Answering System Research Articles

Measuring Convective Organization

Abstract Organized systems of deep convective clouds are often associated with high-impact weather and changes in such systems may have implications for climate sensitivity. This has motivated the derivation of many organization indices that attempt to measure the level of deep convective aggregation in models and observations. Here we conduct a comprehensive review of existing methodologies and highlight some of their drawbacks, such as only measuring organization in a relative sense, being biased toward particular spatial scales, or being very sensitive to the details of the calculation algorithm. One widely used metric, Iorg, uses statistics of nearest-neighbor distances between convective storms to address the first of these concerns, but we show here that it is insensitive to organization beyond the meso-β scale and very contingent on the details of the implementation. We thus introduce a new and complementary metric, Lorg, based on all-pair convective storm distances, which is also an absolute metric that can discern regular, random, and clustered cloud scenes. It is linearly sensitive to spatial scale in most applications and robust to the implementation methodology. We also derive a discrete form suited to gridded data and provide corrections to account for cyclic boundary conditions and finite, open boundary domains of nonequal aspect ratios. We demonstrate the use of the metric with idealized synthetic configurations, as well as model output and satellite rainfall retrievals in the tropics. We claim that this new metric usefully supplements the existing family of indices that can help to understand convective organization across spatial scales. Significance Statement The clustering and organization of convection is associated with high-impact weather and changes could impact climate sensitivity, but no consensus exists on how to best measure organization. Here we suggest a new metric that is robust to the calculation details and can classify scenes as random, clustered, or regular. This new metric can therefore account for spacing of organized convective systems and convective storms on scales spanning tens of kilometers to the entire tropics. We suggest that the new metric Lorg addresses many shortcomings of existing measures and can act as a useful additional tool to further understanding of convective organization.

Knowledge-based Visual Question Answering about Named Entities

This thesis is positioned at the intersection of several research fields, Natural Language Processing, Information Retrieval (IR) and Computer Vision, which have unified around representation learning and pre-training methods. We have defined and studied a new multimodal task: Knowledge-based Visual Question Answering about Named Entities (KVQAE). We were particularly interested in cross-modal interactions and different ways of representing named entities. We also focused on data used to train and, more importantly, evaluate Question Answering systems through different metrics. We annotated a dataset for this purpose, the first in KVQAE comprising various types of entities. We also defined an experimental framework for dealing with KVQAE in two stages through an unstructured knowledge base and identified IR as the main bottleneck of KVQAE, especially for questions about non-person entities. To improve the IR stage, we studied different multimodal fusion methods, which are pre-trained through an original task: the Multimodal Inverse Cloze Task. We found that these models leveraged a cross-modal interaction that we had not originally considered, and which may address the heterogeneity of visual representations of named entities. These results were strengthened by a study of the CLIP model, which allows this cross-modal interaction to be modeled directly. Awarded by : Université Paris-Saclay, Orsay, France on 8 November 2023. Supervised by : Olivier Ferret and Camille Guinaudeau. Available at : https://www.theses.fr/s247993.

Answering Binary Causal Questions Using Role-Oriented Concept Embedding

Answering binary causal questions is a challenging task, and it requires rich background knowledge to answer such questions. Extracting useful causal features from the background knowledge base and applying them effectively in a model is a crucial step to answering binary causal questions. The state-of-the-art approaches apply deep learning techniques to answer binary causal questions. In these approaches, candidate concepts are often embedded into vectors to model causal relationships among them. However, a concept may play the role of a cause in one question, but it could be an effect in another question. This aspect has not been extensively explored in existing approaches. Role-oriented causal concept embeddings are proposed in this paper to model causality between concepts. We also propose leveraging semantic concept similarity to extract causal information from concepts. Finally, we develop a deep learning framework to answer binary causal questions. Our approach yields accuracy that is comparable to or better than the benchmark approaches. <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Impact Statement</u> —Understanding causality is crucial for automatic question-answering systems, which are useful in extracting and distributing human knowledge. An automatic question-answering system with causal knowledge can be used to check whether there is causal relationship between two concepts. Existing approaches to answer binary causal questions often answer such questions with close to 55% accuracy due to the limited usage of causal and contextual features. The deep learning framework we propose in this paper uses a role-oriented concept embedding to address such issues. Our approach achieves better accuracy by up to 3.6%, compared to the state-of-the-art benchmark approaches. The proposed approach can be used in a variety of fields, including prescriptive analysis, event prediction, and any other area where entity relationships are essential. It could also be used to improve the retrieval of causality-related inquiries in web search engines.

