devadvance / relfinder Goto Github PK

Since multiple restrictions of various types can be put on the result set,
not all can be visible at once and thus an overview of all restrictions
that are currently in effect is needed.
The column's heads seem to be a good place for that.

Since it is now possible to search relations also between more than two
'starting nodes', the number of 'starting nodes' that are connected by a
relation might be another helpful value to filter the graph. The user
might, for instance, be only interested in relationships that connect all
'starting nodes' or at least three or four. Thus, another sortable table
that let the user fade in/out relationships depending on the number of
'starting nodes' they connect would make sense now.

It should be possible to correct a false disambiguation, i.e. allow the
user to assign her input to another object of the data set.

A straight-forward solution would be to open the manual disambiguation
dialogue if the user clicks on the (already disambiguated) label in the
input field (since it looks like a button, why shouldn't it be one?).

What steps will reproduce the problem?

If the word "Term" is entered in the input field, there are many options for 
auto-completion, e.g. "academic term. Once more letters of the word 
"Terminator" are entered, e.g. "Termin", you get no options. Entering "Termina" 
produces two options, for "Terminat" we get one and for "Terminat" again zero 
options. Only when the entire word "Terminator" is entered, the requested 
resources are displayed.

Such behavior is confusing and gives the impression that the requested resource 
does not exist.

Improve automatic disambiguation: extend rule "if the user input and the
label of the most relevant object are completely equal" to "if the user
input and the label of an object that has a 'count'-value not less than 50%
of the most relevant one are equal".

The current rule is a bit annoying as the user needs to disambiguate very
often.

Works only for cases where count-value is available (current virtuoso version).

(also see next issue)

It seems the latest beta version is adding relationships that are not on a
path between source nodes. It looks like relFinder is adding circular
relationships that involve at least one of the source nodes. 

Example: 

Assuming I have these relationships

A -> B -> C
A -> D -> A

And if I search for relationships between A and C, I end up with 'A -> D ->
A' in the results.

It should be possible to (de)select all elements in the last column of the
sortable table (i.e., all concepts, types, or lengths) at once. An eye icon
in the column's head should be okay for the moment.

The relationship length filter does not work properly. Currently, it is not
possible to deactivate relationships of length 0 - though the icon changes,
nothing happens.

While I am getting working paths between properties, it seems the filters
are not working as they should.

For example, I am filtering out
'http://www.w3.org/1999/02/22-rdf-syntax-ns#type' using
'<ignoredProperties>' in the configuration file, and yet, the resulting
graphs includes a '1.0#subject' node with many 'type' links pointing to it.

I will see if I can find a SPARQL query I can share (our content is behind
a firewall, so I cannot share it as is). 

What I can say though is that SPARQL queries that have a filter on 'type'
returns the following :

?pf2      ?middle        ?ps1
rdf:type  swivt:Subject  rdf:type 

even though they have this FILTER statement :

&& (?pf2 != <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ) 

I am testing relfinder on a local virtuoso sparql endpoint. It fails on the 
initial autocomplete search. 


In the loaded config I use :

      <autocompleteURIs>
        <autocompleteURI>http://www.w3.org/2000/01/rdf-schema#label</autocompleteURI>
        <autocompleteURI>http://xmlns.com/foaf/0.1/Person</autocompleteURI> 
        <autocompleteURI>http://rdvocab.info/uri/schema/FRBRentitiesRDA/Work</autocompleteURI> 
      </autocompleteURIs>

In the error log I extract the following SPARQL query :

SELECT ?s ?l count(?s) as ?count WHERE 
         { ?someobj ?p ?s . 
            ?s <http://www.w3.org/2000/01/rdf-schema#label> ?l UNION 
            { ?s <http://xmlns.com/foaf/0.1/Person> ?l } 
              UNION 
            { ?s <http://rdvocab.info/uri/schema/FRBRentitiesRDA/Work> ?l } . 
          ?l bif:contains "'search'" . 
          FILTER (!regex(str(?s), '^http://dbpedia.org/resource/Category:')). 
          FILTER (!regex(str(?s), '^http://dbpedia.org/resource/List')). 
          FILTER (!regex(str(?s), '^http://sw.opencyc.org/')). 
          FILTER (!isLiteral(?someobj)). } 
ORDER BY DESC(?count) LIMIT 20

It seems to be missing a {} around the 
http://www.w3.org/2000/01/rdf-schema#label triple ... or am I missing something 
? 

We should display the total number of found relationships somewhere to give
the user a feeling on how many relationships exist. We could, for instance,
insert a counter above the filter tabs.

elliptical arrangement for input nodes would better than radial arrangement

Thanks for adding a configuration file for examples !

