|
| 1 | +# Entity and Frequency Extraction Analysis |
| 2 | + |
| 3 | +**Date:** January 27, 2026 |
| 4 | +**Analyzed:** 5 SG Reports + 8 Resolutions (expanded) |
| 5 | + |
| 6 | +## Executive Summary |
| 7 | + |
| 8 | +This analysis examined UN SG reports and their mandating resolutions to determine where information about the **authoring entity** and **reporting frequency** can be reliably extracted, with a focus on **LLM-based extraction** rather than regex patterns. |
| 9 | + |
| 10 | +### Key Findings |
| 11 | + |
| 12 | +| Information Type | Best Source | Extraction Method | Reliability | |
| 13 | +|------------------|-------------|-------------------|-------------| |
| 14 | +| **Authoring Entity** | MARC 710__a + Manual list | Direct lookup | High | |
| 15 | +| **Report Frequency** | Resolution operative paragraphs | LLM extraction | Medium-High | |
| 16 | +| **Report Scope/Content** | Resolution operative paragraphs | LLM extraction | High | |
| 17 | +| **Previous Resolutions** | Resolution preamble | LLM extraction | High | |
| 18 | +| **Mandate Duration** | Resolution operative paragraphs | LLM extraction | High | |
| 19 | +| **Target Session/Date** | Resolution operative paragraphs | LLM extraction | High | |
| 20 | + |
| 21 | +--- |
| 22 | + |
| 23 | +## 1. Resolution Structure Analysis |
| 24 | + |
| 25 | +### Consistent Document Structure |
| 26 | + |
| 27 | +All resolutions follow a predictable structure that LLMs can reliably parse: |
| 28 | + |
| 29 | +``` |
| 30 | +[Header: Symbol, Title, Adopting Body, Date] |
| 31 | +
|
| 32 | +The [Body], |
| 33 | +
|
| 34 | +[PREAMBLE - "Recalling", "Recognizing", "Noting", etc.] |
| 35 | + - References to previous resolutions |
| 36 | + - Context and background |
| 37 | + - Concerns and observations |
| 38 | +
|
| 39 | +[OPERATIVE PARAGRAPHS - numbered] |
| 40 | + 1. Decides/Takes note/Welcomes... |
| 41 | + 2. Requests the Secretary-General to... |
| 42 | + 3. Invites Member States to... |
| 43 | + ... |
| 44 | +``` |
| 45 | + |
| 46 | +### Different Body Patterns |
| 47 | + |
| 48 | +| Body | Session Reference | Date Format | Example | |
| 49 | +|------|-------------------|-------------|---------| |
| 50 | +| **General Assembly** | "at its [Nth] session" | Session numbers (78th, 79th) | "at its eightieth session" | |
| 51 | +| **Security Council** | Specific dates | "until [date]" | "until 31 October 2025" | |
| 52 | +| **ECOSOC** | Session years | "[year] session" | "2020 session" | |
| 53 | +| **Human Rights Council** | Session numbers | "its [Nth] session" | "its thirty-first session" | |
| 54 | + |
| 55 | +--- |
| 56 | + |
| 57 | +## 2. Extractable Data Fields (LLM Targets) |
| 58 | + |
| 59 | +### 2.1 Frequency Information |
| 60 | + |
| 61 | +**Finding:** Frequency is often **implicit** rather than explicit. LLMs must infer from context. |
| 62 | + |
| 63 | +#### Explicit Frequency Patterns |
| 64 | +Some resolutions contain explicit frequency terms: |
| 65 | +- "annual report" / "annually" |
| 66 | +- "biennial" / "every two years" |
| 67 | +- "on a regular basis" |
| 68 | + |
| 69 | +**Example from A/RES/75/233 (QCPR):** |
| 70 | +> "Reiterates its request to present **annual reports** to the Economic and Social Council" |
| 71 | +
|
| 72 | +#### Implicit Frequency (Target Session) |
| 73 | +More commonly, frequency must be inferred from target session: |
| 74 | +- "submit to the General Assembly at its **eightieth session**" (from 78th → 80th = biennial) |
| 75 | +- "report to the Human Rights Council, starting from its **thirty-first session**" |
| 76 | + |
