Nicht nur in der Mailingliste (Apache Mahout Status by Grant Ingersoll) des Projekts nimmt der Traffic zu, auch erste Projekte die Mahout erfolgreich einsetzen werden bekanntgegeben.

So geschehen bei dem (mobilen) News-Aggregator und Organisator Mippin aus Großbritannien. Mippin sammelt Informationen über das Surf-Verhalten seiner Nutzer und erstellt profilbezogene Vorschläge Themen-ähnlicher Websites. D.h. je mehr ein Benutzer die Software “mitlesen” lässt und je mehr Benutzer Mippin benutzen, desto besser funktioniert das Vorschlagwesen – sowas nennt man dann Collaborative Filtering.

Mippin nutzt dafür das Recommendation System in Apache Mahout und bedient sich damit denselben Tricks wie Amazons Empfehlungs-System. Laut Sean Owen (Mipping & Committer bei Apache Mahout) soll mit Apache Mahout das Ziel erreicht werden, Machine Learning Algorithmen für alle Entwicklern zur Verfügung zu stellen, damit diese Disziplin keine Raketenwissenschaft bleibt.

Hier noch einige Quellen zum Thema Apache Mahout:

• InfoQ: Apache Mahout, Highly Scalable Machine Learning Algorithms
• Cyberstrategics: Running k-means Clustering
• Iletken Blog: Mahout Review (Kritisches Benchmark)

It is already two months since I was announcing the Nutch 1.0 release. Finally after a 1.0 release candidate Nutch was released in version 1.0 last week! Congratulations!

If you are also into Machine Learning you should check out the release of Mahout in its early version 0.3 and wether you are new on it, check out Grant Ingersoll’s presentation on Mahout from ApacheCon Europe 2009 here.

P.S.: Also notice the new Apache Droids Tab in the Lucene navigation bar!

Unfortunately I was not be able to attend at the ApacheCon US 2008 in New Orleans this year – way too far away from good ol’ germany! But I reviewed the given sessions on Lucene (and Solr, Mahout, Tika) afterwards the conference to get some inspiration. Some comments on it:

* Advanced Indexing Techniques with Apache Lucene (Michael Busch)

Detailled presentation about Indexing capabilities of Apache Lucene and a very interesting part on how to use Token Payloads and POS-Tagging with the new TokenStream API.

* Apache Solr: Out of the Box (Chris Hostetter)

Introduction to Solr from installation and administration (Admin Console, Luke), querying (Facets, Highlighting) and configuration (Analyzers, Multiple Indexes, Replication).

* Introducing Mahout: Apache Machine Learning (Grant Ingersoll)

Already posted this session in my recent Breakfast Links. A nice presentation about what Machine Learning stands for and the approach of Mahout.

* Apache Solr: Beyond the Box (Chris Hostetter)

Presentation about Solr’s history and real world examples such like Geo search.

* Content analysis for ECM with Apache Tika (Paolo Mottadelli)

Impressive and extensive presentation about Apache Tika and its Alfresco integration for content extraction.

Breakfast Links 10.11.2008

* Introducing Mahout: Apache Machine Learning

Grant Ingersoll posted his presentation about Mahout from last week on the ApacheCon US 2008. The presentation is in general about machine learning, different approaches, methods and the current status of Mahout. Check it out!

* Multi Core Chips and Search Performance

Interesting article on Search Performance in terms of scalability through hardware.

Breakfast Links 03.11.2008

* Exalead-Labs: voxalead

The parisian search engine company Exalead released their video search engine voxalead. It combines Exalead’s indexing (Entity Recognition) and search technology with third partie’s Voice recognition (LIMSI). Currently they indexed mostly french and us channels and a small amount of german channels. Beside the search through the voice-to-text extracted contents you can also view the original extracted text which shows off the quality of voxalead!

* Metager at Alt Search Engines

Metager now has its place at AltSearchEngines.com in english and german!

* Mahout: Twenty Newsgroups Classification

This weekend I stumbled upon a nice Mahout Classification example. It is all about creating a classification model based on twenty clusters of newsgroup messages. All you need is Mahout and Hadoop to get the example running in approx. 20 minutes.