Natural Language Processing and Sentiment Analysis

The communication protocol between humans is natural language. Humans communicate with each other and express their private states using language in the form of text or speech. Furthermore, language may be considered as the projection of human knowledge and intelligence. Hence, natural language is the right source of data to understand humans, to represent their intelligence and to communicate with them. However, natural language is unstructured data and it cannot be directly used by a computer system.

In this tutorial, we will describe how to represent natural language in order to be processed by a computer, or in other words, how to build features that represent natural language. We will introduce Deep Learning methods for the extraction of rich linguistic features from written utterances, not limited to superficial views of the text as a bag or sequence of words, but including the syntactic (structural) analysis of the text. Moreover, we will show the usefulness of these linguistic features for the development of opinion classification systems, which is crucial for Sentiment Analysis tasks. Likewise, we will focus on the development of polarity classification methods grounded in Deep Learning.

Deep Learning

In recent years, several types of neural networks, especially deep neural networks, commonly called Deep Learning (DL) models, have shown high capabilities of learning from data. The main reasons behind this success is i) the emergence of new large labeled databases, ii) the advances in GPU (Graphics Processor Unit) technology, which have made this device ideal for accelerating the mathematical operations involved in Neural Networks and (iii) advances in learning algorithms and optimization techniques.

This tutorial presents the main concepts of neural networks in general. In the theoretical part, we will present the most important concepts of the learning of these networks: Feedforward, gradient descent and regularization. In the hands-on lab part, we will show the functioning of the neural networks using two simple examples, one regression problem and one image classification problem with shallow neural networks.

Applying computer learning to detect malware in Android: AndroPyTool and OmniDroid

The possibilities and advantages of applying Computational Learning to solve a multitude of problems are beyond doubt. It has been demonstrated how this broad set of techniques can help solve problems in areas as diverse as computer vision, natural language processing, fraud detection, robotics and bioinformatics, among others. Our aim in this tutorial is to present the possibilities offered by this field when dealing with a current, complex problem and críıtico: the detection of malware on Android devices. As we will show in this tutorial, Computational Learning techniques such as classification and clustering algorithms, evolutionary computing, and deep learning are currently being used to detect those malware samples whose behavior shows malicious patterns. In addition, we will explain the different tools developed to analyze Android malware and perform reverse engineering. Finally, we will first describe our software AndroPyTool, designed to extract a wide range of features from Android applications in order to deeply characterize their behavior. Secondly, we will talk about Omnidroid, a wide set of data on characteristics extracted from Android applications, both benign and malignant.