| 77 | +**Example from A/RES/78/70:** |
| 78 | +> "Requests the Secretary-General to submit to the General Assembly at its **eightieth session** a report on the implementation" |
| 79 | +
|
| 80 | +This implies biennial reporting (78th session resolution → 80th session report). |
| 81 | + |
| 82 | +#### Mandate Extension Patterns (Security Council) |
| 83 | +SC resolutions often extend mandates by specific periods: |
| 84 | +> "Decides to extend the mandate of the Verification Mission until **31 October 2025**" |
| 85 | +
|
| 86 | +This implies the next report/review is due around that date. |
| 87 | + |
| 88 | +### 2.2 Report Content/Scope Requirements |
| 89 | + |
| 90 | +Resolutions often specify what reports should cover: |
| 91 | + |
| 92 | +**Example from A/RES/78/70 (Mine Action):** |
| 93 | +> "report on the implementation of the present resolution and the progress made in mine action" |
| 94 | +
|
| 95 | +**Example from A/HRC/RES/28/6 (Albinism):** |
| 96 | +> Mandate includes: |
| 97 | +> - "(c) To promote and report on developments towards and the challenges and obstacles to the realization of the enjoyment of human rights" |
| 98 | +> - "(d) To gather, request, receive and exchange information... on violations of the rights" |
| 99 | +
|
| 100 | +### 2.3 Previous/Related Resolutions |
| 101 | + |
| 102 | +The preamble typically lists all predecessor resolutions: |
| 103 | + |
| 104 | +**Example from A/RES/79/150:** |
| 105 | +> "Recalling its resolutions 44/82 of 8 December 1989, 50/142 of 21 December 1995, 52/81 of 12 December 1997, 54/124 of 17 December 1999, 56/113 of 19 December 2001..." |
| 106 | +
|
| 107 | +This provides: |
| 108 | +- Complete resolution chain history |
| 109 | +- Dates showing reporting pattern over time |
| 110 | +- Related topic references |
| 111 | + |
| 112 | +### 2.4 Responsible Entity |
| 113 | + |
| 114 | +**Primary entity assignment:** |
| 115 | +> "**Requests the Secretary-General** to submit..." |
| 116 | +> "**Requests the Independent Expert** to..." |
| 117 | +> "**Invites the Special Rapporteur** to..." |
| 118 | +
|
| 119 | +**Supporting entities:** |
| 120 | +> "in collaboration with relevant stakeholders" |
| 121 | +> "working with the Special Jurisdiction for Peace" |
| 122 | +
|
| 123 | +### 2.5 Mandate Duration (for Special Procedures) |
| 124 | + |
| 125 | +**Example from A/HRC/RES/28/6:** |
| 126 | +> "Decides to appoint, **for a period of three years**, an Independent Expert" |
| 127 | +
|
| 128 | +--- |
| 129 | + |
| 130 | +## 3. LLM Extraction Strategy |
| 131 | + |
| 132 | +### Recommended Prompt Structure |
| 133 | + |
| 134 | +``` |
| 135 | +Given the following UN resolution text, extract: |
| 136 | +
|
| 137 | +1. REPORTING MANDATE (if any): |
| 138 | + - Target session/date for next report |
| 139 | + - Implied frequency (annual/biennial/etc.) |
| 140 | + - Reporting entity (Secretary-General, Special Rapporteur, etc.) |
| 141 | + - Report topic/scope requirements |
| 142 | +
|
| 143 | +2. RESOLUTION CHAIN: |
| 144 | + - List of previous related resolutions mentioned |
| 145 | + - Pattern of sessions/years (for frequency inference) |
| 146 | +
|
| 147 | +3. MANDATE DETAILS (if establishing/extending a mandate): |
| 148 | + - Duration |
| 149 | + - Key tasks assigned |
| 150 | + - Review/renewal date |
| 151 | +
|
| 152 | +Return structured JSON with confidence scores. |
| 153 | +``` |
| 154 | + |
| 155 | +### Expected Output Schema |
| 156 | + |
| 157 | +```json |
| 158 | +{ |
| 159 | + "reporting_mandate": { |
| 160 | + "exists": true, |
