Large Language Models (LLMs) have demonstrated remarkable factual accuracy in producing human-like and AI-generated texts across a wide range of natural language tasks, including question answering. Despite these advances, their tendency to hallucinate and produce fabricated, false or incorrect responses is a persistent limitation. This limitation undermines their reliability and remains a critical challenge, especially in areas where high precision and trustworthiness are required. To address this challenge, we investigate whether the features derived from Knowledge Graphs (KGs) align with the accuracy of answers produced by the LLMs. In particular, we focus on entropy-based KG features, which capture diversity and uncertainty within structured knowledge. By analyzing the correlation between the entropy-based KG features and the accuracy of LLM responses, we are able to identify “blind spots” where LLMs are prone to hallucination. This provides insights not only into when an LLM is correct, but also into the conditions under which it fails. We present results across several datasets, including two developed for this study, demonstrating that entropy-based KG features can effectively align with the accuracy of LLM responses. Motivated by these findings, we propose a probing strategy for assessing LLM accuracy by focusing on areas where LLM accuracy is weak. The experimental results confirm that KG features can guide the probing effectively, highlighting the importance of using structured features from KGs in building more reliable and hallucination-free AI based systems.

Abstract Commonsense explanation generation refers to reasoning and explaining why a commonsense statement contradicts commonsense knowledge, such as why the statement “My dad grew volleyballs in his garden” is nonsensical. While such reasoning is trivial for humans, it remains a challenge for AI systems. Despite their notable performance in tasks like text generation and reasoning, large language models (LLMs) often fall short of consistently generating coherent and accurate commonsense explanations. To bridge this gap, we propose a novel Two-stage Identification and Prompting (TIP) framework for enhancing LLMs’ ability to handle the task of commonsense explanation generation. Specifically, in the first stage, TIP identifies the nonsensical concept in the given statement, pinpointing the specific element that contradicts commonsense knowledge. In the second stage, TIP generates implicit knowledge based on the identified nonsensical concept and then leverages this implicit knowledge to guide the adopted LLMs in generating explanations. In order to demonstrate the effectiveness of the proposed TIP framework for commonsense explanation generation, we conducted extensive experiments based on the ComVE dataset and a newly constructed CSE dataset, where a variety of LLMs are evaluated. The experimental results show that TIP consistently outperforms all baseline methods across multiple metrics, demonstrating its effectiveness in improving LLMs’ commonsense reasoning and explanation generation capabilities.

Stream processing enables real-time data analysis. Recent stream processing engines (SPEs) execute stream processing in a distributed manner for real-time analysis of massive amounts of data produced by IoT devices and sensors. It has been widely adopted in various applications that support critical decision making. To explain the results of stream processing, ensuring provenance is indispensable. Provenance clarifies the relationship between input data and output data in the processing. With provenance, we can understand what input data contributed to the output. Existing frameworks for providing provenance for stream processing generate provenance or additional information to construct provenance at runtime. However, these approaches impose substantial overhead in ordinary stream processing. In this paper, we propose a new framework, named LPStream, for fine-grained lazy provenance. LPStream is the first framework to support lazy provenance for stream processing. In the ordinary execution mode, LPStream executes stream processing with checkpointing but without provenance generation. If provenance is necessary for some target output tuples, it replays the processing from an appropriate checkpoint and generates the provenance for the target tuple. We explain the design and implementation of LPStream and evaluate its performance by comparing LPStream with stream processing without provenance and with eager provenance. The experimental results demonstrate the effectiveness of our proposal.

Predicting missing attribute values in data is extremely important in improving the accuracy in many applications. Existing algorithms ignore the difference between the records used for learning and predicting. The accuracy is not good enough and can be further improved. This paper proposes two solutions: (1) Maximization-based approach (MP) and (2) Distance-ratio-based approach (DP). MP and DP ensure that the incomplete records with the missed values are similar to the records used to learn the parameters as much as possible. MP and DP learn all possible parameters not only from the k nearest neighboring set (k-NN) but from the k-Sets, which are all possible combinations of k complete records. The parameters learnt from the records that are most similar to the repaired candidates of the incomplete records are chosen. Experimentally, MP and DP significantly outperform the existing approaches.

