pbusienei/conference_helper

This is Nashville Analytics Summit conference helper in searching for the top three sessions of your interest.

Nashville Analytics Summit conference helper

Overview

Since its inception in 2013 Nashville Analytics Summit event is held once a year in the months of September-October. There are dozens of talks covering analytics, data science, AI, career growth, new tools, specific industries and more. It has seen tremendous growth in sessions and participation increase.

The problem

The recent exponential need for data and its solutions, companies are looking forward to ways to leverage data in their organizations in order meet their business objectives. This has made it challenging for participants to navigate so many sessions and presentations at Nashville Analytics Summit. Therefore, if you are thinking of attending, Nashville Analytics Summit which talks might be of interest to you? Using a sentence similarity model, we'll pick out the top 3 talks for a participant based on their description of what they hope to get out of the Session/Talk.

This project utilizes Transformers sentence similarity model multi-qa-MiniLM-L6-cos-v1 to search for abstracts that have cosine similarity that matches the search query.The user will input a phrase of or a word of interest and the model will search for similarity contexts that matches the phrase and output the top 3 sessions that matches the phrase using Transformers model semantic search over conference abstracts.

The model

Semantic search models embed all the entries in the corpus (the abstracts) into a vector space. When the user inputs the query, its embed into the same vector space and the closest embeddings from the corpus (abstract) are found. Hence, the entries with high semantic overlap with query are ranked (high similarity at the top).

Two types of Semantic Search:

1. Symmetric - query and entries in corpus are about the same length and have same amount of content
2. Asymmetric - short query and output longer paragraphs
   This project uses Asymmetric semantic search.

Huggingface spaces

Please take a look at my spaces at Huggingfaces

Model Card

Model card is found here

Critical Analysis

• speech-to-text: Create a trained text to speech app using Speech2Text Models in conjunction with Wav2Vec2 (speech translation)
• domain-specific: Using domain specific may increase the accuracy of the model
• Different-models Trying out different models (models tuned for dot-product) that prefers retrieval of longer documents

Code Demonstration

To run the streamlit app install streamlit in your machine

From command line/shell navigate to the abstract-search folder

Run the command below:

streamlit run app.py

This will open a web interface in your browser like the one below;

Deploying the Streamlit App

Accessing the app locally

Local URL: http://localhost:8501 Network URL: http://10.76.227.68:8501

Or click the link that was provided once you run the streamlit in your command line You can enter your search query in the text box given and you will get your top three Sessions of interest.

Summit event 2022

2022 Analytics Summit will be held on Wednesday Sep 28, 2022 and Thursday Sep 29, 2022. See you there!

Link to the Recording

• Recoding on conference helper

Resource Links

• Sentence Transformer
• Semantic Search S-BERT
• Huggingface