Previous work on answering complex questions from knowledge bases usually separately addresses two types of complexity: with constraints and multiple hops relations. In this paper, we handle both complexity at the same time. Motivated by observation that early incorporation into query graphs can more effectively prune search space, propose a modified staged graph generation method flexible ways to generate graphs. Our experiments clearly show our achieves state art three benchmark KBQA datasets.

Multi-hop Knowledge Base Question Answering (KBQA) aims to find the answer entities that are multiple hops away in Knowl- edge (KB) from question. A major challenge is lack of supervision signals at intermediate steps. Therefore, multi-hop KBQA algorithms can only receive feedback final answer, which makes learning unstable or ineffective. To address this challenge, we propose a novel teacher-student approach for task. In our approach, stu- dent network correct query, while teacher tries...

Knowledge base question answering (KBQA) aims to answer a over knowledge (KB). Recently, large number of studies focus on semantically or syntactically complicated questions. In this paper, we elaborately summarize the typical challenges and solutions for complex KBQA. We begin with introducing background about KBQA task. Next, present two mainstream categories methods KBQA, namely semantic parsing-based (SP-based) information retrieval-based (IR-based) methods. then review advanced...

Lei Wang, Wanyu Xu, Yihuai Lan, Zhiqiang Hu, Yunshi Roy Ka-Wei Lee, Ee-Peng Lim. Proceedings of the 61st Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Knowledge base question answering (KBQA) aims to answer a over knowledge (KB). Early studies mainly focused on simple questions KBs and achieved great success. However, their performances complex are still far from satisfactory. Therefore, in recent years, researchers propose large number of novel methods, which looked into the challenges questions. In this survey, we review advances KBQA with focus solving questions, usually contain multiple subjects, express compound relations, or involve...

Math word problem (MWP) solving faces a dilemma in number representation learning. In order to avoid the issue and reduce search space of feasible solutions, existing works striving for MWP usually replace real numbers with symbolic placeholders focus on logic reasoning. However, different from common reasoning tasks like program synthesis knowledge graph reasoning, has extra requirements numerical other words, instead value itself, it is reusable property that matters more Therefore, we...

Making use of knowledge bases to answer questions (KBQA) is a key direction in question answering systems. Researchers have developed diverse range methods address this problem, but there are still some limitations with the existing methods. Specifically, neural network-based for KBQA not taken advantage recent "matching-aggregation" framework sequence matching, and when representing candidate entity, they may choose most useful context matching. In paper, we explore match answers questions....

Knowledge base question answering (KBQA) is an important task in natural language processing. Existing methods for KBQA usually start with entity linking, which considers mostly named entities found a as the starting points KB to search answers question. However, relying only on linking look answer candidates may not be sufficient. In this paper, we propose perform topic unit where units cover wider range of KB. We use generation-and-scoring approach gradually refine set units. Furthermore,...

While Math Word Problem (MWP) solving has emerged as a popular field of study and made great progress in recent years, most existing methods are benchmarked solely on one or two datasets implemented with different configurations. In this paper, we introduce the first open-source library for MWPs called MWPToolkit, which provides unified, comprehensive, extensible framework research purpose. Specifically, deploy 17 deep learning-based MWP solvers 6 our toolkit. These advanced models solving,...

Federated Learning (FL) has emerged as a de facto machine learning area and received rapid increasing research interests from the community. However, catastrophic forgetting caused by data heterogeneity partial participation poses distinctive challenges for FL, which are detrimental to performance. To tackle problems, we propose new FL approach (namely GradMA), takes inspiration continual simultaneously correct server-side worker-side update directions well take full advantage of server's...

Knowledge Base Question Answering (KBQA) has attracted much attention and recently there been more interest in multi-hop KBQA. In this paper, we propose a novel iterative sequence matching model to address several limitations of previous methods for Our method iteratively grows the candidate relation paths that may lead answer entities. The prunes away less relevant branches incrementally assigns scores paths. Empirical results demonstrate our can significantly outperform existing on three...

