Lorraine Susan (Lori) Levin is an American computer scientist and computational linguist specializing in natural language processing, particularly involving syntax, morphosyntax, and languages with small corpora. She is a research professor in the Language Technologies Institute of the Carnegie Mellon University School of Computer Science, and one of the founders of the North American Computational Linguistics Open Competition. == Education and career == Levin has a 1979 bachelor's degree in linguistics (summa cum laude) from the University of Pennsylvania, and a 1986 Ph.D. in linguistics from the Massachusetts Institute of Technology. Her dissertation, Operations on Lexical Forms: Unaccusative Rules in Germanic Languages, was jointly supervised by Joan Bresnan and Kenneth L. Hale. She worked as an assistant professor of linguistics at the University of Pittsburgh from 1983 until 1988, when she joined the Carnegie Mellon University Language Technologies Institute. == Recognition == Levin was named as a Fellow of the Association for Computational Linguistics in 2025, "for pioneering work on the use of phonetics, syntax, lexical semantics and dialogue modeling in machine translation and in the transfer of NLP technologies to low resource languages, as well as an enduring contribution to the North American Computational Linguistics Olympiad". Levin was awarded the Antonio Zampolli prize of the ELRA Language Resources Association at the LREC 2026 conference.
Globetrooper
Globetrooper is a free travel app known for assisting travelers in finding partners for group trips and world adventures. Globetrooper offers a free social travel platform that helps people find travel partners. == History == Globetrooper was developed and released in 2010 by a couple; Todd Sullivan and Lauren McLeod who are two travel-minded individuals that wanted to make it easier for travelers to plan a journey and see the world. With their backgrounds in business, software & design, and a love for travel, both left the corporate world and launched Globetrooper on Lauren’s birthday 28 March 2010. Globetrooper was first launched as an information portal with a view to making it more social, but after some months, the content quickly grew and changed to the ‘travel partner’ concept.
SD-WAN
A Software-Defined Wide Area Network (SD-WAN) is a wide area network that uses software-defined networking technology, such as communicating over the Internet using overlay tunnels which are encrypted when destined for internal organization locations. If standard tunnel setup and configuration messages are supported by all of the network hardware vendors, SD-WAN simplifies the management and operation of a WAN by decoupling the networking hardware from its control mechanism. This concept is similar to how software-defined networking implements virtualization technology to improve data center management and operation. In practice, proprietary protocols are used to set up and manage an SD-WAN, meaning there is no decoupling of the hardware and its control mechanism. A key application of SD-WAN is to allow companies to build higher-performance WANs using lower-cost and commercially available Internet access, enabling businesses to partially or wholly replace more expensive private WANs connection technologies such as MPLS. When SD-WAN traffic is carried over the Internet, there are no end-to-end performance guarantees. Carrier MPLS VPN WAN services are not carried as Internet traffic, but rather over carefully controlled carrier capacity, and do come with an end-to-end performance guarantee. == History == WANs were very important for the development of networking in general and for a long time one of the most important applications of networks both for military and enterprise applications. The ability to communicate data over long distances was one of the main driving factors for the development of data communications, as it made it possible to overcome the distance limitations, as well as shortening the time necessary to exchange messages with other parties. Legacy WANs allowed communication over circuits connecting two or more endpoints. Earlier networking supported point-to-point communication over a slow speed circuit, usually between two fixed locations. As networking progressed, WAN circuits became faster and more flexible. Innovations like circuit and packet switching (in the form of X.25, ATM and later Internet Protocol or Multiprotocol Label Switching) allowed communication to become more dynamic, supporting ever-growing networks. The need for strict control, security and quality of service (QOS) meant that multinational corporations were very conservative in leasing and operating their WANs. National regulations restricted the companies that could provide local service in each country, and complex arrangements were necessary to establish truly global networks. All that changed with the growth of the Internet, which permitted entities around the world to connect to each other. However, over the first years, the uncontrolled nature of the Internet was not considered adequate or safe for private corporate use. Independent of safety concerns, connectivity to the Internet became a necessity to the point where every branch required Internet access. At first, due to safety concerns, private communications were still done via WAN, and communication with other entities (including customers and partners) moved to the Internet. As the Internet grew in reach and maturity, companies started to evaluate how to leverage it for private corporate communications. During the early 2000s, application delivery over the WAN became an important topic of research and commercial innovation. Over the next decade, increasing computing power made it possible to create software-based appliances that were able to analyze traffic and make informed decisions without delays, making it possible to create large-scale overlay networks over the public Internet that could replicate all the functionality of legacy WANs, at a fraction of the cost. SD-WAN combines several networking aspects to create full-fledged private networks, with the ability to dynamically share network bandwidth across the connection points. Additional enhancements include central controllers, zero-touch provisioning, integrated analytics and on-demand circuit provisioning, with some network intelligence based in the cloud, allowing centralized policy management and security. Networking publications started using the term SD-WAN to describe this new networking trend as early as 2014. With the rapid shift to remote work as a result of lockdowns and stay at home orders during the COVID-19 pandemic, SD-WAN grew in popularity as a way of connecting remote workers. == Overview == WANs allow companies to extend their computer networks over large distances, connecting remote branch offices to data centers and to each other, and delivering applications and services required to perform business functions. Due to the physical constraints imposed by the propagation time over large distances, and the need to integrate multiple service providers to cover global geographies (often crossing nation boundaries), WANs face important operational challenges, including network congestion, packet delay variation, packet loss, and even service outages. Modern applications such as VoIP calling, videoconferencing, streaming media, and virtualized applications and desktops require low latency. Bandwidth requirements are also increasing, especially for applications featuring high-definition video. It can be expensive and difficult to expand WAN capability, with corresponding difficulties related to network management and troubleshooting. SD-WAN products are designed to address these network problems. By enhancing or even replacing traditional branch routers with virtualization appliances that can control application-level policies and offer a network overlay, less expensive consumer-grade Internet links can act more like a dedicated circuit. This simplifies the setup process for branch personnel. SD-WAN products can be physical appliances or software based only. === Components === The MEF Forum has defined an SD-WAN architecture consisting of an SD-WAN edge, SD-WAN gateway, SD-WAN controller and SD-WAN orchestrator. ==== SD-WAN edge ==== The SD-WAN edge is a physical or virtual network function that is placed at an organization's branch/regional/central office site, data center, and in public or private cloud platforms. MEF Forum has published the first SD-WAN service standard, MEF 70 which defines the fundamental characteristics of an SD-WAN service plus service requirements and attributes. ==== SD-WAN gateway ==== SD-WAN gateways provide access to the SD-WAN service in order to shorten the distance to cloud-based services or the user, and reduce service interruptions. A distributed network of gateways may be included in an SD-WAN service by the vendor or setup and maintained by the organization using the service. By sitting outside the headquarters in the cloud, the gateway also reduces headquarters traffic. ==== SD-WAN orchestrator ==== The SD-WAN orchestrator is a cloud hosted or on-premises web management tool that allows configuration, provisioning and other functions when operating an SD-WAN. It simplifies application traffic management by allowing central implementation of an organization's business policies. ==== SD-WAN controller ==== The SD-WAN controller functionality, which can be placed in the orchestrator or in an SD-WAN gateway, is used to make forwarding decisions for application flows. Application flows are IP packets that have been classified to determine their user application or grouping of applications to which they are associated. The grouping of application flows based on a common type, e.g., conferencing applications, is referred to as an Application Flow Group in MEF 70. Per MEF 70, the SD-WAN Edge classifies incoming IP packets at the SD-WAN UNI (SD-WAN user network interface), determines, via OSI Layer 2 through Layer 7 classification, which application flow the IP packets belong to, and then applies the policies to block the application flow or allow the application flows to be forwarded based on the availability of a route to the destination SD-WAN UNI on a remote SD-WAN Edge. This helps ensure that application performance meets service level agreements (SLAs). == Required characteristics == The Gartner research firm has defined an SD-WAN as having four required characteristics: The ability to support multiple connection types, such as MPLS, last mile fiber optic network or through high speed cellular networks e.g. 4G LTE and 5G wireless technologies The ability to do dynamic path selection, for load sharing and resiliency purposes A simple interface that is easy to configure and manage The ability to support VPNs, and third party services such as WAN optimization controllers, firewalls and web gateways == Features == Features of SD-WANs include resilience, quality of service (QoS), security, and performance, with flexible deployment options; simplified administration and troubleshooting; and online traffic engineering. === Resilience === A resilient SD-WAN reduces network downtime. To
Influencer speak
Influencer speak is a speech pattern commonly associated with English-speaking digital content creators, particularly on platforms such as TikTok. This style is characterized by linguistic features such as uptalk, where intonation rises at the end of declarative sentences, and vocal fry, a low, creaky vibration in speech. These features are often used to engage audiences. == Characteristics == Influencer speak is commonly associated with: Uptalk – a rising intonation at the end of statements Vocal fry – a creaky sound often occurring at the end of sentences Use of filler words and slang – contributes to a conversational tone that resonates with audiences == Origins == The origins of "influencer speak" are linked to the "Valley Girl" accent, which became prominent in the 1980s. This earlier style included features such as uptalk and vocal fry, which have been adapted for digital platforms. Linguists have noted that these patterns are often led by young women, who are recognized as linguistic innovators in sociolinguistic research. == Sociolinguistic significance == "Influencer speak" is used to maintain audience engagement. Features such as uptalk help speakers retain the "conversational floor," ensuring continuous attention from listeners. A study conducted by UCLA researchers has shown that creators adjust their speech styles based on the platform and audience. For example, a comedic tone may be emphasized on TikTok, while a more professional tone may be used on platforms such as LinkedIn or YouTube.
Web content development
Web content development is the process of researching, writing, gathering, organizing, and editing information for publication on websites. Website content may consist of prose, graphics, pictures, recordings, movies, or other digital assets that could be distributed by a hypertext transfer protocol server, and viewed by a web browser. == Web developers and content developers == When the World Wide Web began, web developers either developed online content themselves, or modified existing documents and coded them into hypertext markup language (HTML). In time, the field of website development came to encompass many technologies, so it became difficult for website developers to maintain so many different skills. Content developers are specialized website developers who have content generation skills such as graphic design, multimedia development, professional writing, and documentation. They can integrate content into new or existing websites without using information technology skills such as script language programming and database programming. Content developers or technical content developers can also be technical writers who produce technical documentation that helps people understand and use a product or service. This documentation includes online help, manuals, white papers, design specifications, developer guides, deployment guides, release notes, etc. == Search engine optimization == Content developers may also be search engine optimization specialists, or internet marketing professionals. High quality, unique content is what search engines are looking for. Content development specialists, therefore, have a very important role to play in the search engine optimization process. One issue currently plaguing the world of web content development is keyword-stuffed content which are prepared solely for the purpose of manipulating search engine rankings. The effect is that content is written to appeal to search engine (algorithms) rather than human readers. Search engine optimization specialists commonly submit content to article directories to build their website's authority on any given topic. Most article directories allow visitors to republish submitted content with the agreement that all links are maintained. This has become a method of search engine optimization for many websites today. If written according to SEO copywriting rules, the submitted content will bring benefits to the publisher (free SEO-friendly content for a webpage) as well as to the author (a hyperlink pointing to his/her website, placed on an SEO-friendly webpage). == New content types == Web content is no longer restricted to text. Search engines now index audio/visual media, including video, images, PDFs, and other elements of a web page. Website owners sometimes use content protection networks to scan for plagiarized content.
Neural style transfer
Neural style transfer (NST) software algorithms are able to manipulate digital images, or videos, in order to adopt the appearance or visual style of another image. NST algorithms are characterized by their use of deep neural networks for the sake of image transformation. Common uses for NST are the creation of artificial artwork from photographs, for example by transferring the appearance of famous paintings to user-supplied photographs. Several notable mobile apps use NST techniques for this purpose, including DeepArt and Prisma. This method has been used by artists and designers around the globe to develop new artwork based on existent style(s). == History == NST is an example of image stylization, a problem studied for over two decades within the field of non-photorealistic rendering. The first two example-based style transfer algorithms were image analogies and image quilting. Both of these methods were based on patch-based texture synthesis algorithms. Given a training pair of images–a photo and an artwork depicting that photo–a transformation could be learned and then applied to create new artwork from a new photo, by analogy. If no training photo was available, it would need to be produced by processing the input artwork; image quilting did not require this processing step, though it was demonstrated on only one style. NST was first published in the paper "A Neural Algorithm of Artistic Style" by Leon Gatys et al., originally released to ArXiv 2015, and subsequently accepted by the peer-reviewed CVPR conference in 2016. The original paper used a VGG-19 architecture that has been pre-trained to perform object recognition using the ImageNet dataset. In 2017, Google AI introduced a method that allows a single deep convolutional style transfer network to learn multiple styles at the same time. This algorithm permits style interpolation in real-time, even when done on video media. == Mathematics == This section closely follows the original paper. === Overview === The idea of Neural Style Transfer (NST) is to take two images—a content image p → {\displaystyle {\vec {p}}} and a style image a → {\displaystyle {\vec {a}}} —and generate a third image x → {\displaystyle {\vec {x}}} that minimizes a weighted combination of two loss functions: a content loss L content ( p → , x → ) {\displaystyle {\mathcal {L}}_{\text{content }}({\vec {p}},{\vec {x}})} and a style loss L style ( a → , x → ) {\displaystyle {\mathcal {L}}_{\text{style }}({\vec {a}},{\vec {x}})} . The total loss is a linear sum of the two: L NST ( p → , a → , x → ) = α L content ( p → , x → ) + β L style ( a → , x → ) {\displaystyle {\mathcal {L}}_{\text{NST}}({\vec {p}},{\vec {a}},{\vec {x}})=\alpha {\mathcal {L}}_{\text{content}}({\vec {p}},{\vec {x}})+\beta {\mathcal {L}}_{\text{style}}({\vec {a}},{\vec {x}})} By jointly minimizing the content and style losses, NST generates an image that blends the content of the content image with the style of the style image. Both the content loss and the style loss measures the similarity of two images. The content similarity is the weighted sum of squared-differences between the neural activations of a single convolutional neural network (CNN) on two images. The style similarity is the weighted sum of Gram matrices within each layer (see below for details). The original paper used a VGG-19 CNN, but the method works for any CNN. === Symbols === Let x → {\textstyle {\vec {x}}} be an image input to a CNN. Let F l ∈ R N l × M l {\textstyle F^{l}\in \mathbb {R} ^{N_{l}\times M_{l}}} be the matrix of filter responses in layer l {\textstyle l} to the image x → {\textstyle {\vec {x}}} , where: N l {\textstyle N_{l}} is the number of filters in layer l {\textstyle l} ; M l {\textstyle M_{l}} is the height times the width (i.e. number of pixels) of each filter in layer l {\textstyle l} ; F i j l ( x → ) {\textstyle F_{ij}^{l}({\vec {x}})} is the activation of the i th {\textstyle i^{\text{th}}} filter at position j {\textstyle j} in layer l {\textstyle l} . A given input image x → {\textstyle {\vec {x}}} is encoded in each layer of the CNN by the filter responses to that image, with higher layers encoding more global features, but losing details on local features. === Content loss === Let p → {\textstyle {\vec {p}}} be an original image. Let x → {\textstyle {\vec {x}}} be an image that is generated to match the content of p → {\textstyle {\vec {p}}} . Let P l {\textstyle P^{l}} be the matrix of filter responses in