This model was developed specifically for the LockIn browser extension that boosts productivity by blocking irrelevant videos. It is not intended to be a general-purpose relevance classifier. There are both .onnx and .safetensors models

Input

The model takes two inputs:

goal — the user's current focus goal
text — the content of the web page, video, or post

'text' is constructed in a form:

text

Title: <title>

Description: <description>

Tags: <up to 5 tags>

Headings H1: <H1 headings>

Headings H3: <H3 headings>

Paragraphs: <paragraphs>agraphs:

Headings and Paragraphs are basically desciption and comments in a reddit post.

Only first five of tags, Headings H3 and Paragraphs are taken to save tokens.

The two texts are passed to the model as a text pair.

Example

json

{
  "goal": "learn chemistry",
  "text": "Title: 19. Spectroscopy: Probing Molecules with Light\nDescription: MIT 5.61 Physical Chemistry, Fall 2017\nInstructor: Professor Robert Field\nView the complete course: \nYouTube Playlist: \n\nThis lecture discusses time-dependent quantum mechanics.\n\nLicense: Creative Commons BY-NC-SA\nMore information at \nMore courses at\nTags: 5-61-physical-chemistry-fall-2017, dipole approximation, fermi's golden rule, linear response, quantum mechanics",
  "label": 1
},
In this example, the video is matched with the goal, therefore allowed

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

MODEL_NAME = "nsagatov1/youtube-video-relevance-classifier"

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)

goal = "learn chemistry"

text = """
Title: Atomic spectra | Physics | Khan Academy
Description: Courses on Khan Academy are always 100% free. Start practicing—and saving your progress—now!

Electrons only exist at specific, discrete energy levels in an atom.
Tags: online learning, online class, video class, video tutorial, online education
"""

inputs = tokenizer(
    goal,
    text_pair=text,
    truncation=True,
    max_length=512,
    return_tensors="pt"
)

with torch.no_grad():
    outputs = model(**inputs)

prediction = torch.argmax(outputs.logits, dim=-1).item()

labels = {
    0: "denied",
    1: "allowed"
}

print(labels[prediction])