Zero-shot Visual Question Answering (VQA) is a prominent vision-language task that examines both the visual and textual understanding capability of systems in absence training data. Recently, by converting images into captions, information across multi-modalities bridged Large Language Models (LLMs) can apply their strong zero-shot generalization to unseen questions. To design ideal prompts for solving VQA via LLMs, several studies have explored different strategies select or generate...

This paper proposes a new depression detection system based on LLMs that is both interpretable and interactive. It not only provides diagnosis, but also diagnostic evidence personalized recommendations natural language dialogue with the user. We address challenges such as processing of large amounts text integrate professional criteria. Our outperforms traditional methods across various settings demonstrated through case studies.

The task of Question Generation over Knowledge Bases (KBQG) aims to convert a logical form into natural language question. For the sake expensive cost large-scale question annotation, methods KBQG under low-resource scenarios urgently need be developed. However, current heavily rely on annotated data for fine-tuning, which is not well-suited few-shot generation. emergence Large Language Models (LLMs) has shown their impressive generalization ability in tasks. Inspired by Chain-of-Thought...

Recently, numerous new benchmarks have been established to evaluate the performance of large language models (LLMs) via either computing a holistic score or employing another LLM as judge. However, these approaches suffer from data leakage due open access benchmark and inflexible evaluation process. To address this issue, we introduce TreeEval, benchmark-free method for LLMs that let high-performance host an irreproducible session essentially avoids leakage. Moreover, performs examiner raise...

Large language models (LLMs) have recently been shown to deliver impressive performance in various NLP tasks. To tackle multi-step reasoning tasks, few-shot chain-of-thought (CoT) prompting includes a few manually crafted step-by-step demonstrations which enable LLMs explicitly generate steps and improve their task accuracy. eliminate the manual effort, Zero-shot-CoT concatenates target problem statement with "Let's think step by step" as an input prompt LLMs. Despite success of...

Attracted by the impressive power of Multimodal Large Language Models (MLLMs), public is increasingly utilizing them to improve efficiency daily work. Nonetheless, vulnerabilities MLLMs unsafe instructions bring huge safety risks when these models are deployed in real-world scenarios. In this paper, we systematically survey current efforts on evaluation, attack, and defense MLLMs' images text. We begin with introducing overview text understanding safety, which helps researchers know detailed...

Developing automatic Math Word Problem (MWP) solvers has been an interest of NLP researchers since the 1960s. Over last few years, there are a growing number datasets and deep learning-based methods proposed for effectively solving MWPs. However, most existing benchmarked soly on one or two datasets, varying in different configurations, which leads to lack unified, standardized, fair, comprehensive comparison between methods. This paper presents MWPToolkit, first open-source framework In we...

Yunshi Lan, Jing Jiang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Conditional question answering on long documents aims to find probable answers and identify conditions that need be satisfied make the correct over documents. Existing approaches solve this task by segmenting into multiple sections, attending information at global local tokens predict corresponding conditions. However, natural structure of document discourse relations between sentences in each section are ignored, which crucial for condition retrieving across as well logical interaction To...

The security concerns surrounding Large Language Models (LLMs) have been extensively explored, yet the safety of Multimodal (MLLMs) remains understudied. In this paper, we observe that can be easily compromised by query-relevant images, as if text query itself were malicious. To address this, introduce MM-SafetyBench, a comprehensive framework designed for conducting safety-critical evaluations MLLMs against such image-based manipulations. We compiled dataset comprising 13 scenarios,...

Natural Language Processing (NLP) aims to analyze text or speech via techniques in the computer science field. It serves applications domains of healthcare, commerce, education and so on. Particularly, NLP has been widely applied domain its have enormous potential help teaching learning. In this survey, we review recent advances with focus on solving problems relevant domain. detail, begin introducing related background real-world scenarios where could contribute. Then, present a taxonomy...

Unsupervised Text Style Transfer (UTST) has emerged as a critical task within the domain of Natural Language Processing (NLP), aiming to transfer one stylistic aspect sentence into another style without changing its semantics, syntax, or other attributes. This is especially challenging given intrinsic lack parallel text pairings. Among existing methods for UTST tasks, attention masking approach and Large Models (LLMs) are deemed two pioneering methods. However, they have shortcomings in...