| 161 | + "target_session": "80th session", |
| 162 | + "target_date": null, |
| 163 | + "implied_frequency": "biennial", |
| 164 | + "frequency_confidence": 0.85, |
| 165 | + "responsible_entity": "Secretary-General", |
| 166 | + "scope": "implementation of the present resolution and progress made in mine action", |
| 167 | + "explicit_frequency_mentioned": false |
| 168 | + }, |
| 169 | + "resolution_chain": [ |
| 170 | + {"symbol": "A/RES/76/74", "date": "2021-12-09"}, |
| 171 | + {"symbol": "A/RES/74/80", "date": "2019-12-13"} |
| 172 | + ], |
| 173 | + "mandate_details": { |
| 174 | + "duration": null, |
| 175 | + "tasks": [], |
| 176 | + "review_date": null |
| 177 | + } |
| 178 | +} |
| 179 | +``` |
| 180 | + |
| 181 | +### Validation Approach |
| 182 | + |
| 183 | +1. **Cross-reference with historical data**: Compare LLM-inferred frequency with actual publication history |
| 184 | +2. **Session arithmetic**: Verify target session math (78th + 2 = 80th for biennial) |
| 185 | +3. **Entity normalization**: Map extracted entities to canonical department names |
| 186 | + |
| 187 | +--- |
| 188 | + |
| 189 | +## 4. Resolutions Without Report Mandates |
| 190 | + |
| 191 | +Not all resolutions mandate reports. Some resolutions: |
| 192 | +- Establish procedures (E/RES/2020/5 on statistics coordination) |
| 193 | +- Make declarations |
| 194 | +- Request actions other than reporting |
| 195 | + |
| 196 | +**LLM should indicate `reporting_mandate.exists: false` for these.** |
| 197 | + |
| 198 | +--- |
| 199 | + |
| 200 | +## 5. Sample Analyses |
| 201 | + |
| 202 | +### A/RES/78/70 (Assistance in Mine Action) |
| 203 | + |
| 204 | +| Field | Extracted Value | |
| 205 | +|-------|-----------------| |
| 206 | +| Target session | 80th session | |
| 207 | +| Implied frequency | Biennial | |
| 208 | +| Entity | Secretary-General | |
| 209 | +| Scope | "implementation of the present resolution and progress made in mine action" | |
| 210 | +| Previous resolutions | 76/74, many others since 1990s | |
| 211 | + |
| 212 | +### S/RES/2754 (2024) (Colombia Verification Mission) |
| 213 | + |
| 214 | +| Field | Extracted Value | |
| 215 | +|-------|-----------------| |
| 216 | +| Target date | 31 October 2025 | |
| 217 | +| Implied frequency | Annual (mandate extension) | |
| 218 | +| Entity | Secretary-General (via Verification Mission) | |
| 219 | +| Scope | Implementation of 2016 Final Peace Agreement | |
| 220 | +| Mandate duration | Extended until Oct 2025 | |
| 221 | + |
| 222 | +### A/HRC/RES/28/6 (Independent Expert on Albinism) |
| 223 | + |
| 224 | +| Field | Extracted Value | |
| 225 | +|-------|-----------------| |
| 226 | +| Target session | 31st session (HRC) | |
| 227 | +| Implied frequency | Annual (to HRC and GA) | |
| 228 | +| Entity | Independent Expert on Albinism | |
| 229 | +| Mandate duration | 3 years | |
| 230 | +| Scope | 8 specific mandate areas listed | |
| 231 | + |
| 232 | +--- |
| 233 | + |
| 234 | +## 6. Implementation Recommendations |
| 235 | + |
| 236 | +### Phase 1: LLM Extraction Pipeline |
| 237 | + |
| 238 | +1. **Fetch resolution fulltext** (already implemented) |
| 239 | +2. **Run LLM extraction** with structured prompt |
| 240 | +3. **Store extracted fields** in new table `resolution_mandates`: |
| 241 | + ```sql |
| 242 | + CREATE TABLE resolution_mandates ( |
| 243 | + resolution_symbol TEXT PRIMARY KEY, |
| 244 | + target_session TEXT, |
| 245 | + target_date DATE, |
| 246 | + inferred_frequency TEXT, |
| 247 | + frequency_confidence FLOAT, |
| 248 | + responsible_entity TEXT, |
| 249 | + report_scope TEXT, |
