app.py

import streamlit as st
from streamlit_tags import st_tags, st_tags_sidebar
from keytotext import pipeline
from PIL import Image

import json
from sentence_transformers import SentenceTransformer, CrossEncoder, util
import gzip
import os
import torch
import pickle
import yake

############
## Main page
############

st.write("# Demonstration for Etsy Query Expansion（Etsy-QE）")

st.markdown("***Idea is to build a model which will take query as inputs and generate expansion information as outputs.***")
image = Image.open('top.png')
st.image(image)

st.sidebar.write("# Top-N Selection")
maxtags_sidebar = st.sidebar.slider('Number of query allowed?', 1, 20, 1, key='ehikwegrjifbwreuk')
#user_query = st_tags(
#    label='# Enter Query:',
#    text='Press enter to add more',
#    value=['Mother'],
#    suggestions=['gift', 'nike', 'wool'],
#    maxtags=maxtags_sidebar,
#    key="aljnf")

user_query = st.text_input("Enter a query for the generated text: e.g., gift, home decoration ...")

# Add selectbox in streamlit
option1 = st.sidebar.selectbox(
     'Which transformers model would you like to be selected?',
     ('multi-qa-MiniLM-L6-cos-v1','louis030195/multi-qa-MiniLM-L6-cos-v1-de-ecommerce','null'))

option2 = st.sidebar.selectbox(
     'Which corss-encoder model would you like to be selected?',
     ('cross-encoder/ms-marco-MiniLM-L-6-v2','null','null'))

st.sidebar.success("Load Successfully!")

#if not torch.cuda.is_available():
#    print("Warning: No GPU found. Please add GPU to your notebook")

#We use the Bi-Encoder to encode all passages, so that we can use it with sematic search
bi_encoder = SentenceTransformer(option1,device='cpu')
bi_encoder.max_seq_length = 256    #Truncate long passages to 256 tokens
top_k = 32                          #Number of passages we want to retrieve with the bi-encoder

#The bi-encoder will retrieve 100 documents. We use a cross-encoder, to re-rank the results list to improve the quality
cross_encoder = CrossEncoder(option2, device='cpu')

passages = []

# load pre-train embeedings files
embedding_cache_path = 'etsy-embeddings-cpu.pkl'
print("Load pre-computed embeddings from disc")
with open(embedding_cache_path, "rb") as fIn:
  cache_data = pickle.load(fIn)
  passages = cache_data['sentences']
  corpus_embeddings = cache_data['embeddings']

kw_extractor = yake.KeywordExtractor()
language = "en"
max_ngram_size = 3
deduplication_threshold = 0.9
numOfKeywords = 20
custom_kw_extractor=yake.KeywordExtractor(lan=language, n=max_ngram_size, dedupLim=deduplication_threshold, top=numOfKeywords, features=None)

# This function will search all wikipedia articles for passages that
# answer the query
def search(query):
    st.write("Input question:", query)
    ##### Sematic Search #####
    # Encode the query using the bi-encoder and find potentially relevant passages
    query_embedding = bi_encoder.encode(query, convert_to_tensor=True)
    hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=top_k)
    hits = hits[0]  # Get the hits for the first query

    ##### Re-Ranking #####
    # Now, score all retrieved passages with the cross_encoder
    cross_inp = [[query, passages[hit['corpus_id']]] for hit in hits]
    cross_scores = cross_encoder.predict(cross_inp)

    # Sort results by the cross-encoder scores
    for idx in range(len(cross_scores)):
        hits[idx]['cross-score'] = cross_scores[idx]

    # Output of top-N hits from bi-encoder
    #st.write("\n-------------------------\n")
    #st.subheader("Top-N Bi-Encoder Retrieval hits")
    #hits = sorted(hits, key=lambda x: x['score'], reverse=True)
    #for hit in hits[0:maxtags_sidebar]:
    #    st.write("\t{:.3f}\t{}".format(hit['score'], passages[hit['corpus_id']].replace("\n", " ")))

    # Output of top-N hits from re-ranker
    st.write("\n-------------------------\n")
    st.subheader("Top-N Cross-Encoder Re-ranker hits")
    hits = sorted(hits, key=lambda x: x['cross-score'], reverse=True)
    #for hit in hits[0:maxtags_sidebar]:
    #    st.write("\t{:.3f}\t{}".format(hit['cross-score'], passages[hit['corpus_id']].replace("\n", " ")))
    hit_res = []
    for hit in hits[0:1000]:
      q = passages[hit['corpus_id']].replace("\n", " ")
    if q not in hit_res:
        hit_res.append(q)
    for res in hit_res[0:maxtags_sidebar]:
        keywords = custom_kw_extractor.extract_keywords(res)
        for kw in keywords:
            st.write(kw)

st.write("## Results:")
if st.button('Generated Expansion'):
    out = search(query = user_query)
    #st.success(out)