Wojciech Rutkowski

Semantic Compression for Specialised Information Retrieval Systems

The aim of this work is to present methods some of the ongoing research done as a part of development of Semantically Enhanced Intellectual Property Protection System - SEIPro2S. Main focus is on description of methods that allow for creation of more concise documents preserving semantically the same meaning as their originals. Thus, compacting methods are denoted as a semantic compression.

Semantically Enhanced Intellectual Property Protection System - SEIPro2S

The aim of this work is to present some of the capabilities of a Semantically Intellectual Enhanced Property Protection System. The system has reached a prototype phase where experiments are possible. It uses an extensive semantic net algorithms for Polish language that enable it to detect similarities in two compared documents on a level far beyond simple text matching. SEIPro2S benefits both from using a local document repository and from Web based resources. Main focus of this work is to give a reader overview of architecture and some actual results.

Quality of semantic compression in classification

Article presents results of implementation of semantic compression for English. An idea of semantic compression is reintroduced with examples and steps taken to perform experiment are given. A task of re-engineering available structures in order to apply them to already existing project infrastructure for experiments is described. Experiment demonstrates validity of research along with real examples of semantically compressed documents.

Towards Knowledge Acquisition with WiSENet

This article is a continuation of research work started with an idea of semantic compression. As authors proved that semantic compression is viable concept for English, they decided to focus on potential applications. An algorithm is presented that employing WiSENet allows for knowledge acquisition with flexible rules that yield high precision results. Detailed discussion is given with description of devised algorithm, usage examples and results of experiments.

Towards the Lexicon-Based Sentiment Analysis of Polish Texts: Polarity Lexicon

Due to the increasing amount of information available on the Web, sentiment analysis aiming at an automatic identification of the emotional load of texts is growing in importance. The aim of our research is to devise a reliable method for analsing sentiment in Polish texts, which requires developing adequate polarity lexical resources. In this paper, we discuss a method of building a fine-grained polarity lexicon for Polish based on custom-built review corpora. The compiled lexicon is subsequently tested in the field of sentiment analysis reaching the accuracy level of up to 79%.

Automated Acquisition of Semantic Relations for Information Retrieval Systems

Considering a continuous rise of world’s information resources, it is necessary for companies and other organizations to obtain, aggregate, process and utilize them in an appropriate manner in order to maximize the effectiveness of activities that are being conducted. Since the document libraries (or a number of sources to filter from) are becoming bigger and bigger, it is crucial to provide a trusted system which would be able to find the resources relevant to user’s needs. This is a main goal of information retrieval (IR) systems. (Daconta et al. 2003) Traditionally, the effectiveness of IR systems is measured by two basic factors: recall and precision. Both are quantified by a percentage or a value between 0 and 1. Suppose we have a set of documents. A user has specific information needs, represented by a query. Task of IR system is to provide the user with relevant documents from the set. Recall equals a relation of relevant documents returned by IR system to the number of all relevant documents, and precision is a relation of returned relevant documents to all returned documents. Were everything perfect, the recall and precision of IR system would reach 100%. This is a goal of developing and improving retrieval systems.

Polarity Lexicon for the Polish Language: Design and Extension with Random Walk Algorithm

Sentiment analysis aims at an automatic assignment to a portion of text a value expressing an emotional attitude towards its content. Out of numerous efficient methods for investigating sentiment, the authors decided to opt for the lexicon-based approach. A necessary prerequisite for adopting it was the availability of specific lexical resources for the investigated language. While there are substantial readily accessible polarity resources for English, those for Polish are meagre and, to the best of our knowledge, none of them is able to fully support sentiment analysis. Accordingly, the main objective of the presented work is to plug this gap in academic research by creating in an automated manner, a polarity lexical resource for the Polish language. In this paper, we present the motivation for the study and the key mechanisms underlying the development of the polarity lexicon, elucidate the linguistic phenomena to be reckoned with in the process, as well as discuss the random walk algorithm used to extend the obtained polarity resources. Finally, the results of the conducted experiments and the newly compiled polarity lexicon are demonstrated.

Domain based semantic compression for automatic text comprehension augmentation and recommendation

This works presents an application of semantic compression where domain frequency dictionaries are used to augment comprehension of documents. This is achieved by incorporating users feedback into proposed solution. Experiments and examples of actual output are given. Moreover, a measure that allows for evaluation of changes in a structure of available groups is defined and presented.

Linguistically Aware Semantic Network for Automated Information Tracking

The paper presents the result of an ongoing research into extending the WiSENet semantic network by adding an extra tier of data, which could be applied in a number of text processing tasks. It was envisaged that this new tier of data could provide a more detailed description of adjectives and adverbs contained in the semantic network. This detailed description was centrally concerned with the construction of sentiment vectors that convey data on whether a given term of interest is positive, negative or neutral. To this end, an experiment was carried out which made it possible to build such sentiment vectors. In the building phase a corpus of over 12000 documents containing opinions on hotels, books and movies along with corresponding satisfaction scores was used.. In order to check if sentiment vectors are a useful instrument for performing text processing tasks, a test case was proposed to help prepare the satisfaction level prognosis for the tested documents. In the course of the research it became apparent that not all sentiment vectors were eligible for use in the prognosis. In order to uncover plausible reasons for the ineligibility a linguistic analysis was carried out. Finally, an experiment designed to evaluate the usefulness of sentiment vectors and the level of accuracy of the prognosis based on them was performed. It had a success rate of 61.33%, which is highly satisfactory in comparison to other studies.

Sentiment analysis resources for Polish: a collaborative approach

With a rapid increase in the amount of information available on the web, there is a growing need for sentiment analysis. The overall objective of our ongoing research is to work out a robust method of detecting sentiment in Polish texts by means of developing relevant polarity lexical resources. The present study attempts to illuminate the process of creating a general polarity lexicon for Polish based on outputs provided by a customised Markov Random Walk algorithm, operating on an independently built semantic network for the Polish language. The customisations include a specially developed application that allows for a collaborative evaluation of the algorithms output so that it can be used as a newest of parameters in a subsequent iteration of experiments. The results are compared with both the findings of our previous studies as well as those reported by other researchers.