Transcript Analysis – Graph + NLP
This program extracts insights from Zoom Meeting Transcripts (.vtt) using TigerGraph and NLTK.
In order to run this program, modify the auth.ini file with your proper graph solution credentials
and file paths. Then, simply run main.py. A sample transcript has been provided, but feel free to add your own into the \a_raw_transcripts directory!
As of now, this program performs the following tasks:
-
Convert
.vttinto compact version (stored in\b_cmt_transcripts) -
NLP analysis of compact transcript (using NLTK)
- Sentiment analysis
- Trigrams (collocations)
- Frequency of words (plotted)
- Meaningful words (shown as wordcloud)
- Number of speakers, names of speakers
- Who spoke the longest, least, average
-
Graph analysis of compact transcript (using TigerGraph)
- Analyze relationships between speakers
- Asked the most/least questions
- Pair w/ the most back-and-forth
- (TODO): Linking topics in semantic graph
- (TODO): Named-Entity Recognition
-
Visual output of all determined insights
Usage
A TigerGraph Cloud Portal solution (https://tgcloud.io/) will be required to run this program.
Kindly find the GraphStudio link here: https://transcript-analysis.i.tgcloud.io/
The schema utilized in this graph is fleshed out below:
Vertex: speaker
- (PRIMARY ID) name – STRING
Edge: asked_question
- text – STRING
Edge: answered_question
Here is an example of the graph populated with the sample transcript provided:
Analysis
Here is a screenshot of the command-line output produced:
Here is a frequency chart of meaningful words generated:
Here is a word cloud that visualizes common, key terms:
More features coming soon! In the meantime, feel free to continue creating and adding new insights ??
