Question Answering Task Research Articles

Reverse Multi-Choice Dialogue Commonsense Inference with Graph-of-Thought

With the proliferation of dialogic data across the Internet, the Dialogue Commonsense Multi-choice Question Answering (DC-MCQ) task has emerged as a response to the challenge of comprehending user queries and intentions. Although prevailing methodologies exhibit effectiveness in addressing single-choice questions, they encounter difficulties in handling multi-choice queries due to the heightened intricacy and informational density. In this paper, inspired by the human cognitive process of progressively excluding options, we propose a three-step Reverse Exclusion Graph-of-Thought (ReX-GoT) framework, including Option Exclusion, Error Analysis, and Combine Information. Specifically, our ReX-GoT mimics human reasoning by gradually excluding irrelevant options and learning the reasons for option errors to choose the optimal path of the GoT and ultimately infer the correct answer. By progressively integrating intricate clues, our method effectively reduces the difficulty of multi-choice reasoning and provides a novel solution for DC-MCQ. Extensive experiments on the CICERO and CICERO_v2 datasets validate the significant improvement of our approach on DC-MCQ task. On zero-shot setting, our model outperform the best baseline by 17.67% in terms of F1 score for the multi-choice task. Most strikingly, our GPT3.5-based ReX-GoT framework achieves a remarkable 39.44% increase in F1 score.

Improving Automatic VQA Evaluation Using Large Language Models

8 years after the visual question answering (VQA) task was proposed, accuracy remains the primary metric for automatic evaluation. VQA Accuracy has been effective so far in the IID evaluation setting. However, our community is undergoing a shift towards open-ended generative models and OOD evaluation. In this new paradigm, the existing VQA Accuracy metric is overly stringent and underestimates the performance of VQA systems. Thus, there is a need to develop more robust automatic VQA metrics that serve as a proxy for human judgment. In this work, we propose to leverage the in-context learning capabilities of instruction-tuned large language models (LLMs) to build a better VQA metric. We formulate VQA evaluation as an answer-rating task where the LLM is instructed to score the accuracy of a candidate answer given a set of reference answers. We demonstrate the proposed metric better correlates with human judgment compared to existing metrics across several VQA models and benchmarks. We hope wide adoption of our metric will contribute to better estimating the research progress on the VQA task. We plan to release the evaluation code and collected human judgments.

Object-Aware Adaptive-Positivity Learning for Audio-Visual Question Answering

This paper focuses on the Audio-Visual Question Answering (AVQA) task that aims to answer questions derived from untrimmed audible videos. To generate accurate answers, an AVQA model is expected to find the most informative audio-visual clues relevant to the given questions. In this paper, we propose to explicitly consider fine-grained visual objects in video frames (object-level clues) and explore the multi-modal relations (\textit{i.e.}, the object, audio, and question) in terms of feature interaction and model optimization. For the former, we present an end-to-end object-oriented network that adopts a question-conditioned clue discovery module to concentrate audio/visual modalities on respective keywords of the question and designs a modality-conditioned clue collection module to highlight closely associated audio segments or visual objects. For model optimization, we propose an object-aware adaptive-positivity learning strategy that selects the highly semantic-matched multi-modal pair as \textit{positivity}. Specifically, we design two object-aware contrastive loss functions to identify the highly relevant question-object pairs and audio-object pairs, respectively. These selected pairs are constrained to have larger similarity values than the mismatched pairs. The positivity-selecting process is adaptive as the positivity pairs selected in each video frame may be different. These two object-aware objectives help the model understand \textit{which objects are exactly relevant to the question} and \textit{which are making sounds}. Extensive experiments on the MUSIC-AVQA dataset demonstrate the proposed method is effective in finding favorable audio-visual clues and also achieves new state-of-the-art question-answering performance. The code is available at https://github.com/zhangbin-ai/APL.

