Predicting product advertisement links using hybrid learning within social networks

• Published in: The Journal of supercomputing Vol. 79; no. 13; pp. 15023 - 15050
• Main Authors: Ebadi Jokandan, Seyed Mohsen, Bayat, Peyman, Farrokhbakht Foumani, Mehdi
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.09.2023; Springer Nature B.V
• Subjects: Advertisements; Advertising; Algorithms; Compilers; Computer Science; Context; Datasets; Interpreters; Matrix methods; Processor Architectures; Programming Languages; Recommender systems; Social networks; System effectiveness; SOM-SPM; Product’s advertisements; Recommender systems; Contextual information
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-023-05213-3

Advertising has been one of the most effective and valuable marketing tools for many years. Since companies spend vast amounts of money on product advertising, they put their best effort into making the advertising effective, consequently increasing product sales. Advertising recommender systems can significantly help make posted ads more effective. Due to the incredible popularity of social networks in the last two decades and the massive amount of user's activity on these networks, a great opportunity has been provided to elicit user data and post product promotions. Recommender systems for social networks can use various data to make recommendations, including content, post features, individuals, and even context-based information (e.g., activity, location, time, individual, and relational data). One of the upcoming challenges is the study of feature interactions and their valuation. The critical question is: “Can this data, especially contextual data, lead to improved advertising recommendations?” Considering the instantaneous growth of social network data, it is also imperative to verify scalability based on time criteria. Since earlier studies have investigated the influence of a limited number of contextual features on the recommender system, it is necessary to conduct a study that comprehensively utilizes the diverse contextual data available in social networks. Therefore, we use a hybrid context-based SOM-SPM approach to meet the challenges raised. Our method improves the accuracy of offered product’s advertisements for target users. The proposed method improves the performance of the advertising recommender system by constituting a post-context feature matrix based on context dimensions and SPM algorithms to learn behavioral patterns. We experiment with this method on two datasets and evaluate the results with MSE, RMSE, MAE metrics and Davies-Bouldin Index. The results indicate that our method outperforms algorithms like FM, SBS, AFM and MF-LOD. The experiment results also indicated the time scalability of the proposed approach. We also analyzed the influence of various context combinations on recommendation accuracy and then calculated each context-to-context feature interaction to better understand each context's ad recommendation efficiency. For example, the feature interaction matrix analysis in the Instagram dataset revealed that features related to the broadcast time of posts and users are more critical than other contextual features. Therefore, companies can plan more accurately on the time and user's behavioral attributes to achieve advertising effectiveness and minimize spam ads.