شناسايي حساب‌هاي چندكاربره بر اساس ويژگي‌هاي شخصيتي كاربران و علاقه‌مندي آن‌ها به ژانر فيلم‌ها

Identifying multi-user accounts based on users personality traits an‎d their interest in movie genres

مهندسي فناوري اطلاعات

Abstract Movie streaming systems provide users with a personalized experience by offering broad access to diverse content an‎d leveraging recommender systems. However, the use of a single account by multiple users, in addition to causing financial losses for service providers, reduces the accuracy an‎d effectiveness of recommendations. Identifying such accounts is challenging because of the need to distinguish between the natural diversity of a single user’s interests an‎d the heterogeneity resulting from the presence of multiple users; variability in behavioral patterns an‎d preferences does not necessarily imply that an account is shared. Therefore, comparing an account’s preferences with patterns independent of its behavior an‎d assessing the degree of inconsistency can be effective in identifying shared accounts. Furthermore, after identifying such accounts, determining the primary user an‎d reconstructing their preferences is important. This research proposes a method based on users’ personality traits that pursues two main objectives: identifying shared accounts based on the relationship between personality traits an‎d genre preferences, an‎d identifying the primary user an‎d more accurately reconstructing their interests to improve recommendation personalization. The main idea is based on the assumption that, in single-user accounts, movie ratings are consistent with the user’s personality traits, an‎d significant inconsistency in these relationships may indicate multi-user use. To this end, machine learning algorithms are first used to extract the relationships between personality traits an‎d users’ preferences across 18 major movie genres. The recorded ratings of each account are then compared with the predicted patterns to identify shared accounts. Subsequently, for these accounts, a deep learning model receives genre ratings an‎d the primary user’s personality traits as conditional variables, reconstructs the portion that is more consistent with the primary user, an‎d extracts their genre preferences. Finally, the account data are filtered based on the primary user’s preferences an‎d provided to the recommender system. eva‎luations are conducted on the MovieLens dataset, which includes movie ratings an‎d users’ personality traits, while shared accounts are simulated by ran‎domly combining single-user accounts. The results show that the proposed method achieves an accuracy of 0.87 an‎d an F1-score of 0.79 in identifying shared accounts. In addition, for identifying the primary user’s preferred genres, accuracy an‎d F1-score of 0.72 an‎d 0.82, respectively, are obtained. eva‎luation of the recommender system also shows that, after data filtering, NDCG increases from 0.36 to 0.38 an‎d HR increases from 0.63 to 0.69. These results demonstrate the role of personality traits in identifying shared accounts, reconstructing the primary user’s preferences, an‎d improving recommendation performance.

6

171648