BLIVA: A Simple Multimodal LLM for Better Handling of Text-Rich Visual Questions

Vision Language Models (VLMs), which extend Large Language Models (LLM) by incorporating visual understanding capability, have demonstrated significant advancements in addressing open-ended visual question-answering (VQA) tasks. However, these models cannot accurately interpret images infused with text, a common occurrence in real-world scenarios. Standard procedures for extracting information from images often involve learning a fixed set of query embeddings. These embeddings are designed to encapsulate image contexts and are later used as soft prompt inputs in LLMs. Yet, this process is limited to the token count, potentially curtailing the recognition of scenes with text-rich context. To improve upon them, the present study introduces BLIVA: an augmented version of InstructBLIP with Visual Assistant. BLIVA incorporates the query embeddings from InstructBLIP and also directly projects encoded patch embeddings into the LLM, a technique inspired by LLaVA. This approach assists the model to capture intricate details potentially missed during the query decoding process. Empirical evidence demonstrates that our model, BLIVA, significantly enhances performance in processing text-rich VQA benchmarks (up to 17.76% in OCR-VQA benchmark) and in undertaking general (not particularly text-rich) VQA benchmarks (up to 7.9% in Visual Spatial Reasoning benchmark), and achieved 17.72% overall improvement in a comprehensive multimodal LLM benchmark (MME), comparing to our baseline InstructBLIP. BLIVA demonstrates significant capability in decoding real-world images, irrespective of text presence. To demonstrate the broad industry applications enabled by BLIVA, we evaluate the model using a new dataset comprising YouTube thumbnails paired with question-answer sets across 11 diverse categories. For researchers interested in further exploration, our code and models are freely accessible at https://github.com/mlpc-ucsd/BLIVA.

CyberQ: Generating Questions and Answers for Cybersecurity Education Using Knowledge Graph-Augmented LLMs

Building a skilled cybersecurity workforce is paramount to building a safer digital world. However, the diverse skill set, constantly emerging vulnerabilities, and deployment of new cyber threats make learning cybersecurity challenging. Traditional education methods struggle to cope with cybersecurity's rapidly evolving landscape and keep students engaged and motivated. Different studies on students' behaviors show that an interactive mode of education by engaging through a question-answering system or dialoguing is one of the most effective learning methodologies. There is a strong need to create advanced AI-enabled education tools to promote interactive learning in cybersecurity. Unfortunately, there are no publicly available standard question-answer datasets to build such systems for students and novice learners to learn cybersecurity concepts, tools, and techniques. The education course material and online question banks are unstructured and need to be validated and updated by domain experts, which is tedious when done manually. In this paper, we propose CyberGen, a novel unification of large language models (LLMs) and knowledge graphs (KG) to generate the questions and answers for cybersecurity automatically. Augmenting the structured knowledge from knowledge graphs in prompts improves factual reasoning and reduces hallucinations in LLMs. We used the knowledge triples from cybersecurity knowledge graphs (AISecKG) to design prompts for ChatGPT and generate questions and answers using different prompting techniques. Our question-answer dataset, CyberQ, contains around 4k pairs of questions and answers. The domain expert manually evaluated the random samples for consistency and correctness. We train the generative model using the CyberQ dataset for question answering task.

Meta-Interpretive LEarning with Reuse

Inductive Logic Programming (ILP) is a research field at the intersection between machine learning and logic programming, focusing on developing a formal framework for inductively learning relational descriptions in the form of logic programs from examples and background knowledge. As an emerging method of ILP, Meta-Interpretive Learning (MIL) leverages the specialization of a set of higher-order metarules to learn logic programs. In MIL, the input includes a set of examples, background knowledge, and a set of metarules, while the output is a logic program. MIL executes a depth-first traversal search, where its program search space expands polynomially with the number of predicates in the provided background knowledge and exponentially with the number of clauses in the program, sometimes even leading to search collapse. To address this challenge, this study introduces a strategy that employs the concept of reuse, specifically through the integration of auxiliary predicates, to reduce the number of clauses in programs and improve the learning efficiency. This approach focuses on the proactive identification and reuse of common program patterns. To operationalize this strategy, we introduce MILER, a novel method integrating a predicate generator, program learner, and program evaluator. MILER leverages frequent subgraph mining techniques to detect common patterns from a limited dataset of training samples, subsequently embedding these patterns as auxiliary predicates into the background knowledge. In our experiments involving two Visual Question Answering (VQA) tasks and one program synthesis task, we assessed MILER’s approach to utilizing reusable program patterns as auxiliary predicates. The results indicate that, by incorporating these patterns, MILER identifies reusable program patterns, reduces program clauses, and directly decreases the likelihood of timeouts compared to traditional MIL. This leads to improved learning success rates by optimizing computational efforts.

