Algorithm curation is the selection of online media by technologies such as recommender systems and personalized search. Curation entails the selective sharing of online content and recommendations based on inferred interests.[1] Curation algorithms leverage this task by implementing different filter approaches such as collaborative filtering and content-based filtering. Examples include search engine and social media products such as the Twitter feed, Facebook’s News Feed, and the Google Personalized Search.[2]
History
editEarly algorithmic curation
editAlgorithmic curation plays an important role in how people connect online today. [3][4]Most information presented today online decides what to save, archive, share, or ignore, which has become a lot of work.[3][4]Therefore, platforms use newsfeed algorithms to decide what to show each user.[3][4] These algorithms are complicated, so it is not easy to know how they shape communication.[3][4]
Information overload
editEarly on, platforms needed a way to filter information so that users wouldn't get overwhelmed.[3] This led to the first-generation ranking algorithms showing the most recent or most popular posts.[3]Second-generation demonstrated algorithms that curate content to keep people hooked onto the platform for a longer timespan.[3]Algorithmic curation also shapes knowledge, attention, and political exposure.[4] In shaping what people view on a day-to-day basis, this gives reasoning to the algorithm acting as a powerful gatekeeper and deciding what new material people are exposed to.[4]
How algorithm changes users' feeds over time
editAlgorithm curation increases source diversity and also reduces the number of external links, which limits access to outside information.[4] Topics based on political content are made more relevant than some COVID-19 health information.[4]Using agent-based modeling, researchers find the problems within these systems.[4] These adversities and motives increase user engagement while misinformation and polarization worsen.[4] The aim is to discover how user behavior, information, and algorithms all influence each other.[4] Therefore, as a response to information overload, algorithmic curation serves as a response to the massive amount of content on social media.[3][4]
Emergence of AI
editWith AI driving systems, Newsfeed now predicts, personalizes, and optimizes information, which are core AI functions.[3][4] These attributes change human perception and behavior by changing what they see, share, and think.[3][4] Now, researchers have adapted to models of computational simulation to understand how AI-driven curation shapes social outcomes at scale.[3] Other platforms like Twitter have moved to simple chronological feeds.[4] They now use complex, AI-powered ranking systems that personalize information. [4]
Algorithmic curation then evolved onto AI-powered systems focused on developing.[3] Now, due to advanced scales and operations, emerging research uses advanced modeling tools to keep up with AI systems that humans cannot understand.[3] Platforms such as Twitter have moved away from a sequential feed.[4] They switched to AI-powered computational systems to compose personalized information.[3] These systems make decisions that aren't realistic for humans to make.[4]
Approach
editFilter types
editCollaborative filtering
editCollaborative filtering (CF) methods create recommendations based on a person’s usage patterns.[5] CF predicts a person’s desire for an item by matching their interests with people who have similar interests.[5] This process allows for the sharing of ratings between like-minded people.[5] CF is based on human and not machine analysis of content.[5] Users of CF systems (found commonly in social media) rate items that they have interacted with; this rating creates a profile of interests.[5] The CF system then matches that user with other people with similar interests.[5] Once matched, the ratings from those similar users are then used to generate recommendations for the user.[5] The main advantage of collaborative filtering includes the ability to filter by various types of content such as text, art, work, music, and mutual funds; it filter[s] based on complex and hard to represent concepts, such as taste and quality.[6]
Content-base filtering
editAnother popular recommendation system implementation is content-based filtering (CBF). In CBF, a user profile is built to provide information about the types of items that the user likes.[6][7] This is based on keywords used to describe the items.[6][7] In this approach, a recommendation is made by presenting similar items to what the user liked in the past (or items that are similar to what the user is looking for).[7] The CBF method creates a profile for each item based on a set of discrete attributes and features.[7] The system then creates a content-based profile for the user based on a weighted vector of item features.[6] This is made from items the user has previously rated or purchased, or from items the user is currently interested in, or presently viewing.[6] The weights represent the importance of each feature to the user.[6] There are various possible ways of computing these weights, such as Bayesian Classifiers, cluster analysis, decision trees, and artificial neural networks.[6] Regardless of the calculation technique, the goal of the weighted vector is the same: to determine the probability that the user will like a suggested item.[6] One example of content-based filtering to help describe this process is Pandora Radio. When a user visits Pandora, they are prompted to enter artist, genre, or composer to create a station.[6] Pandora then uses CBF to find music with similar attributes to the song, artist, or genre that the user provides.[6]
Technology
editRecommender system
editAn algorithm is curated in part by recommender systems that rank and suggest the most relevant content to users. Content is ranked according to a particular user’s implicit and explicit input.[8] Implicit rankings include elapsed time viewing or engaging with a specific item.[8] The user’s liked items, such as media posts, store pages, news articles and shared items make up the explicit data used by algorithms to recommend content.[8]
Personalized search
editBy utilizing external factors outside the user’s explicit query, personalized search aims to retrieve results most relevant to the user. The user’s past queries, history, and interests create an additional context that refines the algorithm’s output.[9] Social media platforms such as X (formerly Twitter) and Bluesky give users recommendations from similar users and the content they interact with.[10] Additionally, personalized search offers users to explicitly filter search results by giving them the option to block content containing certain phrases or hashtags from being recommended.[11] Personalized search may build off content-based filtering to create an initial context for first-time users.[6] Other types of commercial websites such as search engines and retailers use similar processes to prescribe tailored information and products to a distinct user.
AI Contribution
editArtificial intelligence plays a prominent role in modern algorithmic curation, utilizing machine-learning models capable of handling mass amounts of data.[12] For example, deep learning and reinforcement learning allow curation algorithms to anticipate user preferences with greater precision in conjunction with established approaches.[12] This allows platforms to adjust what users see almost instantaneously.[12] In the context of social media and streaming services, this means AI arranges feeds to highlight what it deems relevant and carries with it the bias from the training data.[13]
Social Media and Potential Impact
editEcho chambers
editSocial media platform algorithms, such as X (formerly Twitter), recommend content it expects the user to positively engage with, which may lead to the curation of an echo chamber feed. Posts and accounts differing in perspectives are less likely to be suggested to one another by the algorithm, which may isolate like-minded users and lead to a decrease in source and topic diversity.[4] For example, Facebook's news feed is specifically designed to propagate content that matches users' interests, reinforcing their existing views.[14] While intended to keep users engaged, promoted content may create filter bubbles that offer little opportunity to engage with content outside preexisting views. Users have the option to actively filter out opposing views by blocking content, further aiding the potential creation of echo chambers.[4]
Over-representation
editA common pattern among social media platforms is the domination of the algorithm by a small set of users. Content produced by the most active users, users with the most followers, or users with the most engagement can compose a small percentage of an individual’s feed.[4]
See also
editReferences
edit- ^ Khan, Sadia; Bhatt, Ibrar (2018), "Curation", The International Encyclopedia of Media Literacy, John Wiley & Sons, Ltd, pp. 1–9, doi:10.1002/9781118978238.ieml0047, ISBN 978-1-118-97823-8, retrieved 2025-11-21
- ^ Berman, Ron; Katona, Zsolt (Sep 2016). "The Impact of Curation Algorithms on Social Network Content Quality and Structure". Working Papers.
- ^ a b c d e f g h i j k l m n Gausen, Anna; Luk, Wayne; Guo, Ce (2022-12-28). "Using Agent-Based Modelling to Evaluate the Impact of Algorithmic Curation on Social Media". J. Data and Information Quality. 15 (1): 2:1–2:24. doi:10.1145/3546915. ISSN 1936-1955.
- ^ a b c d e f g h i j k l m n o p q r s t u Bandy, Jack; Diakopoulos, Nicholas (2021-04-22). "More Accounts, Fewer Links: How Algorithmic Curation Impacts Media Exposure in Twitter Timelines". Proc. ACM Hum.-Comput. Interact. 5 (CSCW1): 78:1–78:28. doi:10.1145/3449152.
- ^ a b c d e f g Herlocker, Jonathan. "Explaining Collaborative Filtering Recommendations".
- ^ a b c d e f g h i j k "Online Recommender Systems – How Does a Website Know What I Want? |". Retrieved 2025-11-21.
- ^ a b c d Wang, Donghui; Liang, Yanchun; Xu, Dong; Feng, Xiaoyue; Guan, Renchu (2018-10-01). "A content-based recommender system for computer science publications". Knowledge-Based Systems. 157: 1–9. doi:10.1016/j.knosys.2018.05.001. ISSN 0950-7051.
- ^ a b c Roy, Deepjyoti; Dutta, Mala (2022-05-03). "A systematic review and research perspective on recommender systems". Journal of Big Data. 9 (1): 59. doi:10.1186/s40537-022-00592-5. ISSN 2196-1115.
- ^ Dou, Zhicheng; Song, Ruihua; Wen, Ji-Rong (2007-05-08). "A large-scale evaluation and analysis of personalized search strategies". Proceedings of the 16th international conference on World Wide Web. Banff Alberta Canada: ACM: 581–590. doi:10.1145/1242572.1242651. ISBN 978-1-59593-654-7.
- ^ Liu, Yuhan; Song, Emmy; Zhang, Owen Xingjian; Merriman, Jewel; Zhang, Lei; Monroy-Hernández, Andrés (2025-10-16). "Understanding Decentralized Social Feed Curation on Mastodon". Proc. ACM Hum.-Comput. Interact. 9 (7): CSCW507:1–CSCW507:25. doi:10.1145/3757688.
- ^ Quelle, Dorian; Bovet, Alexandre (2025-02-26). "Bluesky: Network topology, polarization, and algorithmic curation". PLOS ONE. 20 (2) e0318034. doi:10.1371/journal.pone.0318034. ISSN 1932-6203.
- ^ a b c Lazer, David; Swire-Thompson, Briony; Wilson, Christo (2024-09-01). "A Normative Framework for Assessing the Information Curation Algorithms of the Internet". Perspectives on Psychological Science. 19 (5): 749–757. doi:10.1177/17456916231186779. ISSN 1745-6916.
- ^ Villermet, Quentin; Poiroux, Jérémie; Moussallam, Manuel; Louail, Thomas; Roth, Camille (2021-09-13). "Follow the guides: disentangling human and algorithmic curation in online music consumption". Proceedings of the 15th ACM Conference on Recommender Systems. RecSys '21. New York, NY, USA: Association for Computing Machinery: 380–389. arXiv:2109.03915. doi:10.1145/3460231.3474269. ISBN 978-1-4503-8458-2.
- ^ Papa, Venetia; Photiadis, Thomas (2021-12-15). "Algorithmic Curation and Users' Civic Attitudes: A Study on Facebook News Feed Results". Information. 12 (12): 522. doi:10.3390/info12120522. hdl:20.500.14279/32861. ISSN 2078-2489. Archived from the original on 2025-06-22.