Siren

Abstract

The growing volume of digital data stimulates the adoption of recommender systems in di erent socioeconomic domains, including ecommerce, music, and news industries. While news recommenders help consumers deal with information overload and increase their engagement and satisfaction, their use also raises an increasing number of societal concerns, such as "Matthew e ects", " lter bubbles", and the overall lack of transparency. We argue that focusing on transparency for content-providers is an under-explored avenue. As such, we designed a simulation framework called SIREN 1 (SImulating Recommender E ects in online News environments), that allows news content providers to (i) select and parameterize di erent recommenders and (ii) analyze and visualize their e ects with respect to two diversity metrics with little costs. Taking the U.S. news media as a case study, we present an analysis on the recommender e ects with respect to long-tail novelty and unexpectedness using SIREN. Our analysis o ers a number of interesting ndings, such as the similar potential of certain algorithmically simple (item-based k-Nearest Neighbour) and sophisticated strategies (based on Bayesian Personalized Ranking) to increase diversity over time. Overall, we argue that simulating the e ects of recommender systems can help content providers to make more informed decisions when choosing algorithmic recommenders, and as such can help mitigate the aforementioned societal concerns.

Measuring the benefit of increased transparency and control in news recommendation

“…Other work has not focused on user profiles specifically but used context to explain the credibility of news articles—by comparing them to other similar articles (Bountouridis et al. 2018). This visualization demonstrates which text is similar to other sources, and which text is unique to the article and source currently viewed by the user (see Figure 3).…”

“…Overview of the interactive information layer design proposed in Bountouridis et al. (2018) which uses context to explain the credibility of news articles: corroborated (blue underline) and omitted information (orange underline). Here, hovering over an underlined piece of information reveals information from other news sources omitted in the original story.…”

“…Overview of the interactive information layer design proposed inBountouridis et al (2018) which uses context to explain the credibility of news articles: corroborated (blue underline) and omitted information (orange underline). Here, hovering over an underlined piece of information reveals information from other news sources omitted in the original story.F I G U R E 4Overlap WordCloud Inel, Tintarev, and Aroyo (2020) used as a contextual explanation of video summaries: The words in green are concepts found in the video summary.…”

Enhancing User Profile Accuracy for Personalized Web Systems

“…This research relates to some studies in the field personalized web systems in particular. Personalized web systems are developed to help users; browse news articles (NPA [5], SIREN [10], UP-TreeRec [27]), find scientific and research papers (Pique [14], Personalized reading system [28]), purchase favorite products (Amazon [11], eBay [17]), improve search results (Hide-n-Seek [23], Persona [25]) or even combine some of the previous tasks (Basar [26], Syskill and Webert [19]).…”

Modeling and counteracting exposure bias in recommender systems.

“…Some work focused on the iterated bias of Recommender Systems. Bountouridis et al [3] designed a simulation framework to see the effect of the recommendation models on the diversity and novelty of the Recommendations. They used news data and they compared different state-of-the-art models.…”