orange3-nlp

This provides a collection of widgets for Natural Language Processing.

Installation

Within the Add-ons installer, click on "Add more..." and type in orange3-nlp

Widgets

• General Widgets
  • Abstractive Summary
  • Extractive Summary
  • Named Entity Recognition
  • POS Tagger
  • POS Viewer
  • Question Answering
  • Reference Library
  • Ollama RAG

• Text Splitting Widgets
  • Text Chunker
  • Tokens to Corpus

• Text Embedding Models
  • Doc2Vec
  • E5
  • FastText
  • Gemini
  • Nomic
  • OpenAI
  • Sentence Embedder (SBERT)
  • spaCy
  • USE

• Training of Text Embedding Widget
  • Train Doc2Vec

• For Polish Sentiment Analysis
  • Analiza Sentymentu

Summary Widgets

• Extractive Summary: Selects and joins key sentences or phrases from the original text.

• Abstractive Summary: Generates new sentences that paraphrase and condense the original content (more similar to how humans summarize).

Named Entity Recognition

Named Entity Recognition (NER) is a task in NLP that locates and classifies named entities in text into predefined categories such as:

• PERSON – names of people
• ORG – organizations
• GPE – countries, cities, or locations
• DATE, TIME, MONEY, etc.

Part of Speech Tagging

Part-of-speech (POS) tagging assigns grammatical categories to each word in a sentence.

Common POS Tags

POS tagging is essential for syntactic parsing and downstream NLP tasks.

Part of Speech Viewer

This uses spaCy's displacy HTML renderer to provide a parsed dependency tree of the parts of speech of the input text.

Question Answering

Question Answering (QA) systems aim to extract or generate answers to user questions from a text or knowledge base.

Text Splitting Widgets

Tokens to Corpus

The Tokens to Corpus widget takes the tokens from the Preprocess Text widgets.

Text Chunker

Text Chunker supports 2 chunking strategies to split text. The first is LangChain's RecursiveCharacterTextSplitter and the second is semantic-text-splitter.

Reference Augmented Generation

Reference Augmented Generation (RAG) is a method of enhancing large language model (LLM) responses by providing external documents as supporting context. Instead of relying solely on the model's training data, RAG:

• Retrieves relevant snippets from a document collection (knowledge base).
• Augments the prompt to the LLM by including this retrieved content.
• Generates a more accurate and grounded answer based on the context.

Let's take a look at the Reference Library

And lastly, let's look at the Ollama RAG use.

Polish Sentiment Analysis

Since Polish sentiment analysis support in Orange was limited, Analiza Sentymentu provides a tuned model.