This article describes in-depth research on machine learning methods for sentiment analysis of Czech social media. Whereas in English, Chinese, or Spanish this field has a long history and evaluation datasets for various domains are widely available, in the case of the Czech language no systematic research has yet been conducted. We tackle this issue and establish a common ground for further re...

In this paper, we propose a new computationally efficient framework for learning sparse models. We formulate a unified approach that contains as particular cases models promoting sparse synthesis and analysis type of priors, and mixtures thereof. The supervised training of the proposed model is formulated as a bilevel optimization problem, in which the operators are optimized to achieve the bes...

We leverage vector space embeddings of sentences and nearest-neighbor methods to transform a small amount of labelled training data into a significantly larger training set using an unlabelled corpus. The quality of the larger training set is measured by prediction accuracy on a benchmark sentiment analysis task. Our results indicate it is possible to achieve accuracy within 3-5% of the baselin...

The aim of this project was to use semi-supervised recursive autoencoder provided by [2] and classify the english phrases from movie reviews into five sentiment classes; very positive, positive, neutral, negative and very negative by softmax regression classifier.

In this paper we present a semi-supervised sub-manifold discriminant analysis algorithm. To separate each sub-manifold constructed by each class, we define the within-manifold scatter, between-manifold scatter and total-manifold scatter matrices. The scatter matrices are robust to outlier and diverse-density clusters. Kernelization and direct non-linear embedding are also developed. Experimenta...

Linear discriminant analysis (LDA) is commonly used for dimensionality reduction. In real-world applications where labeled data are scarce, LDA does not work very well. However, unlabeled data are often available in large quantities. We propose a novel semi-supervised discriminant analysis algorithm called SSDACCCP . We utilize unlabeled data to maximize an optimality criterion of LDA and use t...

This paper describes an algorithm that is an extension of mixture-modelling to supervised clustering. It is demonstrated to be as accurate as current state-of-the-art machine learning algorithms across various data sets, and significantly more accurate than distance-based supervised clustering algorithms. Most significantly, it combines the classification itself with the calculation of rich inf...

Feature selection is an important task in effective data mining. A new challenge to feature selection is the socalled “small labeled-sample problem” in which labeled data is small and unlabeled data is large. The paucity of labeled instances provides insufficient information about the structure of the target concept, and can cause supervised feature selection algorithms to fail. Unsupervised fe...

We investigate a novel approach for intuitive interaction with a data set for explorative data analysis. The key idea is that a user can directly interact with a two or three dimensional embedding of the data and actively place data points to desired locations. To achieve this, we propose a variant of semisupervised kernel PCA which respects the placement of control points and maximizes the var...

In this paper, we propose a learning-based supervised discrete hashing method. Binary hashing is widely used for large-scale image retrieval as well as video and document searches because the compact representation of binary code is essential for data storage and reasonable for query searches using bit-operations. The recently proposed Supervised Discrete Hashing (SDH) efficiently solves mixed-...