INTEGRATING IMAGE FEATURES WITH CONVOLUTIONAL SEQUENCE-TO-SEQUENCE NETWORK FOR MULTILINGUAL VISUAL QUESTION ANSWERING

Visual question answering is a task that requires computers to give correct answers for the input questions based on the images. This task can be solved by humans with ease, but it is a challenge for computers. The VLSP2022-EVJVQA shared task carries the Visual question answering task in the multilingual domain on a newly released dataset UIT-EVJVQA, in which the questions and answers are written in three different languages: English, Vietnamese, and Japanese. We approached the challenge as a sequence-to-sequence learning task, in which we integrated hints from pre-trained state-of-the-art VQA models and image features with a convolutional sequence-to-sequence network to generate the desired answers. Our results obtained up to 0.3442 by F1 score on the public test set and 0.4210 on the private test set.

Transformer-based vision-language alignment for robot navigation and question answering

The task of robot navigation and question answering, which is also known as Embodied Question Answering (EQA), places its emphasis on empowering agents to actively explore their environments and deliver answers to user inquiries. Considering the extensive range of potential applications, particularly in the realms of home robots and personal assistants, the Embodied Question Answering task has attracted growing attention from researchers. Owing to the difficulties in bridging the semantic divide between inputs from different modalities and capturing extended connections between widely separated pixels, most existing methods face challenges in attaining adequate performance concerning accuracy in navigation and responses. To address these challenges, we present a transformer-based framework that aligns vision and language information for the task of robot navigation and question answering. Firstly, an information fusion model is designed to utilize object tags as reference points for aligning the vision and linguistic modalities into a coherent semantic space. Secondly, a dedicated transformer block is employed to capture extensive dependencies within visual scenes, enabling the generation of more contextually appropriate actions. Lastly, the two transformer-based components are seamlessly integrated into a cohesive framework, effectively handling the complete Embodied Question Answering task. The results of our experiments clearly indicate that our approach substantially boosts the performance of every module in the system, resulting in a notable 4.1% enhancement in overall accuracy.

Benchmarks for Pirá 2.0, a Reading Comprehension Dataset about the Ocean, the Brazilian Coast, and Climate Change

ABSTRACT Pirá is a reading comprehension dataset focused on the ocean, the Brazilian coast, and climate change, built from a collection of scientific abstracts and reports on these topics. This dataset represents a versatile language resource, particularly useful for testing the ability of current machine learning models to acquire expert scientific knowledge. Despite its potential, a detailed set of baselines has not yet been developed for Pirá. By creating these baselines, researchers can more easily utilize Pirá as a resource for testing machine learning models across a wide range of question answering tasks. In this paper, we define six benchmarks over the Pirá dataset, covering closed generative question answering, machine reading comprehension, information retrieval, open question answering, answer triggering, and multiple choice question answering. As part of this effort, we have also produced a curated version of the original dataset, where we fixed a number of grammar issues, repetitions, and other shortcomings. Furthermore, the dataset has been extended in several new directions, so as to face the aforementioned benchmarks: translation of supporting texts from English into Portuguese, classification labels for answerability, automatic paraphrases of questions and answers, and multiple choice candidates. The results described in this paper provide several points of reference for researchers interested in exploring the challenges provided by the Pirá dataset.

3D Question Answering.

