Article Preview
TopA recommendation system is a system that recommends items or information that best meet users' needs by analyzing their historical behavior and interest preferences. The workflow mainly includes data collection, data preprocessing, feature extraction, similarity calculation, and recommendation calculation. The working principle of a recommendation system is to collect users' historical behavior records, collect basic information and needs, extract user features, calculate similarity between users, find similar users, and recommend items or information that best meet their interests and needs to improve the accuracy and satisfaction of recommendations. A recommendation system that includes music recommendations has practical value in saving users’ time in searching for more accurate products and information. Therefore, recommendation systems have also become a key research object for domestic and foreign experts and practitioners.
Some researchers have found that the problems or characteristics inherent in the data used to train recommendation systems can affect the recommendation results of the system and have designed improved recommendation systems that can solve corresponding problems. Rabiu et al. (2022) found that data sparsity and category imbalance in recommendation systems have a significant impact on recommendation results. Therefore, a recommendation system based on adaptive short-and long-term memory neural network algorithm and collaborative filtering algorithm is designed. The test indicates that the accuracy of the recommendation system on the TOP5 recommendation results of the dataset has been improved by 12.8% compared to the previous method (Rabiu et al., 2022). Gwadabe et al. (2022) proposed a processing method based on an improved generative graph neural network algorithm to address the issue of poor processing and recommendation performance of traditional recommendation systems for long sequence data. This method is used to preprocess data that requires recommendation calculations, and the test indicates that the recommendation results processed by this algorithm have higher recommendation accuracy than those without processing (Gwadabe et al., 2022).
Zhang et al. (2021) found that implicit recommendations have attracted the attention of many scholars. However, the uncertainty of implicit feedback data poses significant challenges to recommendation. Therefore, the authors propose a causal neural fuzzy inference algorithm to model missing data in implicit recommendations. The test showcases that this algorithm has better recommendation performance on three real datasets, and the recommendation calculation speed is faster (Zhang et al., 2021). Zy et al. (2020) believe that the unknown entries in the rating matrix actually contain a significant amount of useful information for prediction. This information is usually discarded in traditional methods. Therefore, Zy et al. designed an improved recommendation strategy based on the idea of semi supervised learning. The experiment showcases that this method significantly outperforms the reference method in recommendation accuracy and has a certain degree of robustness to the diversity of the dataset (Zy et al., 2020). Zhao et al. (2020) found that traditional recommendation techniques are hindered by the simplicity and sparsity of user project interaction data. Most studies focus on a single type of external relationship, without fully utilizing the potential relationship between users and items. Therefore, they propose a recommendation algorithm that integrates heterogeneous networks and applies it to multiple real recommendation task datasets (Zhao et al., 2020). The test indicates that the recommendation accuracy of this recommendation method is significantly higher than the algorithm before improvement, and it can consider more potential relationship information between users and items (Zhao et al., 2020).