Aspects of creating a corporate question-and-answer system using generative pre-trained language models

The article describes various ways to use generative pre-trained language models to build a corporate question-and-answer system. A significant limitation of the current generative pre-trained language models is the limit on the number of input tokens, which does not allow them to work "out of the box" with a large number of documents or with a large document. To overcome this limitation, the paper considers the indexing of documents with subsequent search query and response generation based on two of the most popular open source solutions at the moment – the Haystack and LlamaIndex frameworks. It has been shown that using the open source Haystack framework with the best settings allows you to get more accurate answers when building a corporate question-and-answer system compared to the open source LlamaIndex framework, however, requires the use of an average of several more tokens. The article used a comparative analysis to evaluate the effectiveness of using generative pre-trained language models in corporate question-and-answer systems using the Haystack and Llamaindex frameworks. The evaluation of the obtained results was carried out using the EM (exact match) metric. The main conclusions of the conducted research on the creation of question-answer systems using generative pre-trained language models are: 1. Using hierarchical indexing is currently extremely expensive in terms of the number of tokens used (about 160,000 tokens for hierarchical indexing versus 30,000 tokens on average for sequential indexing), since the response is generated by sequentially processing parent and child nodes. 2. Processing information using the Haystack framework with the best settings allows you to get somewhat more accurate answers than using the LlamaIndex framework (0.7 vs. 0.67 with the best settings). 3. Using the Haystack framework is more invariant with respect to the accuracy of responses in terms of the number of tokens in the chunk. 4. On average, using the Haystack framework is more expensive in terms of the number of tokens (about 4 times) than the LlamaIndex framework. 5. The "create and refine" and "tree summarize" response generation modes for the LlamaIndex framework are approximately the same in terms of the accuracy of the responses received, however, more tokens are required for the "tree summarize" mode.

On sharp estimates for Schrödinger groups of fractional powers of nonnegative self-adjoint operators

Let L be a non negative, selfadjoint operator on L2(X), where X is a metric space endowed with a doubling measure. Consider the Schrödinger group for fractional powers of L. If the heat flow e−tL satisfies suitable conditions of Davies–Gaffney type, we obtain the following estimate in Hardy spaces associated to L:‖(I+L)−β/2eiτLγ/2f‖HLp(X)≤C(1+|τ|)nsp‖f‖HLp(X) where p∈(0,1], γ∈(0,1], β/γ=n|12−1p|=nsp and τ∈R.If in addition e−tL satisfies a localized Lp0→L2 polynomial estimate for some p0∈[1,2), we obtain‖(I+L)−β/2eiτLγ/2f‖p0,∞≤C(1+|τ|)nsp0‖f‖p0,∀τ∈R, provided 0<γ≠1, β/γ=n|12−1p|=nsp and τ∈R. By interpolation, the second estimate implies also, for all p∈(p0,p0′), the strong (p,p) type estimate‖(I+L)−β/2eiτLγ/2f‖p≤C(1+|τ|)nsp‖f‖p. The applications of our theory span a diverse spectrum, ranging from the Schrödinger operator with an inverse square potential to the Dirichlet Laplacian on open domains. It showcases the effectiveness of our theory across various settings.

The Formal Layer of {Brain and Mind} and Emerging Consciousness in Physical Systems

AbstractWe consider consciousness attributed to systems in space-time which can be purely physical without biological background and focus on the mathematical understanding of the phenomenon. It is shown that the set theory based on sets in the foundations of mathematics, when switched to set theory based on ZFC models, is a very promising mathematical tool in explaining the brain/mind complex and the emergence of consciousness in natural and artificial systems. We formalise consciousness-supporting systems in physical space-time, but this is localised in open domains of spatial regions and the result of this process is a family of different ZFC models. Random forcing, as in set theory, corresponds precisely to the random influence on the system of external stimuli, and the principles of reflection of set theory explain the conscious internal reaction of the system. We also develop the conscious Turing machines which have their external ZFC environment and the dynamics is encoded in the random forcing changing models of ZFC in which Turing machines with oracles are formulated. The construction is applied to cooperating families of conscious agents which, due to the reflection principle, can be reduced to the implementation of certain concurrent games with different levels of self-reflection.

