MUSEO: A Recommendation System for Museum Selection
Why Do this Project?
As someone who enjoys visiting museums, I found it time consuming to locate the museums of interest. Thus, I created an app, MUSEO, to:
- Recommend museums based on users’ preference
- Allow users to filter the museums that meet their needs
Data Collecting and Processing
Data from 1600 museums was scraped from TripAdvisor. The graph below highlights the information I collected for each museum.
Next, I converted some categorical features, such as “Recommended length of visit” (<1 hr, 1-2hr, 2-3hr, >3hr), “Fee”(Yes, No), “Country” (US, non-US), into dummy variables. Then I followed the diagram below to generate new features for museum categories, review tags, and textural data such as user review, user quote, and museum description. Missing values were handled by KNN imputation.
Construction of Recommendation System
The algorithm for my recommendation system:
- Read the user input (selected museums)
- Compute and sort cosine similarity across 1600 museums using 219 features (“Review Count” and “Total Things to Do” are removed to avoid bias)
- Remove duplicate recommendations
- Return museums with sorted descending similarity score
Limitation and Improvement
Since there were only 1600 museums in my master data frame on the back-end, the app has some limitations. A potential improvement would be to allow users to select the features that are most important for them and assign weights to those features to further customize the recommendation.
Future Work
- Apply association rule to review tags to see if some tags such as “worth a visit” and “van gogh” would be grouped together
- Predict whether a museum will be featured or not
- Conduct cluster analysis and add the cluster as a new feature for the recommendation system
- Allow users to filter the museums by rating
Impact
- Lure people to explore the museums around the globe and boost the tourism economy
- Save time for people to locate the museums that match with their travel plans by specifying the length of visit and interests (e.g., good for family/kids…etc)
I hope you find this app useful. It’s time to explore a new museum that is right for you! 🙂
Technical Development
See the full code here.
Python:
- Web Scraping
- Adding New Features
- Sentiment Analysis
- Cosine Similarity
R (Shiny App server.R):
- Recommendation tab
- Map tab