The diversity of life produced a large collection of morphological data, and the access to this information is necessary not only for researchers in taxonomy, ethnobotany, ecology and phylogenetics but also for amateurs and a general audience, especially those who contribute to citizen science projects.

The development of Xper2 (Ung et al., 2010), and more recently Xper3 (http://xper3.com) (http://www.optima-bot.org/meetings/XIVAbstracts.pdf, page 57) are dedicated to managing phenotypic descriptions. These descriptions are stored in a digital and structured form allowing automatic analysis and computer aided identification. We favour the use of modular and open technologies such as web services,  with a particular attention to ease-of-use and compatibility with open standard formats (SDD=Structured Descriptive Data (Hagedorn et al., 2005), CSV= Comma-separated values  and Nexus for export and external analyses). Xper3 and the identification webservice use web technologies (e.g. HTML5, Javascript, Ajax) to support collaborative work by multiple users.

Here, we present Xper3, the identification key webservice (Mkey+), and the learning process (Xperience) to improve the identification service. A knowledge base produced with Xper3 is used by a multiaccess identification key webservice to help the contributors of the citizen project SPIPOLL (SPIPOLL= Suivi Photographique des Insectes pOLLinisateurs). The objective is to fill the gaps in our knowledge concerning the changes in diversity of wild pollinators and their plants with changes in land-use (see http://www.spipoll.org/). More than 200 000 pictures have already been collected, identified by the contributors and checked by experts. Last year, the participant behaviour using the key webservice was also studied: which characters they used, the order of choice of these characters, doubts, etc. This information provided feedback on the adequacy of the key service. In order to improve the quality of the identification, we analysed these data and proposed a contextual and dynamic adjustment of the key. For example, confusion of taxa  automatically displays a warning message in the taxonomic forms for the concerned taxa. Errors of character and state choice, for all the taxa and for each taxon, are used to modify the weight of each character and to adjust the way the efficiency of a character is calculated at each step of the key.

Finally, we show how the proposed methods and their development may be applied to other applications using our identification webservice.