ChatGPT is a powerful large language model (LLM) that covers knowledge resources such as Wikipedia and supports natural question answering using its own knowledge. Therefore, there growing interest in exploring whether can replace traditional knowledge-based (KBQA) models. Although have been some works analyzing the performance of ChatGPT, still lack large-scale, comprehensive testing various types complex questions to analyze limitations model. In this paper, we present framework follows...

Knowledge graphs are widely used in industrial applications, making error detection crucial for ensuring the reliability of downstream applications. Existing methods often fail to effectively leverage fine-grained subgraph information and rely solely on fixed graph structures, while also lacking transparency their decision-making processes, which results suboptimal performance. In this paper, we propose a novel Multi-Agent framework Graph Error Detection (MAKGED) that utilizes multiple large...

The structural deformation of foreland thrust zones is notably complex and remains a central focus in geology. This study investigates the characteristics formation mechanisms Hutubi anticline, located southern margin Junggar Basin, through numerical simulations using Underworld software. By designing three experimental setups, we analyzed key controlling factors anticline's development.The primary findings are as follows: (1) simulation results Experiment 1 exhibit high degree similarity to...

In the fields of geological research and engineering applications, fault identification is great significance for understanding structure evolution, predicting disasters, guiding resource exploration development. Traditional methods based on manual interpretation seismic attributes struggle to meet requirements in terms efficiency accuracy when faced with complex conditions massive amounts data. With development deep learning technology, convolutional neural networks have demonstrated...

Formal query building is an important part of complex question answering over knowledge bases. It aims to build correct executable queries for questions. Recent methods try rank candidate generated by a state-transition strategy. However, this generation strategy ignores the structure queries, resulting in considerable number noisy queries. In paper, we propose new formal approach that consists two stages. first stage, predict and leverage constrain We novel graph framework handle prediction...

Query graph construction aims to construct the correct executable SPARQL on KG answer natural language questions. Although recent methods have achieved good results using neural network-based query ranking, they suffer from three new challenges when handling more complex questions: 1) complicated syntax, 2) huge search space, and 3) locally ambiguous graphs. In this paper, we provide a solution. As preparation, extend by treating each clause as subgraph consisting of vertices edges define...

Continual table semantic parsing aims to train a parser on sequence of tasks, where each task requires the translate natural language into SQL based task-specific tables but only offers limited training examples. Conventional methods tend suffer from overfitting with supervision, as well catastrophic forgetting due parameter updates. Despite recent advancements that partially alleviate these issues through semi-supervised data augmentation and retention few past examples, performance is...

Conventional text-to-SQL studies are limited to a single task with fixed-size training and test set. When confronted stream of tasks common in real-world applications, existing methods struggle the problems insufficient supervised data high retraining costs. The former tends cause overfitting on unseen databases for new task, while latter makes full review instances from past impractical model, resulting forgetting learned SQL structures database schemas. To address problems, this paper...

The few-shot problem is an urgent challenge for single-table text-to-SQL. Existing methods ignore the potential value of unlabeled data, and merely rely on a coarse-grained Meta-Learning (ML) algorithm that neglects differences column contributions to optimization object. This paper proposes Meta Self-Training text-to-SQL (MST-SQL) method solve problem. Specifically, MST-SQL based column-wise HydraNet adopts self-training as effective mechanism learn from readily available samples. During...

Augmenting Large Language Models (LLMs) for Question Answering (QA) with domain specific data has attracted wide attention. However, often exists in a hybrid format, including text and semi-structured tables, posing challenges the seamless integration of information. Table-to-Text Generation is promising solution by facilitating transformation into uniformly text-formatted corpus. Although this technique been widely studied NLP community, there currently no comparative analysis on how...

Automatic methods for evaluating machine-generated texts hold significant importance due to the expanding applications of generative systems. Conventional tend grapple with a lack explainability, issuing solitary numerical score signify assessment outcome. Recent advancements have sought mitigate this limitation by incorporating large language models (LLMs) offer more detailed error analyses, yet their applicability remains constrained, particularly in industrial contexts where comprehensive...

Table understanding (TU) has achieved promising advancements, but it faces the challenges of scarcity manually labeled tables and presence complex table structures.To address these challenges, we propose HGT, a framework with heterogeneous graph (HG)-enhanced large language model (LLM) to tackle few-shot TU tasks.It leverages LLM by aligning semantics LLM's parametric knowledge through soft prompts instruction turning deals multi-task pre-training scheme involving three novel...

Multimodal knowledge editing represents a critical advancement in enhancing the capabilities of Large Language Models (MLLMs). Despite its potential, current benchmarks predominantly focus on coarse-grained knowledge, leaving intricacies fine-grained (FG) multimodal entity largely unexplored. This gap presents notable challenge, as FG recognition is pivotal for practical deployment and effectiveness MLLMs diverse real-world scenarios. To bridge this gap, we introduce MIKE, comprehensive...

Recent advancements in generative Large Language Models(LLMs) have been remarkable, however, the quality of text generated by these models often reveals persistent issues. Evaluating models, especially open-ended text, has consistently presented a significant challenge. Addressing this, recent work explored possibility using LLMs as evaluators. While single LLM an evaluation agent shows potential, it is filled with uncertainty and instability. To address issues, we propose MATEval: A...