Identifying explicit and tacit knowledge in a life science knowledge base
Abstract
An alternative to the use of synthetic pesticides and antibiotics in agriculture is to spray local plants extracts, in aqueous or essential oil form. To this end, the Knomana knowledge base [1] compiles various knowledge sets on plant use such as the 42000 descriptions of pesticidal plant uses for plant, animal, and public health presented in the literature. As the One Health approach dictates to be aware of the additional uses of these pesticidal plants to prevent their unintended effects on the animal, the human, and their environment, the challenge for the domain experts (e.g. entomologist, pathologist) is thus to identify the pesticidal plants in Knomana considering the One Health approach.
With the aim to present knowledge to the expert using a compact and comprehensive formalism, in [2], we computed the Duquenne-Guigues basis (DGB) of implications on an excerpt of Knomana, in which each plant is described using its taxonomy (i.e. species, genus, and family), to be consumed as food, and to be used in medical care. The DGB method is based on Formal Concept Analysis (FCA) and provides a cardinality-minimal set of non-redundant implications. By considering a reduced knowledge set, this work identifed 3 types of knowledge elements in the implications: knowledge on plant use at diverse taxonomy levels (e.g. Plants from Meliaceae family are not consumed as food), plant taxonomy (e.g. A plant from Salvia genus is from Lamiaceae family), and side effect of the knowledge set, e.g. a plant from the Piperaceae family is from the genus Piper. This latter illustration is not in accordance with taxonomic referential and thus informs on the extend of knowledge inserted in Knomana. Moreover, as plant taxonomy is known by the experts, removing it from the implications eases their reading but makes it tacit knowledge.
Implementing this method to select pesticidal plants requires to consider Knomana as a multidimensional (ternary) dataset, and thus to use the extension of FCA devoted to this kind of knowledge discovery, i.e. Relational Concept Analysis (RCA). Therefore, computing the DGB of implications based on RCA provides linked set of implications which includes the existential quantifer. Converting this formulation as practical expression is a need for the domain experts.
This poster describes the product line that formulates Knomana knowledge on pesticidal plants as implications, from which the implicit knowledge elements were removed and the side effects are highlighted to alert the expert. This product line was developed using the library fca4j from Cogui software (http://www.lirmm.fr/cogui/), that provides the RCA based DGB of implications, and using a post-process which differentiates the 3 types of knowledge elements within the implications. As an illustration, this poster presents the implications on Spodoptera frugiperda, a highly polyphagous insect that is close to invade South of Europe. The perspective of this work is to identify pesticidal European plants species that share chemical components similarities with plants used to control this pest in its native area.
Domains
Artificial Intelligence [cs.AI]
Fichier principal
Poster_Jobim_johanna_10.pdf (3.71 Mo)
Télécharger le fichier
posters_jobim2021_texte.pdf (190.4 Ko)
Télécharger le fichier
Origin | Files produced by the author(s) |
---|
Origin | Files produced by the author(s) |
---|