However, there is a bug at the moment.

As said in the title, no examples are showing if the example.xml file 
contains only one example.

Everything is fine if it contains two or more examples.

http://relfinder.dbpedia.org/crossdomain.xml is needed because of new
address http://relfinder.semanticweb.org/

Our content is managed in a Semantic MediaWiki, with a local triple store
connector.

A consequence of this is that URIs for properties include a prefix
'Property-3A' (for the 'Property:' namescape in SMW).

Readability of property labels would be improved a lot if this prefix was
automatically detected and removed when displaying property labels on the
graph.

Right now, all labels are displayed as 'Property-3' because of the
truncated string from the property name.

Hi, there was a post on the apache stanbol list, which provides a test case. I 
posted it below and attached the screenshot. The triples are missing from the 
RDF, I checked this, but I didn't reproduce the error. 
All the best,
Sebastian



Hi Sebastian, 

I finally managed some more time to work on this. I've run into good and bad 
news: the good news is that sometimes RelFinder will find non-trivial paths 
(something more complicated than "a and b both have some relationship to c") 
but the results are somewhat wonky and I don't think it finds all of them that 
it should. See the end of this email for the example RDF data that I'm using.

    This query correctly returns document-1 when executed on the Stanbol SPARQL endpoint:
SELECT * WHERE {
?middle ?pf1 <http://some.org/people/person-1>  . 
?middle ?ps1 <http://some.org/people/person-2>  . 
} LIMIT 10

    And the following query correctly returns document-2
SELECT * WHERE {
?middle ?pf1 <http://some.org/people/person-2>  . 
?middle ?ps1 <http://some.org/people/person-3>  . 
} LIMIT 10

   The following query also works, however, I don't see it as one of the Relfinder queries specified in [1]:
SELECT * WHERE {
?middle ?pf1 <http://some.org/people/person-1> . 
?os1 ?ps1 <http://some.org/people/person-3> . 
?os1 ?ps2 ?os2 . 
?middle ?ps3 ?os2 . 
} LIMIT 10

Should the above query be included in Relfinder for max relation lengths > 4, 
and if so, any tips on how to modify relfinder? As it stands, the RDF DATA 
VERSION 1 does have a path connecting persons 1 and 3, but Relfinder doesn't 
find it.

Also, if I modify the rdf data to include inverse author properties (see RDF 
DATA VERSION 2) then RelFinder correctly finds the path through the document 
from person 1 to person 3, but starts adding spurious links when I search 
between person 1 and person 2. For instance, it includes author and creator 
arrows going both ways, like document 1 -> creator -> person2 AND document1 -> 
author -> person 2. The last statement isn't correct and isn't included in the 
RDF. I've attached a screenshot.

[1] 
http://dbpedia.aksw.org/dbpedia_demo/dbpedia/tutorials/relfinder_search/demo_rel
finder.php

RDF DATA VERSION 1:
<?xml version="1.0" encoding="UTF-8"?>
<rdf:RDF
    xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"
    xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
  xmlns:dcterms="http://purl.org/dc/terms/">

<rdf:Description rdf:about="http://some.org/people/person-2">
    <rdf:type rdf:resource="http://xmlns.com/foaf/0.1/Person"/>
    <rdfs:label>Person 2</rdfs:label>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/people/person-3">
    <rdf:type rdf:resource="http://xmlns.com/foaf/0.1/Person"/>
    <rdfs:label>Person 3</rdfs:label>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/people/person-1">
    <rdf:type rdf:resource="http://xmlns.com/foaf/0.1/Person"/>
    <rdfs:label>Person 1</rdfs:label>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/documents/document-1">
    <dcterms:creator rdf:resource="http://some.org/people/person-1"/>
    <rdfs:label>Document 1</rdfs:label>
    <dcterms:creator rdf:resource="http://some.org/people/person-2"/>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/documents/document-2">
    <dcterms:creator rdf:resource="http://some.org/people/person-3"/>
    <rdfs:label>Document 2</rdfs:label>
    <dcterms:creator rdf:resource="http://some.org/people/person-2"/>
</rdf:Description>

</rdf:RDF>

RDF DATA VERSION 2
<?xml version="1.0" encoding="UTF-8"?>
<rdf:RDF
    xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"
    xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
    xmlns:dcterms="http://purl.org/dc/terms/">

