We could publish the command-line applications (batch processing, REST/XML-RPC clustering server, benchmarking) in the OS repository.

Issue: CARROT-11 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Apr 27 2007

The cluster tree in the carrot2-demo-webapp shows inconsistent information about cluster sizes. The root node (All results) eports the 'true' number of documents, while the other nodes report the number of document assignments.

Issue: CARROT-1 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved May 13 2007

In the current version of the Demo Browser, users must configure the Lucene input component every time the application is restarted, which is annoying. It would be a good idea to either let the user save the configuration explicitly, or simply save the configuration in background and restore it when the Demo Browser is launched again (making the process totally transparent for the user).

Issue: CARROT-40 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Mar 14 2007

This nasty bug caused a number of weird errors, e.g. not all documents being displayed in the tuning browser.

Issue: CARROT-56 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Apr 12 2007

This is a result of a heuristic UI evaluation. We might consider adding "tips of the day" for advanced users about e.g.: keyboard shortcuts, how to change clustering algorithms, how to change the results count. The tips might be added on the startup page, with an option to disable them (stored in a persistent cookie).

Issue: CARROT-2 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), updated Mar 20 2012

We could extend the "Search with Carrot2" section to add options similar to those on Clusty: http://clusty.com/sitesearch. Should we also add information about the possibility to link directly to the iframes with the results?

Issue: CARROT-10 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), updated Mar 20 2012

When running the demo browser, JDIC is doing a lot of annoying logging. It dups to the console something along the lines of:

native lib path F:\tmp\br\deps-carrot2-demo-browser-jar*** Error: nspr4.dll doesn't exist under F:\tmp\br\deps-carrot2-demo-browser-jar\ielib
native lib path F:\tmp\br\deps-carrot2-demo-browser-jar

Additionally, it creates a JDIC.log file in the demo browser's main directory.

Is there any way to get rid of this kind of logging?

Issue: CARROT-52 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Apr 03 2007

This could be an interesting source of statistics, potentially useful for improving/ promoting certain clusters in the future. Implementation idea: when the user clicks on a cluster label (for the first time only?), make a background XmlHTTPRequest to send to the server: the query, the cluster label (+ parent labels?), the rank of the cluster?, the number of documents in the cluster? An even more interesting feature would be to track clicks on documents (a cluster whose documents have not been clicked is worse than a cluster whose documents have been clicked).

Issue: CARROT-13 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), updated Mar 20 2012

Microsoft has released v1.0 of their Live Search API:

http://msdn2.microsoft.com/en-us/library/bb251794.aspx

It might be a good idea to see what's new and add new features to Carrot2's MSN input component.

Issue: CARROT-24 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 02 2007

https://hudson.dev.java.net/

Issue: CARROT-51 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Jul 03 2007

When browsing search results for which 'singleton' clusters have been generated (e.g. by the Lingo algorithm), contents of these clusters is not shown in the demo web application. Example query:

http://demo.carrot2.org/demo-head/search?in=Web&q=clinton&s=100&alg=Lingo&opts=h

Issue: CARROT-46 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Mar 14 2007

Currently, the number of input tabs we can have in the webapp is limited by the width of the tabs space – it's not possible to have more than about 7-8 tabs because they wouldn't fit on the screen.

One idea to solve this problem is to have a "more..." tab, shown always at the rightmost position, which would show all the inputs emphasizing those that are currently not available on tabs. After the user clicks one of such inputs, the selected input would replace the rightmost regular tab (not the "more..." tab), the new tab would be automatically activated and the cursor placed in the query text field.

In a more sophisticated scenario, each input on the "more..." tab could have a combo box. Entries with the combo box checked would appear as tabs. In this way, users could have as many tabs as they can fit on their screens. One problem with this is the order/ordering of tabs – alphabetical?

Another important factor is to remember user preferences (last added tab, set of selected tabs depending on the scenario) in a persistent cookie.

Issue: CARROT-8 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 20 2007

Occasionally it happens that search results contain snippets in different languages, which, combined with inaccurate language recognition, can lead to meaningless cluster labels consisting of e.g. stop words. Applying all known stop lists (and not only the one for the recognized language) would fix the problem. Stop word list merging could be implemented on the level of the tokenizer component, and in this way all clustering algorithms using the component would benefit.

Issue: CARROT-37 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Mar 15 2007

As in the title.

Issue: CARROT-22 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Jul 09 2007
Linked issues:

• #227

SOLR (http://lucene.apache.org/solr/) is gaining more and more momentum, so it might be a good idea to add an input component for it. The component is very easy to create – it's essentially a subclass of XmlInputComponent with a specific XSLT file.

Issue: CARROT-17 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Feb 20 2007

Add a "Show in clusters" feature to the webapp, which will highlight all clusters in the tree to which the selected (hovered-on?) document belongs.

Issue: CARROT-7 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Feb 16 2007

Carrot2 does not build with ANT 1.7.0 due to changes in the ANT internals. The Path class, which Carrot2 uses to inherit is dependencies-related taskdefs, has changed the API internally and uses ResourceCollection interface methods rather than String [] list() method.