Visual question answering (VQA) has experienced tremendous progress in recent years. However, most efforts have only focused on 2D image question-answering tasks. In this article, we extend VQA to its 3D counterpart, 3D question answering (3DQA), which can facilitate a machine's perception of 3D real-world scenarios. Unlike 2D image VQA, 3DQA takes the color point cloud as input and requires both appearance and 3D geometrical comprehension to answer the 3D-related questions. To this end, we propose a novel transformer-based 3DQA framework "3DQA-TR", which consists of two encoders to exploit the appearance and geometry information, respectively. Finally, the multi-modal information about the appearance, geometry, and linguistic question can attend to each other via a 3D-linguistic Bert to predict the target answers. To verify the effectiveness of our proposed 3DQA framework, we further develop the first 3DQA dataset "ScanQA", which builds on the ScanNet dataset and contains over 10 K question-answer pairs for 806 scenes. To the best of our knowledge, ScanQA is the first large-scale dataset with natural-language questions and free-form answers in 3D environments that is fully human-annotated. We also use several visualizations and experiments to investigate the astonishing diversity of the collected questions and the significant differences between this task from 2D VQA and 3D captioning. Extensive experiments on this dataset demonstrate the obvious superiority of our proposed 3DQA framework over state-of-the-art VQA frameworks and the effectiveness of our major designs. Our code and dataset will be made publicly available to facilitate research in this direction. The code and data are available at http://shuquanye.com/3DQA_website/.

Taiyi: a bilingual fine-tuned large language model for diverse biomedical tasks.

Most existing fine-tuned biomedical large language models (LLMs) focus on enhancing performance in monolingual biomedical question answering and conversation tasks. To investigate the effectiveness of the fine-tuned LLMs on diverse biomedical natural language processing (NLP) tasks in different languages, we present Taiyi, a bilingual fine-tuned LLM for diverse biomedical NLP tasks. We first curated a comprehensive collection of 140 existing biomedical text mining datasets (102 English and 38 Chinese datasets) across over 10 task types. Subsequently, these corpora were converted to the instruction data used to fine-tune the general LLM. During the supervised fine-tuning phase, a 2-stage strategy is proposed to optimize the model performance across various tasks. Experimental results on 13 test sets, which include named entity recognition, relation extraction, text classification, and question answering tasks, demonstrate that Taiyi achieves superior performance compared to general LLMs. The case study involving additional biomedical NLP tasks further shows Taiyi's considerable potential for bilingual biomedical multitasking. Leveraging rich high-quality biomedical corpora and developing effective fine-tuning strategies can significantly improve the performance of LLMs within the biomedical domain. Taiyi shows the bilingual multitasking capability through supervised fine-tuning. However, those tasks such as information extraction that are not generation tasks in nature remain challenging for LLM-based generative approaches, and they still underperform the conventional discriminative approaches using smaller language models.

Coordinating explicit and implicit knowledge for knowledge-based VQA

Pre-trained models often generate plausible looking statements that are factually incorrect because of the inaccurate implicit knowledge contained in the model’s parameters. Related methods retrieve explicit knowledge from the external knowledge source to help improve the prediction performance and reliability. However, these methods often use weak training signals for the retriever, and require the model to make each prediction based on the retrieved knowledge, even when the retrieved knowledge is not reliable or the model can produce better prediction only using its implicit knowledge. Therefore, it is necessary to enable the pre-trained model to actively select more beneficial knowledge for producing better prediction. This work proposes a novel method to help the model to Coordinate Explicit and Implicit Knowledge (CEIK) for the knowledge-based visual question answering (VQA) task, which is an important direction of pre-trained models. Furthermore, a better training signal is proposed for the retriever according to whether the retrieved knowledge can correct the prediction. Experimental results demonstrate the effectiveness of our method.

