Article Preview
Top1. Introduction
The rapid growth of the Internet has not only contributed to making life easier but has also generated an enormous volume of information, leading to the issue of information overload. It has become challenging for individuals to select their desired items that align with their preferences from the vast array of choices available. To address this, recommender systems (RSs) play a crucial role in alleviating information overload and enhancing user experiences. RSs are intelligent tools and methodologies designed to propose items that are likely to capture the interest of a particular user. These recommendations frequently play a role in various decision-making processes, such as helping users decide which products to purchase, which songs to listen to, and which articles to read. The term “item” commonly refers to the suggestions provided by the system. The system typically presents personalized recommendations in the form of ranked lists. These recommendations are generated by predicting the most suitable items or services for an individual based on their preferences (Ricci et al., 2022).
Three main types of RSs are commonly employed: content-based filtering, collaborative filtering (CF), and hybrid approaches (Da’u & Salim, 2020). These methods rely on the fundamental principle that individuals with similar past preferences are likely to make comparable future choices. CF recommenders specifically utilize data about user-item interactions to formulate their suggestions. Additionally, there exists a subset of RSs that incorporate contextual information related to users and/or items to enhance their recommendations. Among these types, content-based recommenders stand out as a distinct category within RSs. Moreover, certain hybrid RSs amalgamate insights from both user and item perspectives to provide comprehensive recommendations (Batmaz et al., 2019).
Despite the effectiveness of CF methods in various scenarios, the issue of data sparsity remains a significant hurdle (Da’u & Salim, 2020) (Batmaz et al., 2019). This challenge, stemming from an insufficient number of user ratings for each item, often hampers the performance of conventional approaches. Another notable concern is the cold start problem (Shokeen & Rana, 2020a), which significantly impacts both new users and new items. Providing meaningful recommendations without access to relevant historical data proves to be challenging. While novel users can receive suggestions for popular items, this approach falls short of delivering truly personalized recommendations.
Addressing these challenges necessitates the incorporation of supplementary data such as user profiles, product descriptions, and information from social platforms (Sun et al., 2019). Several research projects have explored the collection of additional data, and one approach for tackling data scarcity involves extracting insights from review texts (L. Zheng et al., 2017). In instances where adequate historical user-item interaction data is lacking, leveraging contextual information becomes pivotal in making informed inferences about user preferences. Individuals' preferences can be aggregated through social networks, allowing for the deduction of their tastes. Further data sources, like item tags or categories, can be directly employed to comprehend user preferences. For instance, a user's favorite movie genres or music album categories can shed light on the types of content they prefer (Sun et al., 2019). Contextual factors, such as timing, location, or tags, possess the capacity to influence an individual's perception of a particular item (Chen & Li, 2019). Tag-related data, on the other hand, supplements existing datasets by elucidating item attributes and capturing user preferences through tagging behaviors (J. Zhang et al., 2019b). Consequently, integrating supplementary information into the recommendation process emerges as a viable solution to tackle the challenges posed by data sparsity and the cold start problem in RSs.