Large Language Models (LLMs) have transformed natural language processing, but concerns have emerged about their susceptibility to ideological manipulation, particularly in politically sensitive areas. Previous research has largely focused on LLM biases through a binary Left vs. Right framework, often using explicit ideological prompts and fine-tuning with political question-answering datasets. In this work, we move beyond this binary approach to explore the extent to which LLMs can be influenced across a nuanced spectrum of political ideologies, from Progressive-Left to Conservative-Right. We introduce a novel multi-task dataset designed to reflect diverse ideological positions through tasks such as ideological question-answering, statement ranking, manifesto cloze completion, and Congress bill comprehension. By fine-tuning three LLMs—Phi-2, Mistral, and Llama-3—on this dataset, we evaluate their capacity to adopt and express these nuanced ideologies. Our findings indicate that fine-tuning significantly enhances nuanced ideological alignment, while explicit prompts provide only minor refinements. This highlights the models' susceptibility to subtle ideological manipulation, suggesting a need for more robust safeguards to mitigate these risks.

Polarization is an alarming trend in modern societies with serious implications on social cohesion and the democratic process. Typically, polarization manifests itself in the public discourse in politics, governance and ideology. In recent years, however, polarization arises increasingly in a wider range of issues, from identity and culture to healthcare and the environment. As the public and private discourse moves online, polarization feeds in and is fed by phenomena like fake news and hate speech. The identification and analysis of online polarization is challenging because of the massive scale, diversity, and unstructured nature of online content, and the rapid and unpredictable evolution of polarizing issues. Therefore, we need effective ways to identify, quantify, and represent polarization and polarizing topics algorithmically and at scale. In this work, we introduce POLAR - an unsupervised, large-scale framework for modeling and identifying polarizing topics in any domain, without prior domain-specific knowledge. POLAR comprises a processing pipeline that analyzes a corpus of an arbitrary number of news articles to construct a hierarchical knowledge graph that models polarization and identify polarizing topics discussed in the corpus. Our evaluation shows that POLAR is able to identify and rank polarizing topics accurately and efficiently.

In recent years, the increasing propagation of hate speech in online social networks and the need for effective counter-measures have drawn significant investment from social network companies and researchers. This has resulted in the development of many web platforms and mobile applications for reporting and monitoring online hate speech incidents. In this article, we present MANDOLA, a big-data processing system that monitors, detects, visualizes, and reports the spread and penetration of online hate-related speech using big-data approaches. MANDOLA consists of six individual components that intercommunicate to consume, process, store, and visualize statistical information regarding hate speech spread online. We also present a novel ensemble-based classification algorithm for hate speech detection that can significantly improve the performance of MANDOLA’s ability to detect hate speech. To present the functionality and usability of our system, we present a use case scenario of real-life event annotation and data correlation. As shown from the performance of the individual modules, as well as the usability and functionality of the whole system, MANDOLA is a powerful system for reporting and monitoring online hate speech.

Over the past few years, we have been witnessing the rise of misinformation on the Internet. People fall victims of fake news continuously, and contribute to their propagation knowingly or inadvertently. Many recent efforts seek to reduce the damage caused by fake news by identifying them automatically with artificial intelligence techniques, using signals from domain flag-lists, online social networks, etc. In this work, we present Check-It, a system that combines a variety of signals into a pipeline for fake news identification. Check-It is developed as a web browser plugin with the objective of efficient and timely fake news detection, while respecting user privacy. In this paper, we present the design, implementation and performance evaluation of Check-It. Experimental results show that it outperforms state-of-the-art methods on commonly-used datasets.

Open source software is nowadays widely used and any open source software must carry a prominent license. However, the legal, natural language text of open source licenses is not always easy to interpret and an extensive manual analysis of the text may be required, in order to fully understand its content. Existing approaches present license content based on such manual interpretation. In this paper, we propose an automated license term extraction system (FOSS-LTE) for the identification of the license terms from a specific license text and the creation of a representation of these terms divided into rights, obligations and additional conditions. We present the process employed for the creation of the license term extraction system using NLP techniques and we evaluate its accuracy on a set of sentences from available licenses collected for this purpose.

Licensing decisions for new Open Source Software are not always straightforward. However, the license that accompanies the software is important as it largely affects its subsequent distribution and reuse. License information for software products is captured - among other data - in the Software Package Data Exchange (SPDX) files. The SPDX specification is gaining popularity in the software industry and has been adopted by many organizations internally. In this demonstration paper, we present our tool for the validation of SPDX files regarding proper license use. Software packages described in SPDX format are examined in order to detect license violations that may occur when a product combines different software sources that carry different and potentially contradicting licenses. The SPDX License Validation Tool (SLVT) gives the opportunity to check the compatibility of one or more SPDX files. The evaluation performed on a number of software packages demonstrates its usefulness for drawing conclusions on license use, revealing violations in some of the test projects.