The accuracy of AI-assisted chatbots on the annual assessment test for emergency medicine residents

BackgroundThe incorporation of natural language processing (NLP) models into medical education has accelerated with the introduction of ChatGPT. This study aimed to evaluate and compare the performance of ChatGPT-3.5, Bard and the residents in the annual assessment test for emergency medicine (EM) residents. MethodA total of 90 questions covering 18 distinct topics within the field of EM were asked to residents. The same questions were directed to ChatGPT-3.5 and Bard without excluding those with images. The percentage of correct answers were calculated and represented through a histogram showcasing the distribution of test scores across various bins. Questions were categorized as medical knowledge and clinical reasoning to further assess chatbots' performance based on question types. ResultsChatGPT-3.5 demonstrated a 60% accuracy, securing the 10th position, while Bard achieved an accuracy of 55.5%, placing 21st in the rankings among 46 residents. ChatGPT-3.5 performed better in the 16 subtopics. Bard outperformed ChatGPT-3.5 only on cardiovascular and pulmonary emergencies, accounting 24.4% of questions. Analysis of question types revealed ChatGPT-3.5's higher accuracy in medical knowledge (66%) compared to Bard (56%), while Bard performed better in clinical reasoning (55%) than ChatGPT-3.5 (52.5%). Despite lacking access to ECG images, both models answered several questions by providing text-based ECG interpretations. ConclusionChatGPT-3.5 and Bard demonstrated impressive performance in the task of medical question answering. On the other hand, issues related to ECG interpretations raises questions about the reliability of these models. Our findings highlight the importance of verifying the outputs generated by these models.

Genre: generative multi-turn question answering with contrastive learning for entity–relation extraction

Extractive approaches have been the mainstream paradigm for identifying overlapping entity–relation extraction. However, limited by their inherently methodological flaws, which hardly deal with three issues: hierarchical dependent entity–relations, implicit entity–relations, and entity normalization. Recent advances have proposed an effective solution based on generative language models, which cast entity–relation extraction as a sequence-to-sequence text generation task. Inspired by the observation that humans learn by getting to the bottom of things, we propose a novel framework, namely GenRE, Generative multi-turn question answering with contrastive learning for entity–relation extraction. Specifically, a template-based question prompt generation first is designed to answer in different turns. We then formulate entity–relation extraction as a generative question answering task based on the general language model instead of span-based machine reading comprehension. Meanwhile, the contrastive learning strategy in fine-tuning is introduced to add negative samples to mitigate the exposure bias inherent in generative models. Our extensive experiments demonstrate that GenRE performs competitively on two public datasets and a custom dataset, highlighting its superiority in entity normalization and implicit entity–relation extraction. (The code is available at https://github.com/lovelyllwang/GenRE).

Answering List-Type Questions in Health Domain with Pretrained Large Language Model: A Case for COVID-19 Symptoms.

List-type questions, which can have a varying number of answers, are more common in the health domain where people seek for health-related information from a passage or passages. An example of this type of question answering task is to find COVID-19 symptoms from a Twitter post. However, due to the lack of annotated instances for supervised learning, automatic identification of COVID-19 symptoms from Twitter posts is challenging. We investigated detection of symptom mentions in Twitter posts using GPT-3, a pre-trained large language model, along with few-shot learning. Our results of 5-shot and 10-shot learning on a corpus of 655 annotated tweets demonstrate that few-shot learning with pre-trained large language model is a promising approach to answering list-type questions with a minimal amount of effort of annotation.

Industry-sensitive language modeling for business

We introduce BusinessBERT, a new industry-sensitive language model for business applications. The key novelty of our model lies in incorporating industry information to enhance decision-making in business-related natural language processing (NLP) tasks. BusinessBERT extends the Bidirectional Encoder Representations from Transformers (BERT) architecture by embedding industry information during pretraining through two innovative approaches that enable BusinessBert to capture industry-specific terminology: (1) BusinessBERT is trained on business communication corpora totaling 2.23 billion tokens consisting of company website content, MD&A statements and scientific papers in the business domain; (2) we employ industry classification as an additional pretraining objective. Our results suggest that BusinessBERT improves data-driven decision-making by providing superior performance on business-related NLP tasks. Our experiments cover 7 benchmark datasets that include text classification, named entity recognition, sentiment analysis, and question-answering tasks. Additionally, this paper reduces the complexity of using BusinessBERT for other NLP applications by making it freely available as a pretrained language model to the business community. The model, its pretraining corpora and corresponding code snippets are accessible via https://github.com/pnborchert/BusinessBERT.