<rdf:Description rdf:about="http://some.org/people/person-2">
    <rdf:type rdf:resource="http://xmlns.com/foaf/0.1/Person"/>
    <dcterms:author rdf:resource="http://some.org/documents/document-1"/>
    <dcterms:author rdf:resource="http://some.org/documents/document-2"/>
    <rdfs:label>Person 2</rdfs:label>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/people/person-3">
    <rdf:type rdf:resource="http://xmlns.com/foaf/0.1/Person"/>
    <dcterms:author rdf:resource="http://some.org/documents/document-3"/> 
    <rdfs:label>Person 3</rdfs:label>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/people/person-1">
    <dcterms:author rdf:resource="http://some.org/documents/document-1"/> 
    <rdf:type rdf:resource="http://xmlns.com/foaf/0.1/Person"/>
    <rdfs:label>Person 1</rdfs:label>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/documents/document-1">
    <dcterms:creator rdf:resource="http://some.org/people/person-1"/>
    <rdfs:label>Document 1</rdfs:label>
    <dcterms:creator rdf:resource="http://some.org/people/person-2"/>
</rdf:Description>

<rdf:Description rdf:about="http://some.org/documents/document-2">
    <dcterms:creator rdf:resource="http://some.org/people/person-3"/>
    <rdfs:label>Document 2</rdfs:label>
    <dcterms:creator rdf:resource="http://some.org/people/person-2"/>
</rdf:Description>

</rdf:RDF>

Add a new option to the config to switch between POST and GET for SPARQL
requests

What steps will reproduce the problem?
1. create 3 endpoints in my configs;
2. save it;
3. load it;

What is the expected output? What do you see instead?

expect 3 endpints, but i saw only 1 incorrect endpoint

What version of the product are you using? On what operating system?

the demo

Please provide any additional information below.

Loading: problems with more than two start objects

Saving: must be extended to work with new config properties

Additional objects should only be added by using additional input fields
but not by replacing objects in the first two input fields. Thus, every
object has its own 'input' field and can be removed at any time.

The upper panel of the sidebar should be resizeable such as the other
panels, since the filters and infobox are hardly visible if many objects
are given by the user (and screen resolution is low).

can not parse SPARQL results with namespaces like

<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
xmlns:rs="http://www.w3.org/2005/sparql-results#"
xmlns:xsd="http://www.w3.org/2001/XMLSchema#">
 <rs:results rdf:nodeID="rset">
   <rs:result rdf:nodeID="sol193">
     <rs:binding rdf:nodeID="sol193-0" rs:name="s">
       <rs:value
rdf:resource="http://catalogus-professorum.org/lipsiensis/Grimmer_660"/>
     </rs:binding>
     <rs:binding rdf:nodeID="sol193-1" rs:name="l">
       <rs:value>Andreas Eduard Otto Horst Grimmer</rs:value>
     </rs:binding>
   </rs:result> 

It would be nice if known URIs could be directly entered into the input
boxes. 'Even nicer' would be a solution that allows URIs to be send
directly to the RelFinder (together with the ID of the dataset) and to get
immediately visualized.

I am still trying to find out why I am getting errors with some D2R 
endpoints we have locally.

One issue I noticed is that on two of our D2R SPARQL endpoints, RelFinder 
tries to force a '/sparql' at the end of the URL of the endpoint.

Example:

If I have 'http://servername:port' as the main URL, and if I 
use 'http://servername.port/sparql' as the URL of my endpoint, RelFinder 
tries to contact 'http://servername.port/sparql/sparql' in its queries.

Providing the URL without 'sparql' seems to work, however, it is unsafe to 
assume all endpoints will be named with '/sparql' at the end of their URLs.

Autocomplete queries for SPARQL-Server have a predefined language filter
"FILTER (lang(?l) = 'en')"
This should be removed to avoid problems with Virtuoso Servers without a
language attribute

The RelFinder sourcecode compiled with Flex 3.4 results in a swf file that
cannot produce any graph visualization.
Thus you have to use Flex 3.3!!

IgnoredProperties "+" and "-" buttons don't work in the endpoint config
dialog window

I am getting an HTTP 404 error on autocompletion with a D2R server.

The error message is :

> HTTP ERROR: 400
> This service does not allow the dataset to be specified in the protocol
request
> RequestURI=/sparql
> Powered by Jetty://

The type of SPARQL query RelFinder is sending to the D2R SPARQL endpoint
looks like this :

SELECT ?s ?l WHERE { { ?s <http://www.w3.org/2000/01/rdf-schema#label> ?l }
. FILTER regex(?l, 'POM', 'i') . } LIMIT 20

I tried to run that query from a desktop SPARQL client (Twinkle) and I was
able to get a result set. With the same query, RelFinder returns the error
message.

At this point, I am thinking the error is coming from relFinder but I still
need to probe things a little more before I can tell for sure.

