چكيده لاتين
In todayʹs world, users on social networks are faced with an enormous volume of content, and choosing among them can lead to confusion and reduced quality of user experience. Recommender systems are recognized as an effective solution to address the information overload problem in social networks. These systems provide personalized content by analyzing usersʹ behavioral patterns and preferences. Research shows that long-term psychological factors such as personality traits play an important role in individualsʹ decision-making and choices, while most current recommendation systems ignore these psychological mechanisms. One of the topics of interest in psychology and behavioral sciences is Glasserʹs "Choice Theory". Glasserʹs Choice Theory, presented by social psychologist William Glasser, states that individuals evaluate options based on their personal needs, desires, and values in the decision-making and choice process. According to this theory, every individual strives to satisfy their five basic needs (including survival, power, love and belonging, freedom, and fun) and makes decisions based on these needs. These needs exist in individuals with varying intensities and remain relatively constant throughout life. Considering the commonality between Glasserʹs Choice Theory and recommendation systems (which are based on identifying and predicting individualsʹ choices), it appears that if the basic needs of each user are identified according to Glasserʹs "Choice Theory" and incorporated alongside data and other factors in providing recommendations to them, it will lead to improved accuracy of recommendation systems and better recommendations.
The objective of this research is to design a content recommendation system for social networks using usersʹ past interests and the intensity of their basic needs based on Glasserʹs "Choice Theory". The proposed method is implemented in two phases on the Instagram social network platform. The first phase includes providing a machine learning model that predicts the intensity of each basic need based on usersʹ profile information. For this purpose, features from the personal pages of 127 users along with the intensity of their basic needs, determined through responses to Glasserʹs needs questionnaire, were collected. Then, for each basic need, a dedicated machine learning model was trained using four algorithms (Logistic Regression, Random Forest, Gradient Boosting, and SVM) in the form of multi-label classification. The second phase involves designing a recommendation system to suggest content to users, which employs two approaches: the first approach provides personalized content using collaborative filtering algorithms based on similarity in the intensity of basic needs and past interests among individuals, and the second approach uses the relationship between individuals dominant basic needs and liked content for recommendations.
To evaluate the performance of the proposed recommendation system, offline evaluation was first conducted, resulting in the best performance with MAE of 0.1444 and RMSE of 0.2786. Then, to assess the systemʹs performance in real conditions, online evaluation was conducted on the Instagram platform. The online evaluation results showed that the hybrid recommendation system could demonstrate significant performance compared to the conventional recommendation system, with content acceptance rate improving from 28.5% to 42.7%.
