People go to the web to satisfy their curiosity. The web contains resources that can help: articles, videos, tutorials, online communities and courses, among others. In analogy to the semantic web proposal, which was motivated by a desire to structure the web to be more understandable and usable by machines [Berners-Lee, Hendler, and Lassila, O., 2001], we raise the question: How would we rethink the web from the primary goal of satisfying and fostering human curiosity? We propose the curiosity web based on the intuition that the meaning of documents, such as articles, books, and videos, can be expressed in terms of the questions they address [Paritosh and Marcus, 2016]. It has three representational elements: 1) Curiosity: a URI for a question or information need, 2) Relationships between curiosities: such as relevant, prerequisite, and c) Relationships between curiosities and documents: such as addresses, satis es. The goal of the curiosity web is to provide an exoskeleton for organizing information by the curiosities they address. The curiosity web is a dual of existing semantic networks and knowledge graphs [Collins and Quillian, 1972; Sowa, 2006; Hillis, 2004]. Instead of focusing on describing meaning using analytic primitives and compositions of knowledge, this approach represents meaning in a holistic manner, by the curiosities it addresses.
In areas of information of wide public value, such as healthcare, culture, lifestyle, arts; social, experiential, and subjective knowledge abounds. Consider the information needs of an expectant mother over the course of the pregnancy to childbirth to raising and parenting the child. While we have made great strides in representing and organizing structured knowledge about people, places, and things, not as much progress has been made in organizing such social knowledge.
For example, Wikipedia, knowledge graphs, web markup, and wikidata, to name a few, can tell us a lot about the movie Blade Runner: when it was released, who produced it, the cast, the screenwriter, the ratings on imdb and other sites, etc. These are veri able facts that we have consensus on. But there are more complex curiosities about the movie, e.g., { Why did they name the movie Blade Runner? { What makes it a cult classic? { What are the di erences between the various di erent cuts of this movie? { What does the origami represent in the movie?
These are legitimate curiosities. The state-of-the-art in knowledge graphs and structured data does not have the schemata to represent the knowledge required to understand, let alone answer such questions. In addition, we lack scalable methods of curating such social knowledge as it involves subjectivity and multiple, even con icting, perspectives. Aggregating and ranking such knowledge goes beyond the existing operationalizations of consensus and authority. Sociologists have called this the social stock of knowledge [Berger and Luckman, 1991], library scientists have named this everyday life information seeking [Spink and Cole, 2001], but our communities have not paid enough attention to these domains of knowledge.
There is content addressing these valuable class of information needs, but they are hard to nd. It is prevalent in forums and blogs and reviews and email groups and social media discussions and comments and videos and books. This knowledge is trickier to represent, curate, and evaluate. It is full of subjectivity, ambiguity, disagreement, and di ering perspectives. Nevertheless, this social stock of knowledge is vital for addressing the information needs of patients coping, parents distraught, children curious, and solving real world problems. 2
Knowledge bases such as Wikipedia have too strict of an epistemological position (e.g., objectivity, notability, veri ability) that makes it impossible to talk about most of social knowledge that we are interested in { this knowledge literally has no home! And so it stays in the halfway houses of community QA websites and forums. While Wikipedia is curated and organized purposefully for a wide readership, much of social knowledge on the web is found knowledge created as a by product of human interaction in online communities. To someone who is not part of this community and thread, this content is not nessed and curated for their consumption. More structured knowledge graphs [Bollacker et al., 2008] lack the primitives such as verbs, adjectives, and adverbs, as well as the notion of context and perspective [Guha, 1991], which makes it impossible to represent social knowledge.
Knowledge graphs, topic maps, concept maps provide exoskeletons for organizing the map of knowledge. What are they missing? Topics are too coarse and are not helpful for describing the aspects that you care about, and the connectivity between topics does not include any kind of conceptual or curricular organization. And lastly, in our communities of computer scientists and engineers, we have a epistemological inclination toward precision, truth, accuracy, which is limiting for characterizing social knowledge.
