Hi all,
On behalf of José and myself, we would really like to thank the people who tried out our system and gave us feedback!
Some aspects are left to work on (for example, we have not tested for mobiles, etc.). However, we have made some minor initial changes reflecting some of the comments we received (adding example text for the type box, clarifying that the numbers refer to number of results not Q codes, etc.):
To summarise some aspects of the work and what we've learnt:
* In terms of usability, the principal lesson we have learnt (amongst many) is that it is not clear for users what is a type. For example, when searching for "popes born in Poland", the immediate response of users is to type "pope" rather than "human" or "person" in the type box. In a future version of the system, we might thus put less emphasis on starting the search with type (the original reasoning behind this was to quickly reduce the number of facets/properties that would be shown). Hence the main conclusion here is to try to avoid interfaces that centre around "types".
* A major design goal is that the user is only ever shown options that lead to at least one result. All facets computed are exact with exact numbers. The technical challenge here is displaying these facets with exact numbers and values for large result sizes, such as human:
http://grafa.dcc.uchile.cl/search?instance=Q5
This is achieved through caching. We compute all possible queries in the data that would yield >50,000 results (e.g., human->gender:male, human->gender:male->country:United States, etc.). We then compute their facets offline and cache them. In total there's only a couple of hundred such queries generating that many results. The facets for other queries with fewer than 50,000 results are computed live. Note that we cannot cache for keyword queries (instead we just compute facets for the first 50,000 most relevant results). Also, if we add other features such as range queries or sub-type reasoning, the issue of caching would become far more complex to handle.
In any case, thanks again to all those who provided feedback! Of course further comments or questions are welcome (either on- or off-list). Likewise we will be writing up a paper describing technical aspects of the system soon with some evaluation results. Once it's ready we will of course share a link with you.
Best, Aidan and José
-------- Forwarded Message -------- Subject: Re: GraFa: Faceted browser for RDF/Wikidata [feedback requested] Date: Mon, 15 Jan 2018 11:47:18 -0300 From: Aidan Hogan aidhog@gmail.com To: Discussion list for the Wikidata project. wikidata@lists.wikimedia.org CC: José Ignacio . joshep@live.cl
Hi all!
Just a friendly reminder that tomorrow we will close the questionnaire so if you have a few minutes to help us out (or are just curious to see our faceted search system) please see the links and instructions below.
And many thanks to those who have already provided feedback! :)
Best, José & Aidan
On 09-01-2018 14:18, Aidan Hogan wrote:
Hey all,
A Masters student of mine (José Moreno in CC) has been working on a faceted navigation system for (large-scale) RDF datasets called "GraFa".
The system is available here loaded with a recent version of Wikidata:
Hopefully it is more or less self-explanatory for the moment. :)
If you have a moment to spare, we would hugely appreciate it if you could interact with the system for a few minutes and then answer a quick questionnaire that should only take a couple more minutes:
https://goo.gl/forms/h07qzn0aNGsRB6ny1
Just for the moment while the questionnaire is open, we would kindly request to send feedback to us personally (off-list) to not affect others' responses. We will leave the questionnaire open for a week until January 16th, 17:00 GMT. After that time of course we would be happy to discuss anything you might be interested in on the list. :)
After completing the questionnaire, please also feel free to visit or list something you noticed on the Issue Tracker:
https://github.com/joseignm/GraFa/issues
Many thanks, Aidan and José