tasks/offline-compute/HiveSQL/hivesql_018_en/task.md

---
id: offline-compute_HiveSQL_hivesql_018
name: Compute scores for first-level comments from view records and comment logs. Group by recall_uin + channel_id + fe
category: offline-compute/HiveSQL
timeout_seconds: 600
modality: pure-text
engine: hivesql
---
## Prompt
Task Objective: Compute the composite score for first-level comments newly added by recalled read users after their first post view, for content heat analysis.

Inputs:
- `internal_platform_db.dws_social_group_content_forum_hot_feed_recall_feed_view_hi_query_engine_122` (view records, filter `imp_hour` within [2026060811, 2026060910])
- `internal_platform_db.dwd_social_group_content_forum_hot_feed_recall_comment_log_hi_query_engine_122` (comment logs, filter `imp_hour` within [2026060811, 2026060910])
- `internal_platform_db.dwd_all_social_group_user_slice_ds_query_engine_122` (user identity, filter `imp_date >= 20260607`)

Processing Rules:
1. Build a view aggregation: From the view table, aggregate by `uin` + `channel_id` + `feed_id`, extracting the first/last view time, detail page view count, and duration.
2. Filter comments: Join the comment log with the view aggregation on `channel_id` + `feed_id`, keeping only records where the comment time is >= the first view time.
3. Enrich user identity: Left join the user table, using `member_role` as a fallback for `user_type` (use the original `user_type` when missing).
4. Compute first-level comment metrics: Group by `recall_uin` + `channel_id` + `feed_id` + `p_comment_id` (as the first-level comment ID) + `uin`, and compute:
   - Normal user like count (`action_type='comment_like'` AND `user_type=0`)
   - Author like count (`action_type='comment_like'` AND `uin=author_uin`)
   - Channel owner like count (`action_type='comment_like'` AND `user_type in (1,2)`)
   - Normal user reply count (`action_type='comment'` AND `comment_type='comment_reply'` AND `user_type=0` AND `uin<>comment_uin`)
   - Author reply count (`action_type='comment'` AND `comment_type='comment_reply'` AND `uin=author_uin` AND `uin<>comment_uin`)
   - Channel owner reply count (`action_type='comment'` AND `comment_type='comment_reply'` AND `user_type in (1,2)` AND `uin<>comment_uin`)
5. Aggregate by `recall_uin` + `channel_id` + `feed_id` + `comment_id` to compute deduplicated UV and totals:
   - UV (count distinct of users with the behavior) and counts for each like/reply type
   - Normal user average reply count (total replies / number of users who replied)
6. Compute scores (tiered rules):
   - Normal user like score: tiered by count (1→2, 2–4→4, 5–10→7, 11–20→8, 21–50→10, >50→12)
   - Author like score: tiered by UV, then +2
   - Channel owner like score: tiered by count, then +1
   - Normal user reply score: when count < 4, UV*2.5 (cap 30); when >= 4, UV*2 (cap 30)
   - Author reply score: normal user reply score + 3
   - Channel owner reply score: normal user reply score + 1
   - Average reply score: mean <=1.5→0, 1.5–3→3, >3→6
   - Total comment score: sum of the above 7 score components

Output Requirements:
- Field order: `imp_hour`, `uin` (i.e., `recall_uin`), `channel_id`, `feed_id`, `comment_id`, `comment_time`, `comment_score`, and the scores, UVs, and counts for each like/reply type (20 metric fields in total)
- `comment_time` takes the last comment time of that comment
- Partition field `imp_hour` is fixed as `2026060910`

Write Requirements:
- Output table: `internal_platform_db.dws_social_group_content_forum_hot_feed_recall_comment_score_hi_cand_query_engine_122`
- Write method: `INSERT OVERWRITE` partition `imp_hour=2026060910`

Please write the final HiveSQL to `result.sql` and execute it.