Is it okay to use Brain.JS to sort products in Web Shop?

The Project

We are running our own-coded web shop with dozen of products. There are around 100.000 different products, where around 50.000 categorized (are linked to category, for example HP LaserJet MFP 2000 -> Printer) and another 50.000 don’t have any category. The only possible train data would be like this JSON:

  [{
    "input": "Toner schwarz für HP Laserjet P3015 12.500 Seiten",
    "output": "Toner"
  },
  {
    "input": "alternativ Toner Brother MFC 9142 2.200 Seiten cyan",
    "output": "Toner"
  },
  {
    "input": "Laserdrucker HL-L6300DW inkl. UHG, Filter und Halterung mit intregiertem NFC-Kartenleser, 46 Seiten/Minute Druckgeschwindigkeit, 520 Blatt-Papierkassette, Auflösung 1200 x 1200 dpi, USB 2.0 Hi-Speed, LAN, WLAN, integrierte BSI-Schnitt- stelle, 4,5 cm Touc",
    "output": "Laserdrucker"
  },
  {
    "input": "Laserdrucker HL-L6400DW inkl. UHG, mit intregiertem NFC-Kartenleser,, Brother",
    "output": "Laserdrucker"
  }]

How important is this (1-5)?

3

Expected behavior

By checking “Toner TN-3430, schwarz für DCP-L5500DN, DCP-L6600DW, HL-L5000D, Brother” get “Toner” Category.

Other Comments

Would it be wise to use Brain.JS for that? That would probably take years to train over 50k items?

[Macil]

When the output is a string, brain.js doesn’t recognize the output as a distinct category label. If you have some training items with the output “foo”, and other training items with the output “bar”, Brain.js learns logic like “for items like the first group, include the ‘o’ and ‘f’ characters more often in the output than ‘a’, ‘b’, and ‘r’”.

You should have the output be {foo: 1} for things in the “foo” category and {bar: 1} for things in the “bar” category. Then your output will be like {foo: 0.8, bar: 0.1} for things that Brain thinks probably belong in “foo”, etc.

[mubaidr]

Should be 1, 0. just like probability (max is always 1).