This site contains datasets, applications, tools and other resources publicly available. Other resources can possibly be found on the archived site of the Knowledge Engineering Group.



EUR-Lex text collection

The EUR-Lex text collection provides a large multlabel classification benchmark with up to 4000 different classes.

Medical Concept Embeddings: code and data

Concept vector representations learned from a large labeled background corpus. These were used for computing the semantic similarity between terms from the medical domain.


DIP-SumEval: A Data Set of Human Summary Evaluations

The first data set of judgements of automatic multi-document summarization systems on large variety of quality dimensions. Contains over 400 automatically generated summaries for 49 topics of an data set for multi-document summarization, 1274 judgements according to 11 text and summary quality criteria on a Likert-scale (1 to 5) performed by 26 trained annotators, and 43218 pairwise judgements according to 6 criteria performed by 64 crowd-workers.

Software and Source Code


BOOMER: an algorithm for learning gradient boosted multi-label classification rules

Efficient and scalable scikit-learn implementation for learning gradient boosted multi-label classification rules.


NSS: Framework for Non-Specific Syndromic Surveillance

Software framework including state-of-the-art approaches, statistical baselines and an advanced approach based on sum-product networks.


SeCo for learning multi-label rules

Separate-and-conquer rule-learning framework for learning multi-label head rules.


Extreme Dynamic Classifier Chains

Dynamic classifier chains version of extreme gradient boosted trees (XGBoost) for multi-label classification



Dynamic classifier chains version of recurrent neural networks for predicting one by one, in a sequence, the labels in multi-label classification tasks.


Graded Multilabel Classification, code and new data sets

The code and data used for our paper about pairwise graded multilabel classification. In this setting, a label is not only present or absent, but can have several grades, e.g. stars.

TUD poker framework

Framework for testing end developing computer bots, such as counter-factual regret minimization or neural network based bots.


P³oodle: a personalizing, privacy-protecting browser add-on for searching the web

Personalized web site ranking with different techniques from IR on the own computer.



Learn continuous vector representations jointly for words, documents, and labels. Use corpora with labelled documents and use also descriptions of labels. This enables also to do zero-shot learning, i.e., to predict labels for which no documents were observed during training.


Classification GUI

A graphical user interface that allows to intuitively assign concepts from an ontology to a set of documents in order to quickly and easily develop a (multilabel) classification dataset.


A highly modular framework for the efficient Perceptron algorithm containing a great collection of effective extensions