Vol. 36 · No. 1

Natural Language Generation (NLG) is defined as the systematic approach for producing human understandable natural language text based on non-textual data or from meaning representations. This is a significant area which empowers human-computer interaction. It has also given rise to a variety of theoretical as well as empirical approaches. This paper intends to provide a detailed overview and a classification of the state-of-the-art approaches in Natural Language Generation. The paper explores NLG architectures and tasks classed under document planning, micro-planning and surface realization modules. Additionally, this paper also identifies the gaps existing in the NLG research which require further work in order to make NLG a widely usable technology.

Two well-known approaches for modelling virtual vegetation are grammar-based methods (L-systems) and the Xfrog method, which is based on graph transformations expanding ``multiplier'' nodes. We show that both approaches can be unified in the framework of ``relational growth grammars'', a variant of parallel graph grammars. We demonstrate this possibility and the synergistic benefits of the combination of both methods at simple plant models which were processed using our open-source software GroIMP.

The main goal of the paper is to present the method of design pattern support based on principles of model driven development: the abstraction, semantics and model transformations. More specifically, the method is based on the principle of suggestion of design pattern instances via the semantic marking of model elements or source code fragments and on the subsequent transformations of this way marked models or source code. Thanks to the continual support of the design patterns at more levels of abstraction and thanks to the transformations between particular model levels and source code, the method is targeted to achieve the applicability in the area of the iterative, incremental and model driven development.

Human societies appear in many types of simulations. Particularly, a lot of new computer games contain a virtual world that imitates the real world. A few of the most important and the most difficult society elements to be modelled are the social context and individuals cooperation. In this paper we show how the social context and cooperation ability can be provided using agents that are equipped with internal visions of mutual social relations. Internal vision is a representation of social relations from the agent's point of view so, due to being subjective, it may be inconsistent with the reality. We introduce the agent model and the mechanism of rebuilding the agent's internal vision that is similar to that used by humans. An experimental proof of concept implementation is also presented.

Computer system security is the fastest developing segment in information technology. The conventional approach to system security is mostly aimed at protecting the system, while current trends are focusing on more aggressive forms of protection against potential attackers and intruders. One of the forms of protection is also the application of advanced technology based on the principle of baits – honeypots. Honeypots are specialized devices aimed at slowing down or diverting the attention of attackers from the critical system resources to allow future examination of the methods and tools used by the attackers. Currently, most honeypots are being configured and managed statically. This paper deals with the design of a sophisticated hybrid honeypot and its properties having in mind enhancing computer system security. The architecture of a sophisticated hybrid honeypot is represented by a single device capable of adapting to a constantly changing environment by using active and passive scanning techniques, which mitigate the disadvantages of low-interaction and high-interaction honeypots. The low-interaction honeypot serves as a proxy for multiple IP addresses and filters out traffic beyond concern, while the high-interaction honeypot provides an optimum level of interaction. The proposed architecture employing the prototype of a hybrid honeypot featuring autonomous operation should represent a security mechanism minimizing the disadvantages of intrusion detection systems and can be used as a solution to increase the security of a distributed computer system rapidly, both autonomously and in real-time.

Over decades, I/O optimizations implemented in legacy file systems have been concentrated on reducing HDD disk overhead, such as seek time. As SSD (Solid-State Device) is becoming the main storage medium in I/O storage subsystems, file systems integrated with SSD should take a different approach in designing I/O optimizations. This is because SSD deploys the peculiar device characteristics that do not take place in HDD, such as erasure overhead on flash blocks and absence of seek time to positioning data. In this paper, we present HP-hybrid (High Performance-hybrid) file system that provides a single hybrid file system space, by combining HDD and SSD partitions. HP-hybrid targets for optimizing I/O while considering the strength and weakness of two different partitions, to store large-scale amounts of data in a cost-effective way. Especially, HP-hybrid proposes spatial optimizations that are executed in a hierarchical, fine-grained I/O unit, to address the limited SSD storage resources. We conducted several performance experiments to verify the effectiveness of HP-hybrid while comparing to ext2, ext4 and xfs mounted on both SSD and HDD.

In this paper we propose a symmetric cryptographic approach named Square Key Matrix Management Scheme (SKMaS) in which a sensor node named Key Distribution Server (KDS) is responsible for the security of key management. When the system starts up, the KDS sends its individual key and two sets of keys to sensor nodes. With the IDs, any two valid sensor nodes, e.g. i and j, can individually identify the corresponding communication keys (CKs) to derive a dynamic shared key (DSK) for encrypting/decrypting messages transmitted between them. When i leaves the underlying network, the CKs and the individually keys currently utilized by i can be reused by a newly joining sensor, e.g. h. However, when h joins the network, if no such previously-used IDs are available, h will be given a new ID, CKs and the individually key by the KDS. The KDS encrypts the CKs, with which an existing node q can communicate with h, with individual key so that only q rather than h can correctly decrypt the CKs. The lemmas and security analyses provided in this paper prove that the proposed system can protect at least three common attacks.

The main goal of this work was to experimentally verify the methods for a challenging task of categorization and clustering Polish text. Supervised and unsupervised learning was employed respectively for the categorization and clustering. A profound examination of the employed methods was done for the custom-built corpus of Polish texts. The corpus was assembled by the authors from Internet resources. The corpus data was acquired from the news portal and, therefore, it was sorted by type by journalists according to their specialization. The presented algorithms employ Vector Space Model (VSM) and TF-IDF (Term Frequency-Inverse Document Frequency) weighing scheme. Series of experiments were conducted that revealed certain properties of algorithms and their accuracy. The accuracy of algorithms was elaborated regarding their ability to match human arrangement of the documents by the topic. For both the categorization and clustering, the authors used F-measure to assess the quality of allocation.

A novel semi-supervised kernel feature extraction algorithm to combine an efficient metric learning method, i.e. relevant component analysis (RCA), and kernel trick is presented for hyperspectral imagery land-cover classification. This method obtains projection of the input data by learning an optimal nonlinear transformation via a chunklet constraints-based FDA criterion, and called chunklet-based kernel relevant component analysis (CKRCA). The proposed method is appealing as it constructs the kernel very intuitively for the RCA method and does not require any labeled information. The effectiveness of the proposed CKRCA is successfully illustrated in hyperspectral remote sensing image classification. Experimental results demonstrate that the proposed method can greatly improve the classification accuracy compared with traditional linear and conventional kernel-based methods.

Collaborative tagging which is the keystone of the social practices of web 2.0 has been highly developed in the last few years. In this paper, we propose a new method to analyze user profiles according to their tagging activity in order to improve resource recommendation. We base upon association rules which is a powerful method to discover interesting relationships among large datasets on the web. Focusing on association rules we can find correlations between tags in a social network. Our aim is to recommend resources annotated with tags suggested by association rules, in order to enrich user profiles. The effectiveness of the recommendation depends on the resolution of social tagging drawbacks. In our recommender process, we demonstrate how we can reduce tag ambiguity and spelling variations problems by taking into account social similarities calculated on folksonomies, in order to personalize resource recommendation. We surmount also the lack of semantic links between tags during the recommendation process. Experiments are carried out with two different scenarios: the first one is a proof of concept over two baseline datasets and the second one is a real world application for diabetes disease.