Artificial Intelligence

General

This machine learning model tags English natural language queries relating to precious metals, quantities, weights, purities, and currencies. It handles a broad variety of units across geographical regions and time periods and can infer a variety of additional ones from context. It has support for metals, units, and currencies, in common formats (e.g. symbols, slangs, full names, etc.). This model can serve as a step in the process of developing a simple chatbot, as has been incorporated into HullBreach Studios' Loot Flipping. This initial version of the ML model was release with Loot Flipping Pro v6.0.0 in July 2024. (Later releases added support for more currencies and refined training data in both Loot Flipping Pro and Loot Flipping Lite.).

Type

• Natural Language
• Parts-of-Speech Tagger
• Conditional Random Field Algorithm

Trained Labels

• Currencies: AUD, BRL, CAD, EUR, GBP, MXN, USD
• Metals: gold, iridium, palladium, platinum, rhodium, ruthenium, silver, inference of base metals
• Purities: gold-filled, karat, millesimal, percentage, specific ("dental", "fine", "pure", "ster", "sterling")
• Quantities: integers, decimals, several written forms, indefinite articles
• Units: ancient (drachma, shekel, tael), avoisdupois/imperial, carat, metric, troy, broad ranges

Formatting and Samples

• Input format: Lowercase with punctuation stripped
• Possible tagged labels: CURRENCY, METAL, PURITY, QUANTITY, UNIT, and NONE
• Example labels:
  • what's (NONE) the (NONE) value (NONE) of (NONE) 25.2 (QUANTITY) dwt (UNIT) of (NONE) 14kt (PURITY) gold (METAL)
  • how (NONE) much (NONE) for (NONE) iridium (METAL) per (NONE) troy (UNIT) ounce (UNIT) in (NONE) usd (CURRENCY)
  • spot (NONE) price (NONE) of (NONE) sterling (PURITY) silver (METAL)
  • five (QUANTITY) ozt (UNIT) platinum (METAL) coins (NONE)
  • 85 (QUANTITY) grams (UNIT) of (NONE) 14/20 (PURITY) gold-filled (METAL) in (NONE) euros (CURRENCY)
  • 2 (QUANTITY) pounds (UNIT) sterling (PURITY) silver (METAL) in (NONE) pounds (CURRENCY) sterling (CURRENCY)

Usage (Swift)

import NaturalLanguage

let inputText = ""

if let mlModel = try? MetalSpot(configuration: MLModelConfiguration()).model,
	let metalSpotModel = try? NLModel(mlModel: mlModel) {

	let tagger = NLTagger(tagSchemes: [NLTagScheme("PriceBot")])
	tagger.string = inputText.lowercased()
	tagger.setModels([metalSpotModel], forTagScheme: NLTagScheme("PriceBot"))

	let tags = tagger.tags(in: inputText.startIndex..<inputText.endIndex, unit: .word, scheme: NLTagScheme("PriceBot"), options: [.omitWhitespace, .omitPunctuation, .joinContractions]).map({ tag, range in

		return ["word": String(inputText[range]), "tag": tag?.rawValue ?? ""]

	}).filter({ $0["tag"] != "NONE" })
}
				

General

This model tags two currencies and a quantity for one of those currencies to determine exchange rate conversion.

Type

• Natural Language
• Parts-of-Speech Tagger
• Conditional Random Field Algorithm

Formatting and Samples

• Input format: Lowercase with punctuation stripped
• Possible tagged labels: FROMQUANTITY, FROMCURRENCY, TOCURRENCY, and NONE
• Example labels:
  • five (FROMQUANTITY) dollars (FROMCURRENCY) and (NONE) forty-three (FROMQUANTITY) cents (FROMCURRENCY) in (NONE) british (TOCURRENCY) pounds (TOCURRENCY)
  • what (NONE) is (NONE) two (FROMQUANTITY) benjamins (FROMCURRENCY) to (NONE) canadian (TOCURRENCY) dollars (TOCURRENCY)
  • euros (FROMCURRENCY) to (NONE) japanese (TOCURRENCY) yen (TOCURRENCY)