Wyniki wyszukiwania - BazTech

1

Threats and crisis events detection using machine learning techniques with social media data

Organiściak Patryk, Kuraś Paweł, Kowal Bartosz

Zeszyty Naukowe Wyższej Szkoły Technicznej w Katowicach

|

2023

|

T. 17

123--141

EN

In the paper, the authors present the outcome of web scraping software allowing for the automated classification of threats and crisis events detection. In order to improve the safety and comfort of human life, an analysis was made to quickly detect threats using a modern information channel such as social media. For this purpose, social media services that are popular in the examined region were reviewed and the appropriate ones were selected using the criteria of accessibility and popularity. Approximately 300 unique posts from local groups of cities and other administrative centers were collected and analyzed. The decision of which entry was classified as a threat was defined using the ChatGPT tool and the human expert. Both variants were tested using machine learning (ML) methods. The paper tested whether the ChatGPT tool would be effective at detecting presumed events and compared this approach to the classic ML approach.

PL

W artykule autorzy przedstawiają wyniki prac nad oprogramowaniem web scrapingowym pozwalającym na zautomatyzowaną klasyfikację zagrożeń i wykrywanie zdarzeń kryzysowych. W celu poprawy bezpieczeństwa i komfortu życia ludzi przeprowadzono analizę szybkiego wykrywania zagrożeń z wykorzystaniem nowoczesnego kanału informacyjnego jakim są media społecznościowe. W tym celu dokonano przeglądu popularnych w badanym regionie serwisów społecznościowych i wybrano odpowiednie, kierując się kryteriami dostępności i popularności. Zebrano i przeanalizowano około 300 unikalnych postów z lokalnych grup miast i innych ośrodków administracyjnych. Decyzja o tym, który wpis został sklasyfikowany jako zagrożenie, została określona przy użyciu narzędzia ChatGpt oraz przy udziale osoby (eksperta). Oba warianty zostały przetestowane przy użyciu metod uczenia maszynowego (ML). Dodatkowo, w artykule sprawdzono, czy narzędzie ChatGpt będzie skuteczne w wykrywaniu domniemanych zdarzeń i porównano to rozwiązanie z klasycznym podejściem ML, gdzie dane uczące etykietowano przy udziale ekspretra.

2

Using Inconsistency Reduction Algorithms in Comparison Matrices to Improve the Performance of Generating Random Comparison Matrices with a Given Inconsistency Coefficient Range

Kuraś Paweł, Gerka Alicja

Advances in Science and Technology. Research Journal

|

2023

|

Vol. 17, no 1

222-229

EN

The aim of this paper is to present a new method for generating random pairwise comparison matrices with a given inconsistency ratio (CR) interval using inconsistency reduction algorithms. Pairwise comparison (PC) is a popular technique for multi-criteria decision-making, its purpose is to assign weights to the compared entities, thus ranking them from best to worst. The presented method combines the traditional random generation of comparison matrices supported by inconsistency reduction algorithms: the “Xu and Wei” algorithm and the “Szybowski” algorithm. This paper presents research that shows an increase in performance when generating such matrices relative to the standard random comparison matrix generation procedure using the “Szybowski” algorithm. The other algorithms also improve the process, but to a lesser extent, making the “Szybowski” supporting algorithm the preferred solution for the new process. As a result of the research, a free online tool “PC MATRICES GENERATOR” has also been made available to efficiently generate a large number of comparison matrices with a given CR factor range, any matrix size, and any number of matrices, enabling much more efficient and less time-consuming research in many fields that use comparison matrices, as the analytic hierarchy/network process (AHP/ANP), ELECTREE, PAPRIKA, PROMETHE, VIKOR or the Best-Worst method (BWM).

3

REDUCE – A Python Module for Reducing Inconsistency in Pairwise Comparison Matrices

Kuraś Paweł, Strzałka Dominik, Kowal Bartosz, Mazurek Jiří

Advances in Science and Technology. Research Journal

|

2023

|

Vol. 17, no 4

227--234

EN

This paper introduces REDUCE, a Python module designed to minimize inconsistency in multiplicative pairwise comparisons (PC), a fundamental technique in Multi-Criteria Decision Making (MCDM). Pairwise comparisons are extensively used in various fields, including engineering science and numerical simulation methods, to compare different options based on a set of criteria. However, human errors in perception and judgment often lead to inconsistencies in pairwise comparison matrices (PCM). REDUCE addresses this issue by implementing several algorithms that identify and correct inaccurate data in PCMs, thereby reducing the inconsistency ratio. These algorithms do not require expert intervention, making REDUCE a valuable tool for both scientific research and small to medium enterprises that may not have access to costly commercial software or dedicated decision-making experts. The main functionality of the module is incorporating iterative algorithms for inconsistency reduction. The REDUCE library, written in Python and utilizing auxiliary libraries such as NumPy, SciPy, and SymPy, offers 21 functions categorized into data input helpers, consistency ratio (CR) reduction algorithms, PCM indexes, and support functions. Performance testing indicates that the library can efficiently handle matrices of varying sizes, particularly those ranging from 3x3 to 10x10, and its use significantly accelerates the process compared to spreadsheets, especially when dealing with large quantities of matrices. The library has already been used in several research papers and application tools, and its availability as a free resource opens up opportunities for small and medium-sized enterprises to leverage multi-criteria decision-making methods. Currently, there are no publicly available libraries for this solution. The authors believe that the proposed module may contribute to do better decision-making process in pairwise comparisons, not only for the circle of scientists but also for small and medium enterprises that usually cannot afford expensive commercial software and do not employ full-time experts in decision-making as they rely on the experience of their employees and free online resources. It should also contributes to the transition to Industry 4.0 and advances research in fields such as fuzzy logic, preference programming, and constructive consistent approximations.