layer l {\textstyle l} to the image p → {\textstyle {\vec {p}}} . The content loss is defined as the squared-error loss between the feature representations of the generated image and the content image at a chosen layer l {\displaystyle l} of a CNN: L content ( p → , x → , l ) = 1 2 ∑ i , j ( A i j l ( x → ) − A i j l ( p → ) ) 2 {\displaystyle {\mathcal {L}}_{\text{content }}({\vec {p}},{\vec {x}},l)={\frac {1}{2}}\sum _{i,j}\left(A_{ij}^{l}({\vec {x}})-A_{ij}^{l}({\vec {p}})\right)^{2}} where A i j l ( x → ) {\displaystyle A_{ij}^{l}({\vec {x}})} and A i j l ( p → ) {\displaystyle A_{ij}^{l}({\vec {p}})} are the activations of the i th {\displaystyle i^{\text{th}}} filter at position j {\displaystyle j} in layer l {\displaystyle l} for the generated and content images, respectively. Minimizing this loss encourages the generated image to have similar content to the content image, as captured by the feature activations in the chosen layer. The total content loss is a linear sum of the content losses of each layer: L content ( p → , x → ) = ∑ l v l L content ( p → , x → , l ) {\displaystyle {\mathcal {L}}_{\text{content }}({\vec {p}},{\vec {x}})=\sum _{l}v_{l}{\mathcal {L}}_{\text{content }}({\vec {p}},{\vec {x}},l)} , where the v l {\displaystyle v_{l}} are positive real numbers chosen as hyperparameters. === Style loss === The style loss is based on the Gram matrices of the generated and style images, which capture the correlations between different filter responses at different layers of the CNN: L style ( a → , x → ) = ∑ l = 0 L w l E l , {\displaystyle {\mathcal {L}}_{\text{style }}({\vec {a}},{\vec {x}})=\sum _{l=0}^{L}w_{l}E_{l},} where E l = 1 4 N l 2 M l 2 ∑ i , j ( G i j l ( x → ) − G i j l ( a → ) ) 2 . {\displaystyle E_{l}={\frac {1}{4N_{l}^{2}M_{l}^{2}}}\sum _{i,j}\left(G_{ij}^{l}({\vec {x}})-G_{ij}^{l}({\vec {a}})\right)^{2}.} Here, G i j l ( x → ) {\displaystyle G_{ij}^{l}({\vec {x}})} and G i j l ( a → ) {\displaystyle G_{ij}^{l}({\vec {a}})} are the entries of the Gram matrices for the generated and style images at layer l {\displaystyle l} . Explicitly, G i j l ( x → ) = ∑ k F i k l ( x → ) F j k l ( x → ) {\displaystyle G_{ij}^{l}({\vec {x}})=\sum _{k}F_{ik}^{l}({\vec {x}})F_{jk}^{l}({\vec {x}})} Minimizing this loss encourages the generated image to have similar style characteristics to the style image, as captured by the correlations between feature responses in each layer. The idea is that activation pattern correlations between filters in a single layer captures the "style" on the order of the receptive fields at that layer. Similarly to the previous case, the w l {\displaystyle w_{l}} are positive real numbers chosen as hyperparameters. === Hyperparameters === In the original paper, they used a particular choice of hyperparameters. The style loss is computed by w l = 0.2 {\displaystyle w_{l}=0.2} for the outputs of layers conv1_1, conv2_1, conv3_1, conv4_1, conv5_1 in the VGG-19 network, and zero otherwise. The content loss is computed by w l = 1 {\displaystyle w_{l}=1} for conv4_2, and zero otherwise. The ratio α / β ∈ [ 5 , 50 ] × 10 − 4 {\displaystyle \alpha /\beta \in [5,50]\times 10^{-4}} . === Training === Image x → {\displaystyle {\vec {x}}} is initially approximated by adding a small amount of white noise to input image p → {\displaystyle {\vec {p}}} and feeding it through the CNN. Then we successively backpropagate this loss through the network with the CNN weights fixed in order to update the pixels of x → {\displaystyle {\vec {x}}} . After several thousand epochs of training, an x → {\displaystyle {\vec {x}}} (hopefully) emerges that matches the style of a → {\displaystyle {\vec {a}}} and the content of p → {\displaystyle {\vec {p}}} . As of 2017, when implemented on a GPU, it takes a few minutes to converge. == Extensions == In some practical implementations, it is noted that the resulting image has too much high-frequency artifact, which can be suppressed by adding the total variation to the total loss. Compared to VGGNet, AlexNet does not work well for neural style transfer. NST has also been extended to videos. Subsequent work improved the speed of NST for images by using special-purpose normalizations. In a paper by Fei-Fei Li et al. adopted a different regularized loss metric and accelerated method for training to produce results in real-time (three orders of magnitude faster than Gatys). Their idea was to use not the pixel-based loss defined above but rather a 'perceptual loss' measuring t
