README.md

---
license: mit
language:
- en
pretty_name: Internship Support-Query Resolution Dataset
size_categories:
- 1K<n<10K
task_categories:
- text-classification
tags:
- education
- internship
- student-support
- helpdesk
- query-resolution
- learning-analytics
- tabular
---
# Internship Support-Query Resolution Dataset

Support-query records from an internship programme's helpdesk: how each query
was raised, which theme it's about, and how it was resolved. 4,093 rows, one
row per query. No free-text query content is included — every row's content
is represented only by `query_theme_code`.

## Files

| File | Rows | Contents |
|---|---|---|
| `internship_query_resolution_export_2026-07-08_anonymized.csv` | 4093 | one row per query, 16 columns |
| `query_theme_distribution.csv` | 16 | `query_theme_code`, `query_theme_label`, `description`, `count`, `pct` — what each query was *about* |
| `resolution_pathway_distribution.csv` | 6 | `resolution_pathway_code`, `resolution_pathway_label`, `description`, `count`, `pct` — *how* each query got resolved |
| `LICENSE` | — | MIT License |

The two lookup files are separate because a query's topic and how it got
resolved are independent facts about it.

## Columns

| # | Column | Meaning |
|---|---|---|
| 1 | `query_id` | opaque internal identifier, unique per row |
| 2 | `raised_by_student_id` | pseudonymous asker id (`STUDENT_00001`…), consistent across rows for the same person |
| 3 | `raised_at` | ISO 8601 timestamp the query was raised |
| 4 | `query_theme_code` | what the query is about — look up the name in `query_theme_distribution.csv` |
| 5 | `status` | coarse resolution outcome: `human_resolved`, `resolved`, `resolved_by_asker`, `resolved_by_AI_agent`, `super_escalated` |
| 6 | `resolution_pathway_code` | how the query was resolved — look up the name in `resolution_pathway_distribution.csv` |
| 7 | `resolution_pathway_label` | the resolution-pathway name, inlined here for convenience |
| 8 | `trigger_source` | how the query entered the system: `header_link`, `legacy_hashtag`, `peer_raise`, `AI_agent_proactive`, or `NA` |
| 9 | `resolution_type_raw` | whether an FAQ action was taken while resolving: `find_faq`, `create_faq`, `custom_reply`, or `NA` |
| 10 | `linked_faq_id` | which of 114 existing FAQ entries was reused (`<topic>.<item>`, e.g. `4.3`), or `NA` |
| 11 | `linked_faq_id_source` | which underlying system field that FAQ link was read from, or `NA` |
| 12 | `responded_at` | ISO 8601 timestamp of the response, or `NA` |
| 13 | `resolver_type` | the system's resolver category, 10 values — see *Resolution pathway* below for how this relates to `resolution_pathway_code` |
| 14 | `resolver_student_id` | pseudonymous resolver id, same scheme as column 2, or `NA` — populated only when an individual person (not a script or the asker) resolved the query |
| 15 | `escalation_trigger` | `too_many_skips`, `too_many_ambiguous`, or `NA` |
| 16 | `accepted_at` | ISO 8601 timestamp of acceptance, or `NA` |

All blank cells are written as the literal string `NA`.

## Query theme

What the query is about: one of 16 categories (`T1`–`T16`), assigned by
reading each query's actual text against the category definitions below —
not derived from any other column. Names, full descriptions, counts, and
percentages are in `query_theme_distribution.csv`.

## Resolution pathway

How the query was resolved, independent of its topic. `status` alone is too
coarse to distinguish resolution pathways — for example, every peer-resolved
query (286 of them) carries the same `status = human_resolved` value as an
administrator-resolved query. `resolution_pathway_code` makes this
distinction explicit, grouping the system's 10 raw `resolver_type` values
into 6 categories:

| Code | Label | `resolver_type` values included |
|---|---|---|
| P1 | Admin bulk/script-resolved | `bulk_admin_script`, `bulk_admin_script_legacy` |
| P2 | Admin individually-resolved | `human_admin_direct`, `human_admin_direct_legacy`, `human_resolved_unattributed` |
| P3 | Peer-resolved | `peer_answered_no_formal_approval_record`, `peer_genuine_approve` |
| P4 | AI-agent resolved | `AI_agent` |
| P5 | Self-resolved | `self_resolved_by_asker` |
| P6 | Unresolved/other | `unresolved_or_other` |

`resolver_type` is kept alongside `resolution_pathway_code` because it is
more granular — it is the only column that distinguishes a legacy-era action
from a current-era one (e.g. `bulk_admin_script` vs.
`bulk_admin_script_legacy`, both grouped under `P1`).

## Column notes

**`resolver_student_id`** supports resolver-effort analysis (who resolves
the most queries). Its population does not follow `resolution_pathway_code`
cleanly: within `P1` (bulk-script), 570 of 2,000 rows carry an individual
id even though bulk-script resolution sounds automatic; within `P2` (direct
admin), only 45 of 1,157 rows do — most "direct" admin resolutions in this
data are not attributed to a specific person. This is a property of the
source system, not an artifact of this release.

**`escalation_trigger`** is set by the system's own internal logic (too
many skipped attempts, too much ambiguity) before a resolution pathway is
settled, independent of `resolution_pathway_code`. In this data it only
co-occurs with `P1`, `P2`, or `P5` — `P3` (peer) and `P4` (AI-agent) rows
never carry this flag.

## License

MIT License, see `LICENSE`.