Back to search

Aristotle University Of Thessaloniki (AUTH)

Statistical methods for analysis of data from texts related to information security and technological trends

Abstract

dc:description

The main purpose of this Thesis is to offer insights and solutions on different areas of interest associated with information security and technological trends by utilizing data analysis frameworks, focusing primarily on Statistical Natural Language Processing techniques. One of the primary tasks of each research study included in the Thesis encompasses approaches for topic extraction using topic modeling algorithms as well as word and document clustering techniques. Additionally, a fundamental principle of the Thesis is to contribute to the research community by extending current knowledge baselines offered by previous research in the investigated domains. The research fields that determine the data investigated in the studies of the current Thesis can be categorized into three types: (i) security vulnerabilities and exploits, (ii) patent analysis, and (iii) online social networks analysis. While the types of data used for these categories share similar characteristics associated with textual descriptions and numerical features, their main differences concern the creators, concepts, meaning, and potential usage of their data observations. Apart from the analytical frameworks built upon these three data types, a complete application (or tool) described as Click Text Mining tool (ClickTMtool) is also part of this Thesis. This tool provides functionalities for most of the key algorithms and workflows surrounding the experiments presented in this Thesis. Regarding the first research field, a main challenge towards the prioritization of security vulnerabilities is the assessment of their severity as expressed by multiple factors. This Thesis explores a factor related to evidence or clues confirming the existence of concepts that can be followed to benefit from a security vulnerability and further exploit a targeted system. According to this factor, a binary feature characterizing each vulnerability as exploitable or not is created. This feature can be investigated to evaluate the characteristics of exploitable vulnerabilities and their key differences from the rest. The practical usage of studying security vulnerabilities and exploits is expressed via the development of predictive tools measuring/forecasting the severity of a new security vulnerability from this perspective. Also, the provision of insights into patterns of vulnerabilities and exploits that should be mitigated urgently is a major contribution towards enhancing system security as well. The Thesis explores this factor with the exclusive use of vulnerability descriptions, showing that this feature indeed includes information that can lead to predicting/assessing their exploitability. At the same time, multiple words and topics with high exploitability are addressed and presented in the process. Nevertheless, the Thesis contributes to this research area by diving into the textual descriptions, rather than exploring other factors, and by producing explainable outcomes for understanding vulnerability features frequently associated with exploits. The second research field is described as patent analysis, which is the process of analyzing patented technologies to provide knowledge on the landscapes characterizing patent datasets, usually belonging to a unique area of interest. Patent data reflects on technological interests and investments of both minor and major organizations/individuals, while collective information/insights extracted from datasets of this type can address the needs and trends of the market, existing breakthroughs, and emerging technologies. As a plethora of technological directions exist, the Thesis explores domains that were not previously investigated to the levels that the presented studies thoroughly examined. These areas are related to security vulnerabilities and home automation networks. The main tasks addressed in this Thesis are the identification of the main themes covering a patent dataset, the value or influence assessment of an individual technology as well as the evaluation of the technological position of stakeholders within an area of interest. In summary, the studies contribute to distinguishing key competitors and technologies that can be studied by researchers and organizations for developing impactful innovations as well as designing competitive strategies, with respect to the landscape of a unique technological area. The third research field concerns online social networks. In this Thesis, the investigated content and numerical features are related to posts, i.e. post title, comments, replies, and indicators, e.g. dates, upvotes. The data observations in this case are completely different from the two previous data types, meaning that they should be studied accordingly. Posts and their properties within an online social network primarily refer to user interests, queries, and interactions, while the previous types concern industrial investments and product weaknesses. Post information from online social networks can be proved very useful for multiple viewpoints, including the investigation of human behavior (user interactions) as well as the detection of temporal and essential trends (post content). Apart from these viewpoints, some other more complex tasks concerning content recommendation, question answering, and forecasting tools can be addressed by analyzing posts of this type via text analysis techniques. In this Thesis, the Reddit, Stack Overflow, and Twitter/X platforms are investigated for providing insights on information security-related discussions, queries associated with ChatGPT, and posts of security news, respectively. The common tasks in these studies are the exploration and identification of relevant themes characterizing each dataset of posts coming from these networks while also, additional metadata properties concerning user interactions are analyzed as well. The findings from the respective analysis first help to demystify the key themes reflecting on user interests and experiences. Also, the meta-analysis of user interactions leads to evaluating the overall approval, popularity, difficulty, and tension to interact associated with these key themes within an investigated domain. Like previously, the studies focus on tasks and subjects that were not previously addressed in the related literature, as the ultimate objective of the Thesis is to provide valuable information that can contribute to the research community and platform users. It should be mentioned that the analysis relevant to ChatGPT aims at providing a flexible topic-based trend analysis framework rather than analyzing questions of this kind exclusively. Thus, the main scope of this study is to offer mechanisms for mining questions from Stack Overflow or Stack Exchange communities. In total, apart from the discussed findings, this Thesis implements approaches to overcome some challenges concerning relevant tasks that are not quite common in the related literature. The developed frameworks are built upon algorithms aiming to facilitate advancements for an individual task compared to the existing baselines. As a result, these frameworks offer interesting directions and options as well as ideas for future research that can be adapted to different data types and representations as well as research goals.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Charmanas, Konstantinos
  • Χαρμάνας, Κωνσταντίνος

Subjects

dc:subject × 16

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/58285

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Charmanas, Konstantinos; Χαρμάνας, Κωνσταντίνος. Statistical methods for analysis of data from texts related to information security and technological trends. Aristotle University Of Thessaloniki (AUTH), 2024. http://hdl.handle.net/10442/hedi/58285