Font Size: 

Broadly our objective was to create a system for high-level reasoning to automate the identification of situations of interest among flower-visiting records, thereby simulating the combined inferencing ability of a group of domain experts.

Several experts were interviewed about their knowledge of flower-visiting insects, especially the behaviour of insects and the morphology of flowers as factors affecting the probability of the occurrence of pollination.

Causality was an important consideration, and for this reason we used the process of constructing a Bayesian network as a requirements-gathering tool to identify the factors that are believed to have a causal influence on the transfer of pollen from one flower ‘s anthers to the stigma of another flower of the same plant species.

The nodes of the Bayesian network were then translated into natural language statements describing behavioural interactions constituting the stages of flower-visiting and pollination e.g. (working backwards): Transfer of pollen, collecting or ingesting nectar/pollen/oil, flower-visiting. The knowledge elements required for high-level reasoning were abstracted from these natural language statements and formulated as follows:

It is [degree of probability] that [behavioural interaction] occurred if [combination of causal behavioral/ ecological factors]

e.g. It is [most probable] that [collecting or ingesting of nectar] occurred if [the arthropod species is oligophagous] and if [the arthropod is a female bee or female pollen wasp] and if [the date of the observed behaviour is within the plant species’ flowering period] and if [this plant species produces floral nectar as a reward].

The knowledge representation requirements will be used in two ways:

1) By evaluating an existing flower-visiting ontology against the knowledge representation requiremements we will identify gaps in the ontology and extend, develop and refine the concepts in the ontology;

2) We will create a hybrid knowledge model that will combine the discrete knowledge of an ontology with the predictive power of a Bayesian network.

The broader context of data-mining in biodiversity and implications for restructuring of biodiversity databases will also be discussed.