From Seo Wiki - Search Engine Optimization and Programming Languages

Jump to: navigation, search
The releation between knowlets and triples – redrawn, with his permission, from a slide by Barend Mons

The knowlet is a compact, dynamic, adaptive, semantic, topological unit of knowledge constructed out of connected concepts – a ‘concept cloud’ – representing knowledge about a given concept (the knowlet’s ‘core’ concept) by showing not only the connections of this ‘core’ concept with other concepts, but also the nature as well as the strength of those connections (‘associative distance’ in so-called multidimensional vector space).


Knowledge discovery

Knowlets – the term was coined by Barend Mons[1] – are designed to replicate key elements of the human brain’s associative reasoning functionality. Just as humans use an association matrix of concepts ‘they know about’ to read and understand a text – even more powerfully when they are experts in a field – the knowlet is meant to bring this power to vast and diverse elements of human thought. The ultimate goal of knowlets is to facilitate efficient discovery of new knowledge from ever-growing amounts of relevant information that is, due to its overabundance, not possible to take in and analyze with the traditional method of comprehensively reading the material. The knowlet facilitates both relational discovery[2], on the basis of ontological and observational connections between concepts, and associative discovery, on the basis of inferred connections.

Semantic, topological construct

Knowlets are ‘semantic’ constructs, because they are based on concepts – cognitive units of meaning – as opposed to just the (key)words that are used to describe concepts. The obvious advantage of using concepts rather than words is that the knowledge contained in a knowlet is unambiguous. Concepts can be of different semantic types, such as for instance the biomedical types of ‘proteins’, ‘diseases’, ‘anatomy’, ‘physiology’, and many more.

Knowlets are described as ‘topological’ units of knowledge on account of the fact that they are focused on the connections between concepts rather than, for instance, their similarities in meaning.


Each pair of connected concepts is known as a triple (concept-connection-concept, also known as ‘entity-event-entity’, or ‘subject-predicate-object’ – see also, for instance, the description in the article on Resource Description Framework), and a knowlet is constructed from such triples. A knowlet is ‘compact’ due to the fact that the triples used to construct it are disambiguated and any redundant triples are collapsed into one, representative, triple.

The nature of triples can be divided into three types: 1. ontological (sometimes called ‘factual’), 2. observational, and 3. inferred (also called ‘hypothetical’ or ‘associative’).

1. Ontological connections are those that have been included in authoritative ontologies, or commonly accepted as fact, and curated by experts.

2. Observational connections are those that are the result of observed co-occurrences in authoritative texts (e.g. the peer-reviewed scholarly literature), or co-expression in experiments (e.g. microarrays), and have not yet been formally accepted as facts or included in ontologies.

3. Inferred connections are those where two concept are not directly connected with an ontological or observational connection, but whose own knowlets (concept clouds) overlap to a significant degree. These inferred connections take on a predictive character when this overlap grows over time, allowing the hypothesis that the concepts in question may develop a ‘real’ connection. Inferred connections, particularly those that grow stronger, are the more interesting ones for knowledge discovery.

Dynamic and adaptive knowledge representation

When new ontological connections between two concepts become known, for instance when they are published in the peer-reviewed literature, they override pre-existing observational ones, and new observational ones override inferred connections. This is what makes knowlets ‘dynamic’ and ‘adaptive’, since emerging new knowledge changes them, ensuring that knowlets always represent the most recent state of knowledge. In this way, the knowlet can also detect ‘newness’ of a triple or a set of triples by comparing them with those already categorized in its corpus of known connections.


  1. Mons B, Ashburner M, Chichester C, et al. (2008). "Calling on a million minds for community annotation in WikiProteins". Genome Biol. 9 (5): R89. doi:10.1186/gb-2008-9-5-r89. PMID 18507872. 
  2. Swanson DR (January 1990). "Medical literature as a potential source of new knowledge". Bull Med Libr Assoc 78 (1): 29–37. PMID 2403828. 

Personal tools

Served in 0.646 secs.