A fix for this should not be difficult, but it also means that all the development must be moved to ANT 1.7.0 (creating a compatible version is unlikely).

Issue: CARROT-27 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Feb 21 2007
Attachments: ant170.patch

This suggestion is copied from doc/roadmap-2.1txt. Dawid, please check if it's still valid.

Issue: CARROT-6 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 14 2007

The current WebApp emits element (DIV) identifiers that are integers – this is incorrect according to the spec (validator) because all ID elements must begin with a letter.

Issue: CARROT-23 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Jun 29 2007

eTools (http://www.etools.ch) makes available their search results in a form of an XML feed that Carrot2 could use. Therefore, an input component for eTools would be very useful.

Issue: CARROT-36 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Mar 05 2007

Some users may prefer that all clusters are shown at once, even without clicking the "all clusters" link. To support such users the cluster list could behave in the following way: after the user clicks the "all clusters" (#232) link, at the bottom of the cluster list another link appears (e.g. "always show all clusters"). After the user clicks that link, for all further searches all clusters will appear right away. When immediate showing of all clusters is enabled, at the bottom of the cluster list another link should be visible (e.g. "don't show all clusters right away"), which the user can click to restore the original behaviour (i.e. showinc clusters in groups).

Issue: CARROT-21 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 14 2007

It might be nice feature for the OpenSearch input to be able to configure itself entirely based on the descriptor. Then, the only parameter required by the OpenSearch input would be an URL to the descriptor.

Issue: CARROT-15 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jul 14 2020
Linked issues:

• #234

It looks like the Demo browser shows snippets in clusters is a more or less random order :) Had a brief look at the code, but I guess this needs some more serious debugging.

Issue: CARROT-44 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Mar 12 2007

It would be nice to allow the users to display all clusters in one click rather than by multiple clicks on the "more..." links. To implement this, next to the "more..." link, "all clusters" link should be rendered that reveals all clusters at one time time.

Issue: CARROT-20 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Feb 15 2007

Lingo currently uses its own language processing routines. I think it would make sense if we could reuse the rest
of the framework's infrastructure for this (similarly to STC and other components). Process descriptors would then
become similar with the rest of the framework.

Issue: CARROT-18 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Jul 20 2008

It might be interesting to present clustering results in a tree map similarly to e.g. http://google.hivegroup.com/

Issue: CARROT-34 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), updated Mar 20 2012

Check out Unicode segmentation spec (info from Andrzej Zydroń):

http://www.unicode.org/reports/tr29/tr29-10.html

May be useful for tokenization of certain languages (?).

Issue: CARROT-19 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Jul 09 2007

Update JDIC libraries to the newest available versions. Check if the browser application works.

Issue: CARROT-5 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Mar 09 2007
Linked issues:

• #248

When this is added as an input component for the browser:

/carrot2-lingo-3g-browser/components/input-cached-lucene.bsh:

import com.dawidweiss.carrot.core.local.*;
import com.dawidweiss.carrot.core.local.impl.*;
import com.dawidweiss.carrot.core.local.clustering.*;
import com.dawidweiss.carrot.local.controller.*;
import com.carrot.input.lucene.*;
import carrot2.demo.cache.*;

/**

• Provide a path to the Lucene index to be used below, or use the "lucene.index.path"
• system property to do that.
  */
  String indexPath = System.getProperty("lucene.index.path");

LocalComponentFactory factory = null;

if (indexPath != null)
{
luceneFactory = new LuceneLocalInputComponentFactory(indexPath);

factory = new LocalComponentFactoryBase()
{
    public LocalComponent getInstance()
    {
        return new RawDocumentProducerCacheWrapper(luceneFactory.getInstance());
   }
};

}
else
{
RawDocument errorMessage = new RawDocumentSnippet(
"0",
"Please configure Lucene index first",
"To use Lucene input, please provide a path to a Lucene index in the " + "'lucene.index.path' system property.",
"http://www.carrot2.org", // This url could point to some more material about configuring Lucene index
(float)0.0
);

factory = new LocalComponentFactoryBase() {
public LocalComponent getInstance() {
return new RawDocumentProducerCacheWrapper(new
com.dawidweiss.carrot.core.local.impl.DummyLocalInputComponent(errorMessage,
"Please configure Lucene index"));
}
};
}

return new LoadedComponentFactory(
/* id */ "input-cached-lucene",factory);

we get a weird exception:

Exception message: java.lang.NoClassDefFoundError: A class required by class: global/BlockNameSpace$4 could not be loaded: java.lang.NoClassDefFoundError: IllegalName: global/BlockNameSpace$4

Issue: CARROT-3 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 14 2007

When running the HEAD demo browser from WebStart the HTML control (for showing search results) is not visible at all. When running the STABLE demo browser from WebStart in the same environment the Swing HTML control is used instead of JDIC (is this normal?).

