Travisor

Inspiration

When planning and researching for trip itinerary can take a lot of time. By implementing automation and personalized recommendations to users, it can make their experience more convenient and inspire them with new destinations they have never discovered

What it does

Our personalized recommender automatically suggests user new contents of travel destinations based on their interests and similarity of their historical destination travelled. We are optimizing user’s experiences in finding the best flights, accommodations, or activities for their upcoming trip by leveraging the AI technology

How we built it

We trained a data set using Microsoft Azure Machine Learning studio (AML) to run a recommendation algorithm. The recommendation model we explored is the content-based filtering and collaborative filtering which is available to experiment on AML through a module called "Matchbox Recommender". It is a hybrid recommender combing both type of approaches in recommendation system. We select the relevant data which is the user ID, place ID and the rating, then we split the data using "Split Data" control into testing and training set. We ran the experiment model with "Evaluate Recommender" control and receive about 95% of accuracy.

Challenges we ran into

Acquiring the data set: we were unable to find specific data of travel destinations or user data, so we had to implement the experiment using available data of restaurants.
Error in Microsoft Azure: there were some errors that occurred throughout the process, in the part where we arrange the data set. However, we were able to solve it by reading the guidelines and recheck the data selected.

Accomplishments that we're proud of

Completed an experiment on AML and obtained a high accuracy of 95% from the training set.

What we learned

We learned about the recommendation system in machine learning and the types of recommendation system that can be implemented with our website.
We learning how to use AML program which we never used before
During the experiment, we learned that features of the data set should be available to compute the similarity with the users data.

What's next for Travisor

We can build another kind of recommendation system such as a chatbox recommendation system where the users can directly input the data they wanted and let the AI recommend similar places