Entity Matching (EM) involves identifying different data representations referring to the same entity from multiple data sources and is typically formulated as a binary classification problem. It is a challenging problem in data integration due to the heterogeneity of data representations. State-of-the-art solutions have adopted NLP techniques based on pre-trained language models (PrLMs) via the fine-tuning paradigm, however, sequential fine-tuning of overparameterized PrLMs can lead to catastrophic forgetting, especially in low-resource scenarios. In this study, we propose a parameter-efficient paradigm for fine-tuning PrLMs based on adapters, small neural networks encapsulated between layers of a PrLM, by optimizing only the adapter and classifier weights while the PrLMs parameters are frozen. Adapter-based methods have been successfully applied to multilingual speech problems achieving promising results, however, the effectiveness of these methods when applied to EM is not yet well understood, particularly for generalized EM with heterogeneous data. Furthermore, we explore using (i) pre-trained adapters and (ii) invertible adapters to capture token-level language representations and demonstrate their benefits for transfer learning on the generalized EM benchmark. Our results show that our solution achieves comparable or superior performance to full-scale PrLM fine-tuning and prompt-tuning baselines while utilizing a significantly smaller computational footprint of the PrLM parameters.

Abstract Data lineage allows information to be traced to its origin in data analysis by showing how the results were derived. Although many methods have been proposed to identify the source data from which the analysis results are derived, analysis is becoming increasingly complex both with regard to the target (e.g., images, videos, and texts) and technology (e.g., AI and machine learning (ML)). In such complex data analysis, simply showing the source data may not ensure traceability. For example, ML analysts building image classifier models often need to know which parts of images are relevant to the output and why the classifier made a decision. Recent studies have intensively investigated interpretability and explainability in the AI/ML domain. Integrating these techniques into the lineage framework will help analysts understand more precisely how the analysis results were derived and how the results are trustful. In this paper, we propose the concept of augmented lineage for this purpose, which is an extended lineage, and an efficient method to derive the augmented lineage for complex data analysis. We express complex data analysis flows using relational operators by combining user-defined functions (UDFs). UDFs can represent invocations of AI/ML models within the data analysis. Then, we present a method taking UDFs into consideration to derive the augmented lineage for arbitrarily chosen tuples among the analysis results. We also experimentally demonstrate the efficiency of the proposed method.

Recent advances in location-aware devices have resulted in an exponential increase in the trajectory data streams. A number of applications require real-time processing and analysis of massive moving objects' trajectories. For instance, route guidance in emergency evacuation, patients tracking, etc. Existing scalable trajectory management systems lack support for real-time processing, while the real-time systems do not natively support spatial trajectory processing. This work presents TStream, a real-time and scalable trajectory stream processing and analysis framework. TStream utilizes grid index to support efficient processing of continuous range, kNN and join queries.

With the development of Light Detection and Ranging (LiDAR) technology, point cloud data is a valuable resource to build three-dimensional (3D) models of digital twins. The geospatial 3D model is the principal element to abstract a geographic feature with geometric and semantic properties. The 3D model data provides more efficiency to handle, retrieve, exchange, and visualize geographic features compared to point clouds. However, the construction of 3D models, especially indoor space where various objects exist, usually necessitates expensive time and manual labor resources to organize and extract the geometry information by authoring tools.

With the increase in the use of 3D scanner to sample the earth surface, there is a surge in the availability of 3D spatial data. 3D spatial data contains a wealth of information and can be of potential use if integrated, processed and analyzed in real-time. The 3D spatial data is generated as continuous data stream, however due to its size, velocity and inherent noise, it is processed offline. Many applications require real-time processing and analysis of spatial stream, for-instance, forest fire management, real-time road traffic analysis, disaster engulfed areas monitoring, etc., however they suffer from slow offline processing of traditional systems. This paper presents and demonstrates a robust and scalable pipeline for the real-time processing and analysis of 3D spatial streams. An experimental evaluation is also presented to prove the effectiveness of the proposed framework.