OS: Windows XP Professional
JDK: Sun JRE 1.6.0 (not sure for previous JREs)
Build: HEAD

Issue: CARROT-38 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Mar 09 2007
Linked issues:

• #218

STC does not produce a "junk" cluster from unassigned documents.

Issue: CARROT-30 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Feb 24 2007

From time to time Lingo happens to produce clusters with literally the same labels.

Issue: CARROT-47 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 29 2007
Attachments: repeated-labels.xml

Implemented in trunk. When you specify a global ant property:

ant -Dcarrot2.XXX=value

the property is available in test runs. The prefix "carrot2." is obligatory.

Issue: CARROT-48 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Mar 20 2007

Issue: CARROT-53 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Apr 02 2007

Implement JSON serialization in the Web application to allow interaction with a bunch of scripting languages.

Issue: CARROT-28 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Feb 22 2007

This is a bit of a misleading API – when passing an instance of LuceneSearchConfig, its searcher and analyzer fields can/should be null, as the component factory will create them based on the path and analyzer factory. The problem is that the config class is used both when creating the component factory (where searcher and analyzer fields are irrelevant) and at runtime (where these fields are needed). Therefore, it would be nice to refactor that into two classes – one for use with the constructor and another one for internal use.

Issue: CARROT-45 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 28 2007

Create a tag-cloud visualization for the retrieved clusters? Perhaps some tuning of algorithms would be required to make the cluster labels more compact.

Issue: CARROT-32 (migrated from JIRA), created by Dawid Weiss (@dweiss), updated Mar 20 2012

Issue: CARROT-49 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Apr 03 2007
Linked issues:

• #262

As the Lucene input component supports extraction of snippets, we could add the appropriate option to the configuration dialog in the demo browser.

Issue: CARROT-12 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 28 2007

We could then e.g. integrate different web search tabs into one "Web".

Issue: CARROT-9 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 28 2007

Looks interesting, perhaps we could use it instead of our in-house build dependency resolver?
http://incubator.apache.org/ivy/index.html

Issue: CARROT-31 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Jul 27 2008

One of the missing features of Carrot2's web application is the spell checking functionality allowing the user to correct the spelling mistakes in their queries. One interesting way of solving this problem (on the client side) could be similar to: http://spellify.com/

Issue: CARROT-35 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), updated Mar 20 2012

It might be a good idea to track searches issued from the FireFox plugin. We could e.g. add some extra HTTP parameter and then put it in the logs and handle it in GA.

Issue: CARROT-4 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 14 2007

When preparing customized versions of the Demo Browser, it often happens that specialized processes need to be added. One way of managing this could be adding all possible input component and the corresponding process descriptors, but marking the "non-standard" processes as "inactive" (with a dedicated property). In this way, those who would like to try the customized processes could simply activate them.

One possible problem with this approach is that there is no easy way to "deactivate" (stop the controller from loading) component factories, which, in case of factories that need some specific configuration, might cause initialization exceptions.

Issue: CARROT-43 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jul 13 2007

It might be a good idea to put a link to the Firefox search plugin in the webapp (next to About, Contact etc.). In this way advanced users will be able to make Carrot2 queries much faster.

Issue: CARROT-33 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Jun 28 2007

When starting up the browser on Java 1.6, the splash screen shows up as usual, but the actual browser window opens "in background" (it's not activated, but put under other windows). This can cause a lot of irritation: 1) every time the browser is launched, the user must take an extra action (click, alt+tab) to start working with it, 2) some users might think the browser doesn't work at all (because it doesn't appear on the screen, only on the task bar). Is this behavior intentional?

Issue: CARROT-55 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Apr 03 2007
Linked issues:

• #257

Add desired cluster phrase length customization parameters to STC. Currently these are hardcoded in:

private float calculateModifiedBaseClusterScore(final int effectivePhraseLength, final int documentCount) {
    final double SINGLE_WORD_BOOST = 0.5f;
    final int optimalPhraseLength = 3;
    final int optimalPhraseLengthDev = 2;

Issue: CARROT-29 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Mar 06 2007
Attachments: StcConstants.patch, STCEngine.patch, StcParameters.patch, StcSettingsDialog.patch

When the user selects a cluster, words of the label should be highlighted (e.g. in bold) in the documents shown for that cluster.

Issue: CARROT-16 (migrated from JIRA), created by Stanisław Osiński (@stanislawosinski), resolved Mar 14 2007

The webapp controller servlet should re-read configuration if initial failure occurred. One can then
correct descriptors and simply refresh the home page to reload the application. A possibility
of reloading everything dynamically should also be an option.

Another thing is more descriptive error messages if configuration loading fails for some reason.

Issue: CARROT-54 (migrated from JIRA), created by Dawid Weiss (@dweiss), resolved Apr 18 2007

HAOG-STC does not produce a "junk" cluster from unassigned documents.

Issue: CARROT-39 (migrated from JIRA), created by Andrew Steeley, resolved Mar 14 2007
Attachments: STCGraphRenderer.java.patch