction heuristics differ — treat counts as approximate, layer-aware aggregates.

Category	Full period	Early	Late
`unknown`	17,301	0	17,301
`movement`	12,326	3,508	8,818
`meeting`	10,252	59	10,193
`residence`	6,463	2,078	4,385
`diet`	4,933	229	4,704
`other`	3,429	3,429	0
`party`	1,859	252	1,607
`meal`	1,785	493	1,292
`diplomacy`	396	77	319
`official_work`	228	220	8
`press`	106	24	82
`meeting_like`	72	0	72

Organization ranking — full period (segment-aware)

Full = union of the 2012–2014 and 2014–2020 segments per organization. 2012–2014 / 2014–2020 columns let you spot when a count is dominated by one segment’s extraction methodology rather than real activity.

#	Organization	Category	Full presence	2012–2014	2014–2020	Active days	Meetings
1	首相官邸	`prime_minister_office`	1,670	1,670	0	399	0
2	自由民主党	`liberal_democratic_party`	1,001	657	344	633	97
3	国家安全保障局	`national_security_secretariat`	689	78	611	518	606
4	内閣官房	`cabinet_secretariat`	668	90	578	537	377
5	財務省	`ministry_of_finance`	662	175	487	539	271
6	外務省	`ministry_of_foreign_affairs`	553	291	262	439	119
7	防衛省	`ministry_of_defense`	277	125	152	249	41
8	内閣府	`cabinet_office`	118	118	0	107	0
9	文部科学省	`other`	85	5	80	85	14
10	公明党	`komeito`	63	63	0	59	0
11	外国政府	`foreign_government`	46	46	0	41	2
12	経団連	`business_federation`	39	29	10	36	0
13	経済産業省	`ministry_of_economy_trade_and_industry`	31	31	0	30	0
14	その他	`other`	16	16	0	12	0
15	日本商工会議所	`business_federation`	13	13	0	13	0

Organization monthly trend (full period, top 8)

Sparkline shows the merged monthly presence. Sharp transitions at 2014-12 may reflect the segment switch rather than a real change in activity.

Organization	Total	Trend
首相官邸	1,670
自由民主党	1,001
国家安全保障局	689
内閣官房	668
財務省	662
外務省	553
防衛省	277
内閣府	118

Resolved recorded contacts (segment-aware)

#	Person	Primary org	Full events	2012–2014	2014–2020	Active days	Meetings	Meals	Diplomacy	Diet
1	谷内正太郎	`org-cabinet-sec`	611	0	611	455	606	5	0	0
2	麻生太郎	`org-mof`	584	97	487	486	271	38	0	192
3	菅義偉	`org-cabinet-sec`	478	0	478	393	311	77	0	65
4	茂木敏充	`org-meti`	259	0	259	235	115	15	0	70
5	岸田文雄	`org-mofa`	202	6	196	185	98	26	0	16
6	稲田朋美	`org-cabinet-sec`	105	5	100	101	15	9	0	23
7	甘利明	`org-cabinet-sec`	104	0	104	94	67	9	0	9
8	谷垣禎一	`org-ldp`	102	0	102	90	0	6	0	25
9	下村博文	`org-mext`	80	0	80	80	14	3	0	9
10	二階俊博	`org-ldp`	61	4	57	60	0	25	0	3
11	小野寺五典	`org-mod`	55	2	53	49	26	2	0	21
12	高市早苗	`org-ldp`	41	41	0	38	0	8	0	3
13	細田博之	`org-ldp`	15	15	0	15	0	5	0	0
14	榊原定征	`org-keidanren`	7	0	7	7	0	2	0	0
15	秋葉剛男	`org-mofa`	3	0	3	3	3	0	0	0
16	中西宏明	`org-keidanren`	3	0	3	3	0	3	0	0
17	石破茂	`org-ldp`	2	2	0	2	0	0	0	0
18	江渡聡徳	`org-mod`	1	1	0	1	0	0	0	0
19	加藤勝信	`org-cabinet-sec`	1	1	0	1	0	0	1	0

Caveats

Segment caveat. The 2012–2014 and 2014–2020 segments use different extraction methodologies. Event-count comparisons across the 2014-12-24 boundary should be read alongside the comparability section above, not as evidence of behavior change.
Public-record scope. Internal coordination, off-record phone calls, and non-public meetings are out of scope by construction.
No verbatim text. Raw source text stays under data/raw/ (gitignored). Public artifacts carry only structured fields + SHA-256 hashes of the parsed segments.
Principal vs recorded contacts. Principal coverage is 1.0 by construction. Recorded contacts are persons OTHER than the principal; their counts are segment-dependent.