Online firestorms are rapid collective escalations of highly negative user-generated content and may cause substantial reputational and economic damage. Existing detectors usually work with volume signals, sentiment scores, or predefined linguistic features. Such signals are useful, but they capture contextual meaning shifts in evolving discussion threads only indirectly. This paper proposes an LLM-based detection system with two operating modes. The first mode classifies complete Reddit threads retrospectively by combining local chunk-level assessments into a thread-level judgment. The second mode processes threads sequentially and issues early warnings when a sliding window exceeds calibrated thresholds. In this mode, the language model estimates three firestorm indicators: negativity share, escalation level, and contributor count. On a balanced Reddit dataset, the global mode achieves strong classification performance, while the early warning mode reaches high recall and detects escalating threads after only a small number of comments and distinct contributors. The results indicate that LLMs can be used not only for static judgment tasks, but also as repeated estimators in context-aware monitoring of social media discourse.

Automated vulnerability detection tools are widely used to identify security vulnerabilities in software dependencies. However, the evaluation of such tools remains challenging due to the heterogeneous structure of vulnerability data sources, inconsistent identifier schemes, and ambiguities in version range specifications. In this paper, we present an empirical evaluation of vulnerability detection across multiple software ecosystems using a curated ground-truth dataset derived from the Open Source Vulnerabilities (OSV) database. The dataset explicitly maps vulnerabilities to concrete package versions and enables a systematic comparison of detection results across different tools and services. Since vulnerability databases such as OSV are continuously updated, the dataset used in this study represents a snapshot of the vulnerability landscape at the time of the evaluation. To support reproducibility and future studies, we provide an open-source tool that automatically reconstructs the dataset from the current OSV database using the methodology described in this paper. Our evaluation highlights systematic differences between vulnerability detection systems and demonstrates the importance of transparent dataset construction for reproducible empirical security research.

In recent years, toxic content and hate speech have become widespread phenomena on the internet. Moderators of online newspapers and forums are now required, partly due to legal regulations, to carefully review and, if necessary, delete reader comments. This is a labor-intensive process. Some providers of large language models already offer solutions for automated hate speech detection or the identification of toxic content. These include GPT-4o from OpenAI, Jigsaw's (Google) Perspective API, and OpenAI's Moderation API. Based on the selected German test dataset HOCON34k, which was specifically created for developing tools to detect hate speech in reader comments of online newspapers, these solutions are compared with each other and against the HOCON34k baseline. The test dataset contains 1,592 annotated text samples. For GPT-4o, three different promptings are used, employing a Zero-Shot, One-Shot, and Few-Shot approach. The results of the experiments demonstrate that GPT-4o outperforms both the Perspective API and the Moderation API, and exceeds the HOCON34k baseline by approximately 5 percentage points, as measured by a combined metric of MCC and F2-score.

Cover songs are alternate versions of a song by a different artist. Long being a vital part of the music industry, cover songs significantly influence music culture and are commonly heard in public venues. The rise of online music platforms has further increased their prevalence, often as background music or video soundtracks. While current automatic identification methods serve adequately for original songs, they are less effective with cover songs, primarily because cover versions often significantly deviate from the original compositions. In this paper, we propose a novel method for cover song detection that utilizes the lyrics of a song. We introduce a new dataset for cover songs and their corresponding originals. The dataset contains 5078 cover songs and 2828 original songs. In contrast to other cover song datasets, it contains the annotated lyrics for the original song and the cover song. We evaluate our method on this dataset and compare it with multiple baseline approaches. Our results show that our method outperforms the baseline approaches.

Information about human presence in indoor spaces is crucial for building energy optimization. While there has been a considerable amount of research on using neural networks to automatically detect occupancy from CO2 sensors, their application in practice is limited due to the scarcity of labeled training data. In this paper, we propose Coddora, an off-the-shelf deep learning model pretrained on data from randomized room simulations. Coddora enables quick adaptation to real-world rooms, requiring only minimal data collection. Our contribution includes two model variants for application via fine-tuning or zero-shot classifying, as well as the synthetic dataset providing data from simulations with 100,000 room models.

