Alex Keaveney
Creative Computing / Year 4

alt_text
Alex Keaveney

Alex Keaveney

Creative Computing

Year 4

  • Project Title PlaceSwipe — A Hybrid Location Recommender iOS Application
  • Course BSc [Hons] Creative Computing
  • Year 4
  • Contact Info ak12349@hotmail.com

PlaceSwipe — A Hybrid Location Recommender iOS Application

The aim of the project is to investigate the effectiveness of Hybrid Filtering on recommending locations for people to visit. This investigation is carried out by creating an iOS mobile application that would get recommendations for users using different Hybrid Algorithms. The performance of each can then be evaluated using mathematical and quantitative feedback from users.

Project Description

This project conducts detailed research into the area of Recommendation Systems (RS). RS are software that attempt to provide personalised content and services to users based on that user’s past preferences. The effectiveness of RS are becoming increasingly important due to the overload of information available which needs to be filtered into what a user wants at any given time. Examples of RS include Youtube, for videos, Spotify, for music and Amazon, for products. This project focuses on using Hybrid Filtering to recommend locations to users. This is carried out by developing an iOS and Python Web Application that gives users Location Recommendations. The LightFM library was used to implement the HF system. It was accompanied by the Flask Web Framework to connect the backend to a mobile front end, written in Swift. The front-end allowed the users to like and dislike recommendations with a Tinder-style interface. The user data to allow CF to work was retrieved from Foursquare.com and stored in a MongoDB.

Project Findings

The result of this project was an iOS application that communicates with a Python Web Server and gets recommendations for users based on their previous history i.e., their past likes and dislikes and by filters for maximum distance and type of place e.g., drinks and outdoors. Users can save locations to their favourites. They can also get more information about locations such as descriptions, reviews and directions. One finding was that using the same UI as Tinder-swipe gestures, to interact with recommendations, appeals to users and makes the application more engaging. This project confirmed that Hybrid Filtering is an effective method for recommending locations. It proved that Weighted Approximately Ranked Pairwise loss (WARP) outperforms Bayesian Personalised Ranking (BPR) for classifying location data.

Alex Keaveney
Creative Computing / Year 4