Algorithmic amplification
Algorithmic amplification is the process by which automated ranking and recommendation systems on digital platforms increase the visibility of certain content beyond its initial audience. Major platforms including Facebook, YouTube, TikTok, and X (formerly Twitter) use such systems to determine what appears in users' feeds and search results. The term is used in research on social media and digital media regulation to describe how platform design choices influence the distribution of online information. Unlike chronological feeds, algorithmic systems evaluate content using signals such as engagement rates, viewing duration, and predicted relevance to individual users. Content that performs strongly on these metrics may be promoted to progressively larger audiences through feeds, search rankings, or autoplay systems. The process is distinct from content moderation, which involves removing, labelling, or restricting content under platform rules, although the two can interact in practice. The concept is closely connected to the attention economy. Research has linked algorithmic amplification to the spread of misinformation and the circulation of political content, as well as to effects on young users' mental health. The scale and direction of those effects remain debated, in part because independent researchers have limited access to the internal workings of platform recommendation systems. Governments in the European Union, United Kingdom, United States, and China have pursued differing regulatory approaches to recommendation algorithms. The EU's Digital Services Act and the UK's Online Safety Act 2023 impose obligations on large platforms related to recommendation system transparency and risk, while China became the first country to enact binding legislation specifically targeting such systems. Internal documents and whistleblower testimony reported by the BBC in 2026 described how competitive pressure between Meta and TikTok led to trade-offs between engagement and user safety in the design of their recommendation systems. == Terminology == The term algorithmic amplification is used in media studies, platform governance scholarship and regulatory literature to describe how automated systems influence the distribution of content beyond what organic user sharing alone would produce. It is distinct from viral spread, which refers primarily to user-driven sharing behaviour, and from algorithmic bias, which describes systematic errors or unfairness in algorithmic outputs. The related term algorithmic curation is used for the broader process of selecting and ordering content, of which amplification is one possible outcome. The phrase also appears in regulatory and legislative discussion of recommendation systems. The European Union's Digital Services Act (DSA) identifies recommendation systems as a potential source of systemic risk, and the term appears frequently in academic and policy commentary on the regulation. In the United States, proposals including the Filter Bubble Transparency Act and the Kids Online Safety Act (KOSA) have used it to frame requirements around recommendation system transparency. In the United Kingdom, the House of Commons Science, Innovation and Technology Committee used the term in a 2025 report on how recommendation algorithms contributed to the spread of misinformation during the 2024 Southport riots. A Joint Declaration on AI and Freedom of Expression adopted in October 2025 by four international freedom of expression mandate holders, including the UN Special Rapporteur on Freedom of Opinion and Expression and the OSCE Representative on Freedom of the Media, stated that recommender systems and other AI-powered curation tools exert "a large hidden influence and gatekeeper role" over what information people access and consume. == Background == Early internet platforms typically displayed content in reverse-chronological order or through keyword-based search systems. Although the term is most often applied to social media, the underlying logic predates social media itself. A 2021 overview traced the origins of modern recommendation systems to the early 1990s, when they were first used experimentally for personal email and information filtering. The 1992 Tapestry mail system and the 1994 GroupLens news filtering system were early milestones before recommendation systems spread into e-commerce and other online services. As user bases and content volumes grew during the 2000s, major