• [^] # Re: du boulot!

    Posté par . En réponse au journal Wikum : Résumé et récursion. Évalué à 3.

    Un copier/coller du résumé de l’article :

    We design a workflow to create a summary tree using the idea of recursive summarization of a discussion, where users build summaries of small sections of the discussion, small sets of those summaries are then aggregated and summarized, and so on until the entire discussion is incorporated into the layered summary tree.

    Editors can then continue to create new summaries that can distill both previously-written summaries and unsummarized comments, until we are left with a summary of the entire discussion.

    Even before the summary tree is complete, the summaries that people write in Wikum are embedded in the original discussion and contribute towards making the discussion easier to read. In threaded discussions, the summary of a subtree lives "Between" the comment and its parent.

    Building the Summary Tree For readers of a discussion, Wikum lets them see a visual overview, differentiate between summaries and comments, explore into summaries, and jump between conversations.

    If a summary of a subthread has been written, a person writing a summary at a higher level in the discussion thread can promote the lower summary to their position and build on the summary text; this lower summary can be a useful starting point for authoring the higher-level summary.

    We chose to stop subjects working on both tasks after each Wikum summary was completed because we wanted to use our other user study participants to provide feedback on the Wikum summary qualities as opposed to spending all their study time finishing the Google Docs summaries.

    A thousand words of summary is around two pages long, which may be more than someone is willing to read. However, because of recursive summarization in the Wikum case, users can read a 250-word summary of the entire discussion and drill in to get more detailed summaries.

    (paramètres par défaut)

    http://smmry.com/https://people.csail.mit.edu/axz/papers/wikum.pdf#&SM_LENGTH=7

    À comparer avec le mien :p Il y a des trucs que j’ai pas dit et je dis des trucs qui n’y sont pas. Le automatique n’est pas ultra cohérent étant donné qu’il y a des détails sur l’expérience mais pas les résultats, ce qui est fâcheux. Niveau description des fonctionnalités par contre c’est pas mal.