Score – rate on a scale

The Score node lets Laya rate a text on a scale of ordered levels, for example from “not at all” to “extremely” urgent. Laya returns one number, score, that can lie between two levels, plus the probability of every level.

When to use Score

• The answers go from “little” to “a lot”: urgency, anger, satisfaction, risk, complexity, politeness.
• You want a number to compare against thresholds, for example “escalate from 3 upwards”.

Score vs. Choice

Choice picks exactly one answer. Score understands that the levels build on each other, and its result shows where between the levels the text lies. Example with the levels low, medium, high, critical:

In the second case Choice would report “critical” although Laya leans almost as much towards “low”. The Score value and the probabilities make this visible.

How the score is calculated

Each level gets a number by its position: the first level is 0, the second 1, and so on. Laya assigns a probability to every level. score is the average of the level numbers, weighted by these probabilities:

0 × 0.01 + 1 × 0.17 + 2 × 0.47 + 3 × 0.35 = 2.15

The result 2.15 is a bit above “high”, leaning towards “critical”. With five levels the score lies between 0 and 4. With two levels it lies between 0 and 1.

Settings

Input

Output

Options

Choosing good levels

The level names tell Laya what “low” and “high” mean, so they matter a lot.

Describe the levels in words for important decisions, like Laya’s own templates do. Two examples for the Levels list, one row per line:

no time pressure
needs attention soon
blocking issue or hard deadline

You can also put an explanation after a colon:

none: ordinary request
minor: mildly inappropriate
serious: unsafe advice or abuse
severe: dangerous or illegal

The whole row, including the explanation, is the level name that Laya sees and returns in legend. Laya does not support a separate description field for levels. That is why the list has only one column.

Levels from a variable

Levels from a variable must be a list, because the order matters:

msg.levels = ["calm and neutral", "concerned but civil", "clearly annoyed", "very angry"];

Result

{
  "type": "score",
  "score": 1.6157,
  "legend": { "0": "low", "1": "medium", "2": "high", "3": "critical" },
  "probabilities": { "0": 0.05, "1": 0.4815, "2": 0.33, "3": 0.14 },
  "confidence": 0.3,
  "answer_confidence": 0.4815,
  "action": { "act_probability": 1.0 },
  "model": "laya-rl-agent",
  "usage": { "truncated": false, "...": "…" },
  "routing": { "model": "english", "...": "…" }
}

Getting the name of the level

Quick way: round the score. This gives the level closest to the average:

msg.laya.legend[Math.round(msg.laya.score)]

Most likely level: when Laya is torn, the rounded score can differ from the level with the highest probability. To get that one, use this in a Function node:

var a = msg.laya;
var best = null;
 
Object.keys(a.probabilities).forEach(function (k) {
  if (best === null || a.probabilities[k] > a.probabilities[best]) {
    best = k;
  }
});
 
msg.level = a.legend[best];   // for example "medium"
return msg;

Branching with a Switch node

Order the conditions from strictest to loosest:

msg.laya.score >= 3          // very or extremely urgent → escalate
msg.laya.score >= 1.5        // moderately urgent → normal queue
msg.laya.score < 1.5         // not or slightly urgent → low priority

Batch mode

With Batch on, every record in Text is rated on the same scale in one request (up to 64 per request, more are split automatically).

// object of records → result keeps the IDs
msg.laya.T1.score
 
// list of records → result keeps the order
msg.laya[0].score

Errors