Naslov
Deep learning-based recommendation system in tourism by personality type using social networks big data

Autori
Ambrušec , Martina

Vrsta, podvrsta i kategorija rada
Ocjenski radovi, diplomski rad, diplomski

Mjesto
Dubrovnik / Zagreb

Datum
06.05

Godina
2022

Stranica
34

Mentor
Tolić, Domagoj

Ključne riječi
recommendation system ; deep learning ; big data ; tourism

Sažetak
Recommendation systems are present in many daily activities. They are trying to predict user preferences. Due to the growth of social networks, there is a vast amount of data constantly updated which makes recommendation systems more personalized and efficient. This study aims to apply natural language processing (NLP) and deep learning techniques to obtain a recommendation. NLP is used to analyze the text (i.e., hashtags) from social networks to determine similarity between different points of interest (POI). A pre-trained convolutional neural network (CNN) is used to classify a set of images obtained from social networks to determine which POI is visited by which personality type. The personality type is determined using the Five-Factor (that is, Big Five) model. The Big Five traits are firstly converted into ten personality class labels (High Openness, Low Openness, High Conscientiousness, Low Conscientiousness, High Extraversion, Low Extraversion, High Agreeableness, Low Agreeableness, High Neuroticism, Low Neuroticism) for the classification network. A pre-trained CNN is used in a transfer learning manner to automatically extract features from more than 2.000 manually labeled images and classify them. It is demonstrated that personality traits can be extracted from posted images with an accuracy of 83%. Also, it is shown that those traits can be aggregated for a given set of pictures, such that a representation of a POI can be determined.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA