import streamlit as st import pandas as pd import os import re import preprocessor as p import joblib import base64 project_description = """ # Hotel Data Analysis Project ## Overview I have completed a hotel data analysis project using an instant web scraper. This project involved scraping hotel data and hotel reviews separately, cleaning the data, concatenating it, and performing sentiment analysis on the DataFrame. Additionally, I clustered the hotel reviews, applied sentiment analysis, and passed those clusters to an LLM (Language Model) to extract strengths and weaknesses of hotels. ## Steps ### 1. Scraping Hotel Data - Utilized an instant web scraper to collect hotel data. - Scraped hotel data separately from hotel reviews. ### 2. Data Collection - Collected hotel data and hotel reviews data separately for each hotel. ### 3. Data Cleaning - Cleaned the collected data to remove any inconsistencies or errors. - Applied preprocessing techniques to prepare the data for analysis. ### 4. Data Concatenation - Concatenated the cleaned hotel data and hotel reviews data to create a unified dataset for analysis. ### 5. Sentiment Analysis - Performed sentiment analysis on the concatenated DataFrame. - Utilized the results to understand the overall sentiment of hotel reviews. ### 6. Clustering Hotel Reviews - Clustered the hotel reviews based on their content to identify patterns and similarities. ### 7. Extracting Strengths and Weaknesses - Passed the clustered reviews to an LLM (Language Model) to extract strengths and weaknesses of hotels. - Used the extracted information to gain insights into customer perceptions. ## Conclusion This project demonstrates the use of web scraping, data cleaning, sentiment analysis, and clustering techniques to analyze hotel data. The extracted strengths and weaknesses provide valuable insights for hotel management to improve customer satisfaction and service quality. """ def create_download_link(df, filename): csv = df.to_csv(index=False) b64 = base64.b64encode(csv.encode()).decode() href = f'Download CSV file' return href # Path to the directory containing CSV files directory_path = r'hotel reviews' # Get a list of CSV files in the directory csv_files = [file for file in os.listdir(directory_path) if file.endswith('.csv')] # Function to concatenate selected columns def concatenate_columns(df, selected_columns): concatenated_data = df[selected_columns[0]].tolist() + df[selected_columns[1]].tolist() return pd.DataFrame({'ConcatenatedData': concatenated_data}) # Function to display selected dataset def display_selected_dataset(selected_dataset): dataset_path = os.path.join(directory_path, selected_dataset) selected_df = pd.read_csv(dataset_path) st.subheader(f'Dataset: {selected_dataset}') st.write(selected_df) def clean_tweets(series): REPLACE_NO_SPACE = re.compile("[.;:!\'?,\"()\[\]]") REPLACE_WITH_SPACE = re.compile("()|(\-)|(\/)") tempArr = [] for line in series: # Check if the value is NaN if pd.isnull(line): tempArr.append("") continue # Send to tweet_processor tmpL = p.clean(line) # Remove punctuation tmpL = REPLACE_NO_SPACE.sub("", tmpL.lower()) # Replace specific characters with spaces tmpL = REPLACE_WITH_SPACE.sub(" ", tmpL) # Remove extra spaces tmpL = " ".join(tmpL.split()) tempArr.append(tmpL) return tempArr # Streamlit app def main(): # Create a menu bar menu = st.sidebar.selectbox( 'Navigation', ['Home', 'collected hotel data', 'Display Hotel Data', 'Display hotel reviews Datasets', 'CSV Column Concatenation and Sentiment Analysis'] ) if menu == 'Home': st.markdown(project_description) elif menu == 'collected hotel data': # Display DataFrame df = pd.read_csv('chennai hotes.csv') df1 = pd.read_csv('stream.csv') st.subheader('Collected chennai hotes Data') st.write(df) st.subheader('preprocess applyed data') st.write(df1) elif menu == 'Display Hotel Data': # Display hotel data df = pd.read_csv('stream.csv') css = """ """ st.markdown(css, unsafe_allow_html=True) for index, row in df.iterrows(): st.markdown(f"""

{row['Hotel Name']}

Rating: {row['rating']}

Location: {row['location']} ({row['nearest places']})

Website: Website link

Number of Reviews: {row['number of reviewss 2']}

Room Type: {row['room type']}

Price: {row['price']}

Strengths: {row['Strengths']}

Weaknesses: {row['Weaknesses']}

""", unsafe_allow_html=True) elif menu == 'Display hotel reviews Datasets': selected_dataset = st.selectbox('Select Dataset', csv_files) if selected_dataset: display_selected_dataset(selected_dataset) elif menu == 'CSV Column Concatenation and Sentiment Analysis': st.title('CSV Column Concatenation and Sentiment Analysis') new_names = { 'a3332d346a': 'Reviewer Name', 'afac1f68d9': 'Reviewer Country', 'abf093bdfe': 'Room Type', 'abf093bdfe 2': 'Length of Stay', 'abf093bdfe 3': 'Review Date', 'abf093bdfe 4': 'Traveler Type', 'abf093bdfe 5': 'Second Review Date', 'f6431b446c': 'Overall Rating', 'a53cbfa6de': 'Positive Comments', 'a53cbfa6de 2': 'Negative Comments', 'a3332d346a 2': 'Hotel Response', 'a53cbfa6de 3': 'Hotel Response1' } # File upload uploaded_file = st.file_uploader('Upload CSV file', type=['csv']) if uploaded_file is not None: df = pd.read_csv(uploaded_file) df.rename(columns=new_names, inplace=True) # Show original DataFrame st.subheader('Original DataFrame:') st.write(df) # Select columns selected_columns = st.multiselect('Select columns to concatenate', df.columns) if st.button('Concatenate columns'): if len(selected_columns) == 2: # Concatenate columns new_df = concatenate_columns(df, selected_columns) # Remove null values new_df = new_df.dropna() # Drop duplicates new_df = new_df.drop_duplicates() # Reset the index new_df = new_df.reset_index(drop=True) # Clean tweets new_df['CleanedData'] = clean_tweets(new_df['ConcatenatedData']) # Load the saved model loaded_model = joblib.load('sentiment_analysis_model.pkl') # Apply sentiment analysis new_df['Sentiment'] = loaded_model.predict(new_df['CleanedData']) # Display concatenated, cleaned, and sentiment analyzed DataFrame st.subheader('Concatenated, Cleaned, and Sentiment Analyzed DataFrame:') st.write(new_df) # Create download link st.markdown(create_download_link(new_df, 'concatenated_sentiment_analyzed_data'), unsafe_allow_html=True) else: st.warning('Please select exactly two columns to concatenate.') # Run the app if __name__ == '__main__': main()