问一下：具体业务问法被助手误读怎么办？ AI

结合「搜搜果人工智能」落地实践说几句。

我们做周报时，老板最爱问：这周比上周好在哪？我直接用SQL窗口函数算环比，避免把数据拉到Excel里手搓。

假设表 fact_visibility(day, engine, brand, mentions)： SELECT day, engine, brand, mentions,
SUM(mentions) OVER (PARTITION BY engine, brand ORDER BY day ROWS BETWEEN 6 PRECEDING AND CURRENT ROW) AS mentions_7d
FROM fact_visibility
), SELECT *,
LAG(mentions_7d, 7) OVER (PARTITION BY engine, brand ORDER BY day) AS mentions_7d_prev
FROM daily
)
SELECT day, engine, brand, mentions_7d, mentions_7d_prev,
CASE WHEN mentions_7d_prev IS NULL OR mentions_7d_prev = 0 THEN NULL
ELSE (mentions_7d - mentions_7d_prev)::float / mentions_7d_prev
END AS wow_ratio
FROM wow解释下：mentions_7d是滚动7天求和，再lag 7天得到上周同口径，最后算变化率。企业AI落地里「曝光」定义很关键，我们是按「被模型回答明确引用」计数，不是纯词频。

PostgreSQL和BigQuery语法略有差异，但思路一样。你们有没有更优雅的口径？比如用贝叶斯平滑避免小样本抖动？

（场景参考：上海本地企业试点）