import streamlit as st import pandas as pd import altair as alt from PIL import Image #!pip install -U sentence-transformers from sentence_transformers import SentenceTransformer, util from sentence_transformers import SentenceTransformer from transformers import pipeline import numpy as np import sys import json #from torch.utils.data import DataLoader from sentence_transformers import SentenceTransformer, LoggingHandler, util, models, evaluation, losses, InputExample from torch.utils.data import IterableDataset import tqdm #from torch.utils.data import Dataset import random from shutil import copyfile from urllib.error import URLError ## TEXT ###################################################################################################################### # Load document embeddings # set up title and sidebar st.title (" Nashville Analytics Summit Conference Helper")# (" Your top 3 Important Sessions") st.markdown("**Problem** 🧐:") st.markdown("Since its inception in 2013, Nashville Analytics Summit has seen a growth of over 488%. The Summit prides its itself as the fastest growing locally grown tech events in the south region. With an increasing number of participants and dozens of talks covering a myriad of topics, there is a need to tailor participants needs to their interests") st.markdown("---") st.markdown("**Solution**πŸ’‘:" ) st.markdown("Develop an application in which users can input the description of areas of interest and app returns the top three Sessions matching the description requested.") st.markdown("---") st.markdown("** Approach** πŸ—οΈ :") st.markdown("* For the approach, I used a transformer model, multi-qa-MiniLM-L6-cos-v1, that uses sentence similarity to match the description of the event and the input description.") st.markdown("* The dataset used is Nashville Analytics Summit descriptions of the presentations, which include the Unique ID, Name of presenter, Description of presentation, Activity Code, Start Time, End Time, Location Name") st.markdown("---") #st.markdown("### The Details") # section 1: the context, problem; how to address st.markdown("### Problem") st.markdown("With the growth of the Nashville Analytics Summit participation every year, there have been an increase in the number participants, the number of talks/topics have exponentially increased over time covering a myriad of current issues and developments, analytics, data science AI, career growth, new tools and much more. This has made it challenging for the participants to easily navigate through the events and attend the sessions of interest before reading through the sessions descriptions. Struggling finding the Session that Matches your personal interests? Transformers to the rescue!!! πŸ¦Έβ€β™‚οΈ .") dificult_reading = Image.open('dificult_reading.png') st.image(dificult_reading, caption='Reading sessions descriptions.') st.markdown("---") # section 2: how can transformers help? st.markdown("### How can Transformers Help?πŸͺ„ ") st.markdown("* Semantic search applies the user intent, context, and conceptual meanings to match a user query to the corresponding content. It uses vector search to returns results that aim to match a users query. The components work together to retrieve and rank the results based on the meaning.") st.markdown("**Sentence Similarity**") st.markdown("* In this a similar project we used a dataset from the Nashville Analytics Summit") st.markdown("* Given a search phrase, the model search for an abstract that marches the search phrase using cosine similarity. ") st.markdown("---") # section 4: The process st.markdown("### The Process πŸ”") st.markdown("A participant inputs a search phrase of the sessions he/she would like to attend and the semantic serach would return the title of the session, the abstract of the session, the room number and the time the session will be offered.") st.markdown("The output of the search query \ - Once the user input their phrase or keywords to search sessions of interest, the appropriate sessions are selected by finding the top 3 sessions with the highest similarity scores to the phrase or keyword.") #st.markdown("3. Send (claim, evidence) pairs to a transformer model. Have the model predict whether each evidence supports, refutes, or is not relevant to the claim. (πŸ“ YOU ARE HERE!)") #st.markdown("4. Report back to the user: The supporting evidence for the claim (if any), the refuting evidence for the claim (if any). If no relevant evidence is found, report that the claim cannot be supported or refuted by current evidence.") # section 5: my work #st.markdown("## Climate Claim Fact-Checking with Transformers") #st.markdown("My work focuses on step 3 of the process: Training a transformer model to accurately categorize (claim, evidence) as:") #st.markdown("* evidence *supports* (entails) claim") #st.markdown("* evidence *refutes* (contradicts) claim") #st.markdown("* evidence *does not provide enough info to support or refute* (neutral) claim") #st.markdown("For this project, I fine-tune ClimateBERT (4) on the text entailment task") st.markdown("---") # section 6: analysis st.markdown("## Critical Analysis") st.markdown("What can be done to improve the accuracy?") st.markdown("* Giving more data to be used to create embeddings, this would improve the model performance.") st.markdown("* A few pretrained Sentence Similarity models could be used and pick the one with higher accuracy and generalize well to new semantics!") st.markdown("* Creating an app that can be multilingual and output the search in users preferred language") st.markdown("---") ## EXAMPLE ################################################################################################################### st.markdown("## Let's try it out!") st.markdown("This application is a dashboard for displaying your top 3 matching Sessions at the Nashville summit") doc_emb = np.loadtxt("abstract-embed.txt", dtype=float) # Load data df = pd.read_csv("sessions.csv", usecols=['Unique ID', 'Name', 'Description', 'Activity Code', 'Start Time', 'End Time', 'Location Name']) # front end elements of the web page html_temp = """

Sentence Similarity App for Nashville Analytic Summit 2022

""" def main(): # display the front end aspect st.markdown(html_temp, unsafe_allow_html = True) # Get attributes from dataframe docs = list(df["Description"]) titles = list(df["Name"]) start_times = list(df["Start Time"]) end_times = list(df["End Time"]) locations = list(df["Location Name"]) # Query # Load the model model = pipeline('text-classification')#('sentence-transformers/multi-qa-MiniLM-L6-cos-v1') query = st.text_input("Enter your query: ") if query: #st.text_area('Text area') #age = st.number_input("Age in Years") #Encode query and documents query_emb = model(query)#.astype(float) #Compute dot score between query and all document embeddings #scores = util.dot_score(query_emb, doc_emb.astype(float))#[0].cpu().tolist() #cos_scores = util.pytorch_cos_sim(query_emb, doc_emb)[0] scores = util.pytorch_cos_sim(query_emb, doc_emb)[0] #Combine docs & scores with other attributes doc_score_pairs = list(zip(docs, scores, titles, start_times, end_times, locations)) # top_k results to return top_k=3 print(" Your top", top_k, "most similar sessions in the Summit:") #Sort the results in decreasing order and get the first top_k doc_score_pairs = sorted(doc_score_pairs, key=lambda x: x[1], reverse=True) #Output presentation recommendations for doc, score, title, start_time, end_time, location in doc_score_pairs[:top_k]: st.write("Score: %f" %score) st.write("Title: %s" %title) st.write("Abstract: %s" %doc) st.write("Location: %s" %location) st.write(f"From {start_time} to {end_time}") st.write('\n') if __name__ == "__main__": main() st.markdown("---") # References + Resource Links st.markdown("## Resource Links") st.markdown("### References") st.markdown("1. https://www.sbert.net/examples/applications/semantic-search/README.html") st.markdown("2. https://www.sbert.net/docs/pretrained-models/msmarco-v3.html") st.markdown("3. Semantic search [colab](https://colab.research.google.com/drive/12cn5Oo0v3HfQQ8Tv6-ukgxXSmT3zl35A?usp=sharing)") st.markdown("4. [project code on github](https://github.com/vanderbilt-data-science/abstract-search)")