Application Project with FAST AI Movies

Extracting Semantic Relationships from Unstructured Textual Data in eLearning Video Scripts

Description

FAST AI Movies is an innovative AI startup offering a Software as a Service (SaaS) for eLearning content creation. Their solution uses LLMs to request short scripts from customers and generate coherent training videos. The project focuses on extracting structural and semantic information from textual data. Key components of the system include Chapter Segmentation, Text Generation Transformer, Keyphrase Extraction, Relation Classification, and an Icon Database.

1. Chapter Segmentation: This component breaks down lengthy text into manageable chapters using techniques like sentence similarity and positional encoding to identify content boundaries.

2. Text Generation Transformer: It expands unstructured text (short sentences and bullet points) into understandable sentences, forming the basis for video transcripts. It also controls the tone and prevents hallucinations.

3. Keyphrase Extraction: Identifies and extracts important keywords or phrases from the generated text, which are used to structure the video transcript and capture relationships between main points.

4. Relation Classification: Determines and categorizes relationships between keyphrases, including pair relations (e.g., causal and parallel relationships) and root relations (identifying important phrases). It aims to create a hierarchical relationship taxonomy.

5. Icon Database: Responsible for finding icons that represent key phrases and relations from the Relation Classification component to include in the output video.

The main goal is to automate the extraction and representation of information from textual data, creating a nested taxonomy of keyphrases to enhance e-learning video content. This project aims to eliminate the need for human intervention, making FAST AI Movies' SaaS more efficient and productive.