Browsing by Subject "Travel Attractions Recommendation,"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
Item POI Recommendations with the Use of Knowledge Graph Neural Networks(Pulchowk Campus, 2021-08) Paudyal, NabinRecommendation of the most relevant travel attractions for travellers when they are visiting a new place is a very important problem. The ability to recommend the most relevant tourist attractions that would be of most interest to the visitors can help increase the pro ts for the business and also provide a better traveler experience on the visitor's part. Some of the most common techniques used for recommending travel attractions are collaborative ltering based ones which rely on the information such as places visited by groups of other travelers that share maximum similarity to the visitor. For the purpose of this thesis work, the goal was to devise a method to generate the most relevant travel attraction recommendations for users drawing upon other models that produced excellent results in recommendation for other items like movies and books in the past. In that regard, Knowledge Graphs were combined with Graph Neural Networks in the ensuing KGNN. Knowledge Graphs represent the relationship between entities in the form of a graph while Graph Neural Networks are the form of Neural Networks that apply over data structured as graphs. The experiments were carried out using the three di erent KGNN approaches - KGCN, KGNN-LS and KGAT, on the publicly available location based social network check-in datasets for Foursquare and Gowalla. The performance improvement of 24.19%, 14.51% and 22.58% were achieved respectively for the three methods for top 5 recommendations on Foursquare in terms of F1-score when compared to the best performing baseline of Rank-GeoFM. Similarly, the improvements were 13.20%, 14.86% and 28.30% and 18.27%, 25.58% and 30.23% respectively for top 10 and 20 recommendations for Foursquare. The performance improvement on Gowalla meanwhile was of 19.35%, 29.03% and 43.54% respectively for top 5 recommendations, 10.29%, 129.41% and 29.41% for top 10 recommendations and 6.77%, 15.25% and 25.42% respectively for top 20 recommendations.