Our D2R server is : Jetty(6.1.10)

It would be nice if basic infos about a disambiguated object are shown in
the infobox also in cases where no relationships are found. Currently, the
user never knows for sure if the correct object was matched on his input.
He just selects an object from the list and sees its URI as tooltip - but
no further information. A function that also shows the abstract, picture
and URL in the infobox after the user has clicked on the button in the
input box (that with the object label) would be nice. The button might also
be used for redisambiguation (e.g., after false automatical matches --> see
issue 11). 

If you save the current graph using the URL button, the data source 
appears twice in the list of data sources the next time you click on that 
URL.

The second copy appears at the bottom of the list. Probably a side effect 
of including the data source in the URL.

When an input field (i.e., object) is removed, the node should also be
removed from the graph.

If the user hits the 'add' button, text that she has already entered into
the input fields should not be deleted.

What steps will reproduce the problem?
1. input a chinese string in the input box, e.g. 张三

What is the expected output? What do you see instead?

the string should be displayed normally in the inputbox, and autocompletition 
works

What version of the product are you using? On what operating system?

the demo

Please provide any additional information below.

The error message "Error. Please check SPARQL configiration" appears in
every case, an entered term has no match in the dataset. That's not correct.

Infobox doesn't load Images from URLs with PHP

Customization of RelFinder to a local environment would be greatly improved
if the default datas source could be changed from DBpedia (possibly to the
first data sources found in the configuration file ?).

Similarly, XML configuration tags to set examples for each data source
would help a lot (or a separate XML file for examples if it is more
appropriate for the code).

The pick & pin feature is not working when a node is first hidden (e.g.,
via filtering) and then shown again. Needle symbol doesn't appear in this case.

the "search more..." and seperator entry are added before the list will be
sortet by similarity.

enable possibility to resort the examples in the further disambiguation
popup. (switch sorting between popularity / similarity)

The numbers in the "#num" column of the sortable table are currently
ordered by the visible number, not the total one. This results in a
reordering of the column every time a property is deactivated.

An input field where no input has been entered should just disappear when
'find relations' is executed (currently, the user is prompted for a
disambiguation). Of course, two input fields must remain at any time.

One potentially useful enhancement would be to allow a filter or constraint
on the directionality of paths found.

Namely :

- from A to B only
- from B to A only
- or a mixed path between A and B such as ' A -> C <- B'

Just heard Philipp's talk at ESWC.  This is no bug report but a suggestion 
for future work.  We might actually consider discussing this on the LOD 
mailing list, as it has a wider impact.  If the linked datasets that you are 
using were writable (they are not currently, but I hope they will soon 
become), how about the following: When RelFinder has discovered an indirect 
relation between two resources, how about assisting the user in establishing 
a direct link between them?  E.g. suppose RelFinder finds A -> B1 -> ... -> 
Bn -> C, one could ask the user to enter a name for relations R and R^-1 
such that A -R-> C and C -R^-1-> A.  Then, via e.g. SPARQL Update, those new 
triples could be contributed to the datasets where A and C come from.

"check config" is shown instead of "no relation found" in autocomplete

There is partly HTML syntax in the infobox abstract, e.g. "–" instead
of "-" in case of 'David Hume'.

"is Virtuoso" is alway true for loaded configs

I am missing terms in the autocompletion list. I understand there is a
limit to what autocompletion is returning to preserve performance, but
there is no way to use a node that does exist but is not returned by
autocompletion.

For example, if I have a page called 'Chemistry' and I type 'Chemistry' in
the search form, autocompletion is giving me a list of pages that include
'Chemistry' in the title but it is missing the 'Chemistry' page itself. 

And since there is no way to submit the form without selecting from
autocompletion, I cannot use that page as a starting node.

RelFinder should provide a way to enter a term and validate it exists
without having to go through autocompletion.

I saved two searches with RelFinder using the 'URL' button.

One search against LODD and one search against our local Semantic Web
triple store (RAP).

Clicking on the saved URL from LODD opened RelFinder and launched 'Find
relations' automatically.

Clicking on the saved URL from RAP opened RelFinder in a blank state. I had
to manually press 'Find Relations' to run the search.

Examples should not be hard coded.

Is there a way to specify which properties to look for and use as
'abstracts' for nodes ?

DBpedia has two properties at least (abstract and comment). 

Other data sources may have other properties with the same function (ours
has 'Has description' or 'Has summary' for example). 

It would be nice to have XML configuration tags to indicate which
properties RelFinder should use in the preview window for a node (in the
same way Filtered properties can be configured).