Smart Video Search Enigne

Team: Diana Arita, Kartik Patel, Matthew Kryczka, Selvaganapathy Thirugnanam

Presentation: https://mediaspace.illinois.edu/media/t/1_tpvrr7g7

Demo: https://mediaspace.illinois.edu/media/t/1_mfluhe37

Documentation: https://github.com/Kartikp2/CourseProject/blob/main/Documents/CS-410%20Course%20Project%20Documentation.pdf

Problem Statement

There is no easy way to search by video content. This makes it hard for users who are interested in finding precise locations of the videos where a concept or a term is mentioned.

Abstract

Build a search engine that can support video segment search for Coursera lecture videos with transcripts. This would allow a user to type in a query and see a ranked list of short video segments so that the user can precisely locate which segment to watch in a lecture in order to know more about a concept. We intend to build the search engine based on the scrapping/indexing/ranking concepts learnt from this course. The end solution will provide a User Interface for the user to type in the query and to view the results as links to the video segments.

Project Overview

• Scrape the videos along with the transcripts (Coursera dl)
• Extract documents out of scrapped videos and transcripts and build an association between them
• Use Tokenizer(Stemming and other normalization techniques) to extract lexical units (words)
• Use an Indexer (inverted Index) for faster response from the Search Engine.
• Perform Ranking/Scoring based on Probabilistic retrieval functions (Eg: BM25)

Software Used

• Coursera-dl package: https://github.com/coursera-dl/coursera-dl
• Metapy library: https://github.com/meta-toolkit/metapy

How to run UI!

Download code from github: https://github.com/Kartikp2/CourseProject.git

• Run backend
  • Be in the directory! ( same as server.py)
  • To install the dependencies - npm install
  • Run server.py
• Run Frontend
  • Be in the directory (search-ui -> smart-video-search)
  • To install the dependencies - npm install
  • To run app - npm start
  • Go to browser and hit url https://localhost:3000

Files

Not all of the coursera-dl files were uploaded as they take up a lot of space ( the project is already close to 1GB!)

• Documents
  • Contains proposal, progress report, powerpoint, and final documentation
• Course
  • Contains course data for ranking and line.toml file
• idx/inv
  • Inverted index for faster access to data
• initial_analysis
  • Snippet of what dat looks like coming out of coursera-dl package, segment_extractor.py that converts the srt data and links multiple data into a csv
• search-ui
  • Code base for the frontend
• test_flask
  • Sample flask that returns the results that the frontend expects
• video_thumbnails
  • File that conatins images of video thumbails (obtained from the segment_extractor.py)
• coursera_video_lessons
  • Used by the search engine to build the data and used for ranking
  • Each line is a lecture
• coursera_video_segments
  • Used by backend to further narrow query results
  • Each line is a transcript segment
• search_engine.py
  • Using Metapy library to build the search ranker. Build corpus, inverted index, and use OkapiBM25 to rank queries
• server.py
  • Backend file that fetches the query from UI and runs it on the search engine. Finally returns the most relevant video segment