| 250 | + mandate_duration TEXT, |
| 251 | + previous_resolutions TEXT[], |
| 252 | + extracted_at TIMESTAMPTZ, |
| 253 | + llm_model TEXT |
| 254 | + ); |
| 255 | + ``` |
| 256 | + |
| 257 | +### Phase 2: Report-Resolution Linking Enhancement |
| 258 | + |
| 259 | +1. **Join reports to resolution mandates** |
| 260 | +2. **Compare inferred vs. actual frequency** |
| 261 | +3. **Flag discrepancies** for review |
| 262 | + |
| 263 | +### Phase 3: Survey Integration |
| 264 | + |
| 265 | +Use extracted data to inform survey: |
| 266 | +- "This report is mandated to cover: [scope]" |
| 267 | +- "Current frequency: [actual] vs. Mandated: [inferred]" |
| 268 | +- "Mandate expires: [date]" |
| 269 | + |
| 270 | +--- |
| 271 | + |
| 272 | +## 7. Cost-Benefit Analysis |
| 273 | + |
| 274 | +### LLM vs. Regex |
| 275 | + |
| 276 | +| Aspect | Regex | LLM | |
| 277 | +|--------|-------|-----| |
| 278 | +| **Development time** | High (many patterns) | Low (one prompt) | |
| 279 | +| **Maintenance** | High (new patterns) | Low (self-adapting) | |
| 280 | +| **Accuracy** | Medium (misses context) | High (understands intent) | |
| 281 | +| **Cost per resolution** | ~$0 | ~$0.01-0.05 | |
| 282 | +| **Handles edge cases** | Poor | Good | |
| 283 | +| **Confidence scoring** | No | Yes | |
| 284 | + |
| 285 | +**Recommendation:** Use LLM extraction with validation against historical data. |
| 286 | + |
| 287 | +### Estimated Costs |
| 288 | + |
| 289 | +- ~5,000 resolutions to process |
| 290 | +- ~$0.03 per resolution (GPT-4o mini or Claude Haiku) |
| 291 | +- **Total: ~$150 one-time extraction** |
| 292 | +- Can be run incrementally as new resolutions are added |
| 293 | + |
| 294 | +--- |
| 295 | + |
| 296 | +## Appendix: Resolution Examples |
| 297 | + |
| 298 | +### Example 1: Explicit Reporting Request |
| 299 | + |
| 300 | +From **A/RES/78/70**: |
| 301 | +``` |
| 302 | +Requests the Secretary-General to submit to the General Assembly at its |
| 303 | +eightieth session a report on the implementation of the present resolution |
| 304 | +and the progress made in mine action, and to include in that report an |
| 305 | +appendix containing information provided by Member States. |
| 306 | +``` |
| 307 | + |
| 308 | +### Example 2: Mandate Establishment |
| 309 | + |
| 310 | +From **A/HRC/RES/28/6**: |
| 311 | +``` |
| 312 | +Decides to appoint, for a period of three years, an Independent Expert on the |
| 313 | +enjoyment of human rights by persons with albinism, with the following mandate: |
| 314 | +(a) To engage in dialogue and consult with States... |
| 315 | +(b) To identify, exchange and promote good practices... |
| 316 | +(c) To promote and report on developments... |
| 317 | +(h) To report to the Human Rights Council, starting from its thirty-first session, |
| 318 | + and to the General Assembly; |
| 319 | +``` |
| 320 | + |
| 321 | +### Example 3: Mandate Extension (Security Council) |
| 322 | + |
| 323 | +From **S/RES/2754 (2024)**: |
| 324 | +``` |
| 325 | +1. Decides to extend the mandate of the Verification Mission until 31 October 2025; |
| 326 | +2. Expresses its willingness to work with the Government of Colombia on |
| 327 | + the further extension of the mandate... |
| 328 | +``` |
| 329 | + |
| 330 | +### Example 4: No Report Mandate |
| 331 | + |
| 332 | +From **E/RES/2020/5** (Statistics Coordination): |
| 333 | +- Contains procedural requests to coordinate statistical programmes |
| 334 | +- No reporting mandate to Secretary-General |
| 335 | +- LLM should return `reporting_mandate.exists: false` |
0 commit comments