The curiosity web is an approach to organizing knowledge by curiosities. The underlying intuition is that we can express what a document does, for the purposes of information consumption, by connecting it to the questions it addresses. We choose to start with questions, since they are shorter and simpler than answers. In addition, we can disagree about the answers more than we can about the questions, providing a more stable foundation for the knowledge in answers. It involves curating URIs for curiosities and the relationships between curiosities (such as related, prerequisite), and relationships between curiosities and resources (such as addresses, satis es).
First, it proposes a Curiosity as a rst class semantic primitive for representing information needs. A curiosity has a stable URI, textual descriptions and elaborations in multiple languages. Just Wikipedia curates the topic for the book Alice in Wonderland represents and abstracts away across speci c editions and copies of the books, curiosities will be curated by abstracting from questions. Thus, a curiosity might further point to one or more questions on the web that surface that curiosity.
Second, the curiosity web contains relationships between curiosities. The most simple relationship between two curiosities might be that they are relevant to each other, e.g., someone interested in one might nd the other useful. In addition, there might be curricular relations such as prerequisite, enables, and other relationships such as temporal order, granularity, level of detail, etc.
Third, the curiosity web contains relationships between curiosities and documents. The simplest relationship is that a document addresses a curiosity. We can imagine additional relationships such as evokes, satis es, etc. Figure 1 shows a hypothetical example of a small part of the curiosity web around curiosities related to parenting. 4
There are existing examples of organizing knowledge using questions, e.g., the widespread usage of frequently asked questions (FAQs) to organize community knowledge on forums and usenet [Hammond et al., 1995]. Another starting point for questions on the web might be in schema.org/Question markup [Guha, Brickley, Macbeth, 2016]. Further, community QA sites like stackexchange.com merge similar questions into one thread, so as to not fragment content across multiple threads. Another example is amazon.com, which has a crowdsourced product FAQ page attached to every product on its website. Questions are a compact human understandable and curatable knowledge representation for information needs.
Building and maintaining the curiosity web is feasible, below are some models addressing parts of the task: 1. Community driven: Imagine Curiositypedia, built on top of Wikipedia community principles to curate curiosities and connections between them. 2. Webmaster driven: Imagine schema.org markup for articles and content by webmasters to make their content more available to user questions and search engines. 3. Machine driven: Using technologies such as statistical machine translation, annotate documents and resources on the web with curiosities they address 4. User driven: As users consume content, they can help annotate it with their questions that the content addressed, much like collaborative tagging e orts [Golder and Huberman, 2006]. Everybody can be a curator, and the web gets better as users leave trails! 5
Curiosity is the urge to know. Curiosity is the fascination with the unknown. A curiosity is a desire to know or learn something, e.g., one might want to know how to make their omelettes u er, or why their ve-year old cant shake of his temper tantrums, or the meaning of it all. This might be motivated by a need to solve a problem or explore new domains. This might surface in forms such as questions posed to friends or queries to search engines. It surfaces as piquing of the neck, furrowing of brows, rising pitch in the cadence, the question mark in the sentence, the queries to search engines, the questions for elders and librarians, the communities of inquiry, the processes of accumulating knowledge, such as science and spirituality.
The obvious unknown is the known unknown { the location of the restaurant your friend recommended, the latest on the California res, the weather tomorrow, etc. These are unknowns just on the surface of what we know and need. Resolving them quickly improves our lives and makes us expand our surface of known. Then there is the unknown unknown. Things that are so beyond our stock of known that we wouldnt even know what to ask. And yet, our curiosity is fascinated with these unknowns { the desire to travel, to seek new experiences, to go outside our comfort zone, to get intoxicated, to seek diversity; these are some real risks and commitments to a fascination with this distant unknown. Even though we dont have a speci c question (usually) when we visit a new place, that journey and interactions there could teach us new things, or help us see what we knew in a new light, and lead us to new questions.
The curiosity web can make it easier for the creators, curators, and consumers of content to annotate content with curiosities they address. Someone else, who later has that curiosity, does not have to gure out how their curiosity maps to some taxonomy, or know the right terms to pose queries, but can more directly rediscover such content. In addition, being able to connect curiosities together allows us to build and follow the trails of knowledge that the Memex machine dreamt of [Bush, 1945].