评论解码

2 星 — 问题不小

是什么驱动的:客服 · 会推荐:多半不是

一条用户评论真正在告诉你什么。

Jev 读到的 state

你好——接着我前两封邮件再跟进一下。三月的发票还没批下来,我的团队在它结清之前没法开始下一阶段。我知道大家都忙翻了,但这已经比我们在 SOW 里约定的日期晚了三周。有人能告诉我卡在哪一环吗?需要的话我可以把采购单再发一遍。

9
问题
1
次请求
235毫秒
往返耗时
878
输入 token
US$0.000037
费用
jev-1.13.0
模型

领衔的几条解读

暗含星级

Score

文本本身支撑的星级,与它给出的任何数字无关。

87%
1.03 满分 4 · 2 星 — 问题不小

是什么驱动的

Choice

这个评价实际挂在哪个方面上。

39%
客服 占 49%,共 6 个选项
49% 客服

会推荐

Noul

他们会不会让朋友买。

多半不是
答案为「是」的概率
18%
抛硬币

同一次请求里的其余部分

读起来真实

Noul

亲历细节对抗泛泛的赞美。

偏向否
答案为「是」的概率
34%
抛硬币

可挽回

Noul

一次好的回应能不能把这个客户拉回来。

多半是
答案为「是」的概率
79%
抛硬币

细节

Score

产品团队能据以行动的东西有多少。

52%
2.71 满分 4 · 好几个具体问题

情绪强度

Score

不论哪个方向,感受有多强烈。

88%
1.92 满分 4 · 明显在意

提到竞品

Noul

文本里有一处比较。

几乎肯定不是
答案为「是」的概率
1%
抛硬币

值得回复

Noul

哪些地方公开回应是划算的。

多半不是
答案为「是」的概率
18%
抛硬币

分享这条解读

拿到链接的人都能看,并且会出现在「浏览」里。

发到 X
产生这条结果的确切请求

POST https://api.typesafe.ai/v1/systemone

{
  "state": "你好——接着我前两封邮件再跟进一下。三月的发票还没批下来,我的团队在它结清之前没法开始下一阶段。我知道大家都忙翻了,但这已经比我们在 SOW 里约定的日期晚了三周。有人能告诉我卡在哪一环吗?需要的话我可以把采购单再发一遍。",
  "model": "jev-latest",
  "questions": {
    "implied_rating": {
      "type": "score",
      "instructions": "What star rating does this review text imply?",
      "criteria": [
        "1 star — would warn others off",
        "2 stars — significant problems",
        "3 stars — mixed, real caveats",
        "4 stars — good with minor gripes",
        "5 stars — unreserved recommendation"
      ]
    },
    "driving_aspect": {
      "type": "choice",
      "instructions": "Which aspect most drives this reviewer’s verdict?",
      "criteria": {
        "Build quality": "Materials, durability, physical feel",
        "Software": "Apps, firmware, bugs, updates",
        "Value": "Price relative to what you get",
        "Support": "The experience of getting help",
        "Delivery": "Shipping, packaging, fulfilment",
        "Performance": "Speed, battery, accuracy, capacity"
      }
    },
    "is_authentic": {
      "type": "noul",
      "instructions": "Does this read as a genuine first-hand review?",
      "criteria": {
        "true": "Specific lived detail an actual owner would know",
        "false": "Generic, templated, or promotional in a way real users are not"
      }
    },
    "would_recommend": {
      "type": "noul",
      "instructions": "Would this reviewer recommend the product to a friend?"
    },
    "is_recoverable": {
      "type": "noul",
      "instructions": "Could a good response from the company turn this reviewer around?",
      "criteria": {
        "true": "The complaint is fixable and the reviewer is open to it",
        "false": "They are gone, or the problem is fundamental to the product"
      }
    },
    "detail_level": {
      "type": "score",
      "instructions": "How much actionable detail does this review contain?",
      "criteria": [
        "None — pure reaction",
        "A vague direction",
        "One concrete issue",
        "Several concrete issues",
        "A reproducible account a team could work from"
      ]
    },
    "emotional_intensity": {
      "type": "score",
      "instructions": "How emotionally intense is this review?",
      "criteria": [
        "Flat and factual",
        "Mild feeling",
        "Clearly invested",
        "Strong feeling",
        "Extreme — love or fury"
      ]
    },
    "mentions_competitor": {
      "type": "noul",
      "instructions": "Does the review compare the product to a competitor or alternative?"
    },
    "should_respond": {
      "type": "noul",
      "instructions": "Should the company publicly respond to this review?"
    }
  }
}

把 Jev 指向你自己的文本

九个类型化答案,一次请求,大约半秒。挑一个镜头,或者自己写问题。