We present a dataset of 34,223 comments in German, authored by users of online platforms associated with public discourse in German newspapers. Each comment was annotated for hate speech and the adequacy of contextual information by 29 volunteers using a binary annotation scheme. The inter-rater reliability for hate speech, measured by Fleiss' Kappa, is 0.4428 across all annotators, improving to 0.6078 when focusing on an optimized subset of 12 annotators. Additionally, we provide a baseline text classification using BERT, which achieved an MCC-score of up to 0.32 and an F2-score of up to 0.64 in initial experiments with this corpus. The dataset, named HOCON34k, comprising German hate speech comments from newspapers, is publicly available for research purposes.

We provide a german corpus consisting of 7,061 posts authored by users of social media platforms. A group of volunteers annotated each post according to hatespeech and misogynistic/misogynous hatespeech in a binary fashion. The interrater reliability over all annotators according to Fleiss’ Kappa is 0.6409 for hatespeech and 0.8258 for misogynistic hatespeech. Furthermore, baseline measurements with machine learning based text classification with BERT are presented. Initial experiments with the corpus achieve macro average F1-scores up to 0.79 for hatespeech and 0.75 for misogynistic hatespeech. The dataset of the corpus on German Misogynistic Hatespeech Posts (GMHP7k) is publicly available.

Nowadays, the use of third-party libraries in software is common. At the same time, the number of published libraries continues to increase. An automated classification should help to maintain an overview and identify similar software libraries. This paper investigates if new approaches can be used to classify all software libraries crawled from Apache Maven repositories into defined classes using machine learning. In addition to tags that are not always available or of poor quality, we examine one feature that is always available—the id. Consisting of group-id and artifact-id, the id of an Apache Maven software library contains valuable information that can help in classification. Through a developed preprocessing and an optimized recurrent neural network (RNN), the tokenised ids should allow a classification of most libraries. Furthermore, we present an optimized approach through a hybrid use of id tokens and tags in combination. Based on the dataset including 28,600 labeled entries, a comparison of various approaches was carried out. The RNN achieved a balanced accuracy of 71.36% by training on tokenised ids. A model trained on tags achieved a balanced accuracy of 92%. However, the new hybrid approach, which combines tags and ids, optimizes the result to 94.12%. While a classification on tags achieves a better result than the more general id-based approach, the applicability is limited to software libraries that are tagged. The hybrid approach, on the other hand, takes advantage of the classification results based on tags when these are available, but includes valuable information from the always available ids.

Research towards energy optimization in buildings heavily relies on building-related data such as measured indoor climate factors. While data collection is a labor- and cost-intensive task, simulations are a cheap alternative to generate datasets of arbitrary sizes, particularly useful for data-intensive deep learning methods. In this paper, we present the tool EnergyPlus Room Simulator, which enables the simulation of indoor climate in a specific room of a building using the simulation software EnergyPlus. It allows to alter room models and simulate various factors such as temperature, humidity, and CO2 concentration. In contrast to manually working with EnergyPlus, this tool enhances the simulation process by offering a convenient interface, including a user-friendly graphical user interface (GUI) as well as a REST API. The tool is intended to support scientific, building-related tasks such as occupancy detection on a room level by facilitating fast access to simulation data that may, for instance, be used for pre-training machine learning models.

Dieses Buch führt ausführlich in die Datenkommunikation ein und erläutert eingehend die Protokollmechanismen der Transportprotokolle TCP und UDP.

Finding the right business partner to drive innovation or acquire technology transfer is a labor and time-intensive process. To simplify this process, there is a need for improved methods of automated matchmaking that can quickly identify the best potential collaboration partners. This paper presents a novel approach for semi-automated business matchmaking between companies and research institutes, that is applied to a first case study. For this purpose, we compare two transformer-based text classification models and evaluate how dataset quality affects few-shot learning performance. Flair's TARS classifier performed very well in our use case, requiring only 40 examples per class to achieve an F1 score of about 90

Efficiently synchronizing data with external sources, such as social media feeds, while minimizing well-timed requests is a challenge in various domains. This research investigates prediction algorithms for determining appropriate update intervals for Facebook and Twitter feeds, considering metrics such as the delay (time between a post's publication and retrieval) and requests per post. Due to variations in update intervals, different algorithms yield diverse results. Selecting the most suitable algorithm for each feed is a time-consuming but crucial task for achieving optimal resource usage. We propose three strategies for algorithm selection: baseline (using a single algorithm per feed), optimum (calculating the best algorithm for each feed), and classification (identifying algorithms through classification). Real-world data from Facebook and Twitter are used to evaluate the strategies, comprehensively assessing their strengths and weaknesses. Findings demonstrate that the strategy optimum identifies the best algorithms, while the strategy classification selects fairly good algorithms at significantly reduced computational effort.