EGLR: Two-staged Explanation Generation and Language Reasoning framework for commonsense question answering

In the field of natural language processing, strong commonsense reasoning abilities are often required for intelligent systems to excel in the task of commonsense question answering (QA). To enhance the interpretability of QA systems, a natural approach is to generate textual explanations in addition to the predicted answers, ensuring the understandability of the results. Recent efforts address this by using a prompted language model (LM) with frozen parameters to generate explanations in a few-shot manner. These explanations are then used as additional context to guide a finetuned LM in making the final decision. However, these methods still underutilize the semantic information embedded in the explanatory text. Consequently, the reasoning models tend to rely on word co-occurrence and the knowledge stored in the model itself rather than fully exploiting the explanations. Therefore, we propose a two-staged Explanation Generation and Language Reasoning Framework (EGLR), Our framework takes advantage of the in-context learning capability of LMs to generate explanations and reformulates the reasoning task based on explanations as a semantic matching problem. Through joint prompting and training, our model can select the most appropriate explanation by comparing multiple explanations. Experimental results on three public datasets demonstrate that our framework achieves superior performance on the full dataset while maintaining performance on out-of-domain scenarios.

VL-Meta: Vision-Language Models for Multimodal Meta-Learning

Multimodal learning is a promising area in artificial intelligence (AI) that can make the model understand different kinds of data. Existing works are trying to re-train a new model based on pre-trained models that requires much data, computation power, and time. However, it is difficult to achieve in low-resource or small-sample situations. Therefore, we propose VL-Meta, Vision Language Models for Multimodal Meta Learning. It (1) presents the vision-language mapper and multimodal fusion mapper, which are light model structures, to use the existing pre-trained models to make models understand images to language feature space and save training data, computation power, and time; (2) constructs the meta-task pool that can only use a small amount of data to construct enough training data and improve the generalization of the model to learn the data knowledge and task knowledge; (3) proposes the token-level training that can align inputs with the outputs during training to improve the model performance; and (4) adopts the multi-task fusion loss to learn the different abilities for the models. It achieves a good performance on the Visual Question Answering (VQA) task, which shows the feasibility and effectiveness of the model. This solution can help blind or visually impaired individuals obtain visual information.

Graph Pooling Inference Network for Text-based VQA

Effectively leveraging objects and optical character recognition (OCR) tokens to reason out pivotal scene text is critical for the challenging Text-based Visual Question Answering (TextVQA) task. Graph-based models can effectively capture the semantic relationship among visual entities (objects and tokens) and report remarkable performance in TextVQA. However, previous efforts usually leverage all visual entities and ignore the negative effect of superfluous entities. This article presents a Graph Pooling Inference Network (GPIN), which is an evolutionary graph learning method to purify the visual entities and capture the core semantics. It is observed that the dense distribution of reduplicative objects and the crowd of semantically dependent OCR tokens usually co-exist in the image. Motivated by this, GPIN adopts an adaptive node dropping strategy to dynamically downscale semantically closed nodes for graph evolution and update. To deepen the comprehension of scene text, GPIN is a dual-path hierarchical graph architecture that progressively aggregates the evolved object graph and the evolved token graph semantics into a graph vector that serves as visual cues to facilitate the answer reasoning. It can effectively eliminate object redundancy and enhance the association of semantically continuous tokens. Experiments conducted on TextVQA and ST-VQA datasets show that GPIN achieves promising performance compared with state-of-the-art methods.

Vietnamese Legal Text Retrieval based on Sparse and Dense Retrieval approaches

We introduce the combination of two techniques: Sparse Retrieval and Dense Retrieval, while experimenting with different training approaches to find the optimal method for the Vietnamese Legal Text Retrieval task. Moreover, the Question Answering task was only built on the open domain of UIT-ViQuAD but shown promising results on the in-domain legal dataset. Finally, we also mentioned the data augmentation of legal documents up to 3GB to train the Phobert language model, improve this backbone with Condenser, Cocondenser in this paper. Furthermore, these techniques can be utilized for other information retrieval assignments in languages with limited resources.