Edge-ML GloVe NLP JavaScript Library
Vui Huang Tea
Edge-ML GloVe NLP is a high-speed, low-footprint NLP (Natural Language Processing) EDGE-ML (Edge Machine Learning) JavaScript library that can be used to enhance and derive more info from 'TensorFlow.js' (TensorFlow JavaScript) classification predictions. For example, Google Imagenet has 1000 classes and the labels consists of 1862 unique words. With this client-side NLP library, the textual description can be broaden to 60,000 words.
This is accomplished by applying GloVe-based NLP (Natural Language Processing) on the classification labels. The most frequently associated words of the top labels are gathered, matched, ranked and sorted. This is essentially an ML charades expert that deduces what the unidentified object is by using the top labels as word clues.
TensorFlow suggests using a model that is of 30 MB in size or less in the browser. The NLP data in this ML-Edge library is customizable from 4-34 MB, depending on word-depth requirement from 10-100 of the most frequently used words of each vocabulary.