In this paper, we present AMPEL, a system that assists social media managers in creating successful posts for company pages on social media platforms. The AMPEL workflow classifies posts as either successful or unsuccessful, using Facebook as an example. The system makes a prediction of success for a new post during its creation, prior to its publication, which is a major advantage in comparison to existing systems. Posts that are classified as unsuccessful can be revised by the author until the prediction is successful. The system also evaluates previously published posts for success, allowing for comparison between the predicted success and actual success achieved. The two classification models are built using Random Forest, XGBoost, and neural network-based classification algorithms. The system is evaluated using two separate corpora of posts from different industries. We also demonstrate a prototype of AMPEL and show that it achieves good results with very reliable predictions of success. Our evaluation shows that AMPEL can replace manual review by a human social media manager in many applications.

The paper presents an agent-based modeling and simulation approach to model the forest fuel supply chain for heating purposes (i.e., heating plants). The paper focuses on the simulation of the processes of timber harvesting by forest enterprises and the competition of heating plants for the limited resource of wood chips. In particular, the work identifies different stakeholders having an adaptive behavior-with respect to the overall market conditions and timber prices. The agent-based model developed here-called SimFoMa-uses three types of agents-forest enterprises, heating plants, and traders. The agents are interacting in an environment that has rich information on the forests and road network. The SimFoMa model is applied to a test area, the province of Carinthia, Austria. We defined six different simulation scenarios that cover different market situations-from increasing timber prices, volatile market conditions, or decreasing market conditions-and evaluated the harvest patterns, transport distances and the forest itself. The paper utilizes the agent-based modeling methodology to model the agent's adaptive behavior of the forest fuel supply chain and to model the competition of heating plants for forest fuels. To evaluate this phenomena we mainly analyze transport distances of the simulation runs. For the test area of Carinthia, the experiments show that the behavior of small forest owners influences the supply of forest fuels. Timber prices not meeting the expectations of small forest owners might not motivate them to produce timber and forest fuels. On the long run the overall forest fuel supply does not meet the demand in the test area Carinthia-hence it relies on biomass imports. Furthermore, we witnessed increasing transport distances from harvest site to heating plant. Recommendations for Resource Managers The results of the spatial Agent-based simulation of the forest fuel market with agents competing for the limited resource forest biomass show that transport distances for forest fuels can vary and may increase over time. Hence, the planning of the forest fuels supply and the respective transport distances is crucial to reduce the carbon footprint of the timber for heating purposes. As small forest owners produce timber on a more irregular basis (based on the price in the market), the motivation of small forest owners is crucial for the steady supply of biomass for heating purposes-for the case of Carinthia. In the long run it is not possible to fulfill the demand of biomass for heating purposes for Carinthia, without imports of timber. Again, crucial is the motivation of small forest owners to produce timber.

PurposeThis paper aims to evaluate different approaches for the parallelization of compute-intensive tasks. The study compares a Java multi-threaded algorithm, distributed computing solutions with MapReduce (Apache Hadoop) and resilient distributed data set (RDD) (Apache Spark) paradigms and a graphics processing unit (GPU) approach with Numba for compute unified device architecture (CUDA).Design/methodology/approachThe paper uses a simple but computationally intensive puzzle as a case study for experiments. To find all solutions using brute force search, 15! permutations had to be computed and tested against the solution rules. The experimental application comprises a Java multi-threaded algorithm, distributed computing solutions with MapReduce (Apache Hadoop) and RDD (Apache Spark) paradigms and a GPU approach with Numba for CUDA. The implementations were benchmarked on Amazon-EC2 instances for performance and scalability measurements.FindingsThe comparison of the solutions with Apache Hadoop and Apache Spark under Amazon EMR showed that the processing time measured in CPU minutes with Spark was up to 30% lower, while the performance of Spark especially benefits from an increasing number of tasks. With the CUDA implementation, more than 16 times faster execution is achievable for the same price compared to the Spark solution. Apart from the multi-threaded implementation, the processing times of all solutions scale approximately linearly. Finally, several application suggestions for the different parallelization approaches are derived from the insights of this study.Originality/valueThere are numerous studies that have examined the performance of parallelization approaches. Most of these studies deal with processing large amounts of data or mathematical problems. This work, in contrast, compares these technologies on their ability to implement computationally intensive distributed algorithms.