platforms including Google, YouTube, and Facebook developed machine-learning systems to personalise content delivery and prioritise material predicted to generate engagement. Facebook introduced its News Feed in 2006, which gradually shifted from chronological presentation towards algorithmically ranked content. YouTube altered its recommendation system in 2012 to prioritise watch time rather than clicks, a change the platform said was prompted by concerns that click-based metrics encouraged misleading thumbnails and low-quality videos. TikTok, launched internationally in 2018, adopted a model in which its primary content surface, the For You feed, is driven almost entirely by algorithmic recommendation rather than by a user's social graph. An internal document obtained by The New York Times in 2021 showed that the platform's algorithm optimised for retention and time spent, using signals such as watch duration, replays, likes, and comments to score and rank videos. Algorithmic recommendation also became central to platforms outside social media. Spotify's personalised features, including Discover Weekly, Release Radar, and Home recommendations, use behavioural signals and inferred "taste profiles" to surface tracks and artists beyond a listener's existing library. An ethnographic study of music curators at streaming platforms described this blend of algorithmic and human editorial selection as an "algo-torial" model of gatekeeping. Amazon adopted item-based collaborative filtering for product recommendations in 1998, and its recommendation engine has been described as one of the earliest large-scale deployments of recommendation technology in e-commerce. The same dynamics operate on adult content platforms. Law professor Amy Adler has argued that from 2007 onwards the pornography industry migrated to algorithm-driven streaming platforms, most of which are controlled by a single near-monopoly company, Aylo (formerly MindGeek). These platforms use algorithmic search engines, suggestions, rigid categorisation of content, and AI-driven search term optimisation in ways that produce the same distorting effects found on mainstream speech platforms, including filter bubbles, feedback loops, and the tendency of algorithmic recommendations to alter individual preferences. == Mechanisms == Recommendation systems commonly combine collaborative filtering, which predicts a user's preferences from the behaviour of similar users, with machine-learning models that predict which content a user is likely to engage with from their prior activity. In a common two-stage design, a platform first generates a set of candidate items from a large content pool and then ranks them using a scoring model with objectives such as predicted engagement or user satisfaction. Small changes in ranking criteria can shift exposure at scale, particularly when applied repeatedly across multiple browsing sessions. These systems typically rely on signals including engagement rates, viewing duration, click-through rates, and network relationships between users. Modern recommendation pipelines continuously update predictions as new behavioural data arrives, allowing platforms to adjust rankings in near real time. Users' revealed preferences, expressed through behaviour such as clicks and viewing time, do not always align with their stated preferences, expressed through explicit feedback such as surveys or content controls. Popularity signals can create feedback dynamics in which early engagement increases the likelihood that content will be shown to additional users. Experimental research on online cultural markets has demonstrated how such feedback processes can produce unequal visibility outcomes even when initial differences in content quality are small. == Beneficial and public-interest uses == Recommendation systems can help users navigate large volumes of content by surfacing material predicted to match their interests or needs, which can improve discoverability on platforms with large content libraries. In public health communication, platforms can help health authorities distribute timely information at scale, though the same recommendation systems also risk amplifying misinformation alongside official guidance. Sociologist Zeynep Tufekci has argued that the shift from independent blogs to large centralised platforms transferred gatekeeping power from traditional media to corporate algorithms. In the case of the Egyptian uprising of 2011, she noted that ordinary users