Authors:
Sara Jarrad
1
;
Hubert Naacke
1
;
Stephane Gancarski
1
and
Modou Gueye
2
Affiliations:
1
LIP6, Sorbonne University, Paris, France
;
2
Department of Mathematics and Computer Science, Cheikh Anta Diop University, Dakar, Senegal
Keyword(s):
Point of Interest (POI), Next POI Recommendation, Word Embedding, Word2Vec, Similarity Metrics.
Abstract:
Social media platforms allow users to share information, including photos and tags, and connect with their peers. This data can be used for innovative research, such as proposing personalized travel destination recommendations based on user-generated traces. This study aims to demonstrate the value of using embeddings, which are dense real-valued vectors representing each visited location, in generating recommendations for the next Point of Interest (POI) to visit based on the last POI visited. The Word2Vec language model is used to generate these embeddings by considering POIs as words and sequences of POIs as sentences. This model captures contextual information and identifies similar contexts based on the proximity of numerical vectors. Empirical experiments conducted on a real dataset show that embedding-based methods outperform conventional methods in predicting the next POI to visit.