Semantic Analytics Group @ Uniroma2

The SAG team (Castellucci, Croce, Basili) participated to the Evalita Sentiment Analysis in Twitter in Italian challenge (Sentipolc 2016) with a Deep Learning architecture based on Convolutional Neural Networks.

The SAG systems, named unitor, achieved the first and the second position among 22 systems in the Subjectivity Detection task. In the Polarity Classification task SAG systems placed second and third among 28 systems.

Details about the systems and the results will be provided in the Evalita 2016 proceedings. In the meanwhile, a very similar architecture is described in “Injecting sentiment information in context-aware convolutional neural networks” (Croce et Al, 2016), available in the SocialNLP 2016 Proceedings.

A new version of KeLP has been released!

In addition to minor bug fixes, this release includes:

– a flexible system to manipulate example-pairs
– new manipulators for performing tree pruning
– new examples for the usage of kelp (package kelp-full)

Check out this new version from our repositories. API Javadoc is already available. Your suggestions will be very precious for us, so download and try KeLP 2.1.0!

A new version of the Java Machine Learning Framework KeLP has been released.

This new release (2.0.2) includes:

• the Nystrom method for large scale kernel learning, following the methodology reported in the recently accepted ECIR 2016 paper.
• New examples for the usage of the Smoothed Partial Tree Kernel and the Compositionally Smoothed Partial Tree Kernel.
• Minor bug fixes.

Check out this new version from our repositories. API Javadoc is already available. Your suggestions will be very precious for us, so download and try KeLP 2.0.2!

A new version of the Java Machine Learning Framework KeLP has been released.

This new release (2.0.1) includes:

• Soft Confidence Weighted Classification algorithm: a brand new online learning algorithm from Wang, J., Zhao, P., Hoi, S.C.: Exact soft confidence-weighted learning. In Proceedings of the ICML 2012. ACM, New York, NY, USA (2012)
• Optimization of the kernel caching mechanism
• The Smooth Partial Tree Kernel and the Partial Tree Kernel now have the possibility to specify a maximum branching factor (parameter: maxSubseqLeng) in the tree fragments considered by the kernel operation.

and minor big fixes.

Check out this new version from our repositories. API Javadoc is already available. Your suggestions will be very precious for us, so download and try KeLP 2.0.1!

The 2.0.0 version of KeLP has been released. This is a major release that includes brand new features as well as a renewed architecture of the entire project.

Now KeLP is organized in four maven projects:

1. kelp-core: it contains the infrastructure of abstract classes and interfaces to work with KeLP. Furthermore, some implementations of algorithms, kernels and representations are included, to provide a base operative environment.
2. kelp-additional-kernels: it contains several kernel functions that extend the set of kernels made available in the kelp-core project.
3. kelp-additional-algorithms: it contains several learning algorithms extending the set of algorithms provided in the kelp-core project.
4. kelp-full: this is the complete package of KeLP. It aggregates the previous modules in one jar. It contains also a set of fully functioning examples showing how to implement a learning system with KeLP. Batch learning algorithm as well as Online Learning algorithms usage is shown here. Different examples cover the usage of standard kernels, Tree Kernels and Graph Kernels, with caching mechanisms.

Moreover, this new release includes consistency check methods during the population of a Dataset object and:

CsvDatasetReader: it allows to read files in CSV format

DCDLearningAlgorithm: it is the implementation of the Dual Coordinate Descent learning algorithm.

Check out this new version from our repositories. API Javadoc is already available. Your suggestions will be very precious for us, so download and try KeLP 2.0.0!

The paper “A Graph-based Model of Contextual Information in Sentiment Analysis over Twitter” by Giuseppe Castellucci, Danilo Croce and Roberto Basili has been accepted at the 2nd Italian Conference on Computational Linguistics!

Abstract: Analyzing the sentiment expressed by short messages available in Social Media is challenging as the information when considering an instance is scarce. A fundamental role is played by Contextual information that is available when interpreting a message. In this paper, a Graph-based method is applied: a graph is built containing the contextual information needed to model complex interactions between messages. A Label Propagation algorithm is adopted to spread polarity information from known polarized nodes to the others.

A new version (1.2.1) of the Java Machine Learning platform KeLP developed within the SAG group has been released.

In this releases, the code for learning relations between pairs of short texts has been introduced, including the approach described in:

Simone Filice, Giovanni Da San Martino and Alessandro Moschitti. Relational Information for Learning from Structured Text Pairs. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics, ACL 2015.

In particular this new release includes:

• TreePairRelTagger: a manipulator that establishes relations between two tree representations (available in the maven project discreterepresentation)
• 5 new kernels on pairs: released in the maven project standard-kernel

Check out this new version from our repositories. API Javadoc is already available. Your suggestions will be very precious for us, so download and try KeLP 1.2.1!

The proceedings of the 6th Italian Information Retrieval Workshop are online. You can download all the workshop papers from the CEUR IIR 2015 system.

Check out our paper “A Compositional Perspective in Convolution Kernels” by Roberto Basili, Paolo Annesi, Giuseppe Castellucci, Danilo Croce.

A new version (1.2.0) of the Java Machine Learning platform KeLP developed within the SAG group has been released.

Two brand new subprojects have been added:

• graph-representation: it contains DirectedGraphRepresentation for representing direct unweighted graphs
• graph-kernel: it contains some state-of-the-art graph kernels, like the shortest path kernel and the Weisfeiler-Lehman Subtree Kernel for Graphs

Furthermore the following components are added:

• StandardizerManipulator: for standardizing the features of the vectors in a dataset
• KernelMultiplication: a new kernel for combining other kernels applying the multiplication operator
• ExperimentUtils: a class providing useful methods for performing experiments, like a n-fold cross validation
• LibsvmDatasetReader: for reading files in LibSVM or LibLinear or SvmLight format

Moreover several demo examples are added in kelp-examples as well as some unit test in kelp-full.

Check out this new version from our repositories. API Javadoc is already available. Your suggestions will be very precious for us, so download and try KeLP 1.2.0!

A new version (1.1.1) of the Java Machine Learning platform KeLP developed within the SAG group has been released.

This is a minor release that fixes some bugs.

Check out this new version from our repositories.

Your suggestions will be very precious for us, so download and try KeLP!