REPORT A PROBLEM
LINKS
Resource link (short)
http://azon.e-science.pl/zasoby/83087Resource link (repository)
https://id.e-science.pl/records/83087Resource metadata
Title |
Computing multidimensional semantic similarity in ontologies (PD-2021-04) |
---|---|
Persons |
Authors:
Paweł Szmeja
Partner: Systems Research Institute Polish Academy of Sciences, Warsaw |
Description |
The notion of semantic similarity has been extensively explored in many different fields, including computer science. Various similarity models, along with specific measures that work within them, have been proposed over the years in many different contexts, and for both generic, and domain–specific applications. A holistic view of the area reveals problems that need to be addressed, including the lack of generic models that would be able to exploit full extent of modern data sources, rich with complex semantic descriptions, as well as the shortcomings in interpretation of similarity scores. The similarity framework SimDim presented in this thesis proposes a specific approach to similarity scoring that allows detailed comparisons of similarity between semantic objects in any domain. It includes the dimensional approach, which adds a layer of meaning to similarity scores themselves, allowing for more informed choice of measures for any given problem, and more awareness in interpretation and comparison of scores. Dimensions of similarity frame the notion of similarity in terms of the meaning of data, that was used to arrive at the result, and not, as it was done before, in terms of model of similarity, or the data format. The dimensional score enables capturing of different kinds of similarity, and, therefore, makes the score more informative. The SimDim framework is based on a similarity model and a generic algorithm for semantic similarity, with implementation. The algorithm is a highly configurable and domain–independent tool that enables introduction of the dimensional approach into practice. It allows a deep view of objects and their features, that includes all available knowledge, or a subset of it, explicitly chosen to address specific tasks. It works within the similarity model, in which calculation of features is encapsulated in concrete functions, that are “first class citizens”. Under this approach, the algorithm may be parametrized in many different ways, making use of various ways to calculate features. Particular configurations (i.e. parametrizations) of the algorithm result in specific measures, examples of which are included as well. The parts of the SimDim framework – the dimensional approach, the model, and the algorithm work in tandem to deliver an integrated solution to, and offer a different (to currently available solutions) perspective on the problem of similarity calculation. (English) |
Keywords | "podobieństwo"@pl, "similarity"@en |
Classification |
Resource type:
thesis Scientific discipline: Dziedzina nauk inżynieryjno-technicznych / informatyka techniczna i telekomunikacja (2018) Destination group: pupils, students, scientists Harmful content: No |
Characteristics |
Place of creation: Warszawa
Creation time: 2021 Number of pages: 144 Supervisor: Maria Ganzha Resource language: English |
License | CC BY-SA 4.0 |
Technical information |
Submitter: Anna Wasilewska Availability date: 17-03-2023 |
Collections | Kolekcja Instytutu Badań Systemowych PAN w Warszawie |
Similar resources
Intuicjonistyczne zbiory rozmyte w komputerowym określaniu podobieństwa dokumentów tekstowych
Adam Niewiadomski, thesis, Systems Research Institute Polish Academy of Sciences, Warsaw, dziedzina nauk technicznych / informatyka (2011)