NLP and its Components: A Detailed Discussion

Natural Language Processing (NLP) encompasses computational techniques for processing and analyzing human language, primarily through Natural Language Understanding (NLU) and Natural Language Generation (NLG). NLU focuses on interpreting language by analyzing phonology (sounds), morphology (word structures), syntax (sentence structures), semantics (meaning), and pragmatics (context). These processes enable machines to comprehend the nuances of human language for accurate interpretation and response generation. NLG, in contrast, involves producing human-like text from structured data. It includes content determination (identifying relevant information), text planning (organizing information), sentence planning (constructing grammatically correct sentences), and surface realization (generating the final text). NLG is crucial for applications like automated report generation and chat bots, where coherent and contextually appropriate responses are essential. The synergy between NLU and NLG underpins many NLP applications. In machine translation, NLU interprets the source text, while NLG generates the translated text. In question-answering systems, NLU processes the query, and NLG formulates the response. Deep learning advancements have significantly enhanced both NLU and NLG, enabling more sophisticated and human-like interactions. This paper explores the components and processes of NLU and NLG, offering a comprehensive understanding of their mechanisms and the advancements driving modern NLP systems.

Research on the Innovative Path of Ideological and Political Education in Colleges and Universities in the Context of Deep Learning

Abstract To construct an automatic question-and-answer system for Civics, a neural network and a self-attention mechanism are combined in this paper. Through the input layer of the neural network, the model remembers the semantic information of the Civics text that has been processed. Dynamic feature extraction is used to categorize and classify Civics questions in the system and determine the type of questions. The GLU network is used to predict the answers to the Civics and Politics questions, and the predicted results are normalized by combining the Softmax function, and the category with the highest probability is selected as the answer. The main mode of teaching afterward was to combine the automatic question-and-answer system with teaching. The results show that in the question and answer system, the highest accuracy rate is the practical cognition type of questions, with the accuracy rate fluctuating in the range of (0.8-0.9), and the accuracy rate of truth type and method type of questions fluctuates up and down in the range of 0.7. The percentage of students in the high marking stage increased by 0.1. In Civic Behavior, students’ patriotism increased by 0.15, and spirituality increased by 0.3.

DEC-NWD: an approach for open set domain adaptation in fault diagnosis

The issue of data-driven cross-domain fault diagnosis for rolling bearings has been effectively addressed through advancements in domain adaptation (DA) methods. However, most existing approaches assume the same set of labels for training data and test data. This assumption often falls short of reality, as new fault types may emerge during the testing phase, resulting in less effective DA methods based on marginal distribution. To address this issue, this study proposes an open set DA method based on domain similarity, entropy, confidence, and nuclear-norm 1-Wasserstein discrepancy (NWD). Within this method, a sample-level transferability criterion is introduced. This criterion quantifies the transferability of target samples and assigns small weights to the unknown class. The complementary nature of entropy and confidence is exploited to improve the discriminability of the network for highly uncertain predictions and to use multiple classifiers to compensate for the possible influence of prediction errors on confidence. Additionally, the NWD is utilized in this method. It treats the classifier as a discriminator and leverages the predicted discriminative information to maximize the alignment of the common classes between the source and target domains. The proposed method has been validated through extensive experiments conducted on two publicly available bearing datasets.

Prevalence and Risk Factors of Urinary Tract Infection in Pregnant Women: Systematic Review and Meta-Analysis

Background: Urinary Tract Infections (UTIs) are the second-most frequent bacterial infections, with an overall incidence of 5-10%. This study aimed to investigate the prevalence and risk factors associated with UTI in pregnant women. Method: A systematic search was conducted in the open domain, electronic databases such as PubMed and Google Scholar. Citation searching was also performed to identify all relevant studies. All the steps were carried out by two independent reviewers to avoid bias. Based on inclusion &amp; exclusion criteria, followed by qualitative evaluation, the studies were included and analyzed by fixed or random effects model using CMA. A sub-group analysis was also performed for Asian and African countries for prevalence and for different categories of risk factors. Results: Out of 45 studies, 41 studies were included for the prevalence of UTI in pregnant women and 19 studies were included to evaluate the risk factors. The pooled prevalence of UTI in African pregnant women was higher when compared to Asian pregnant women (28.6% vs. 20.7%). The risk factors statistically significantly associated with UTI were found to be undergarment texture (silk), history of catheterization, diabetes mellitus, past history of UTI and HIV. Conclusion: UTIs are prevalent in pregnant women in Asian and African pregnant women. Hence, identification of risk factors and frequent screening of UTIs during pregnancy is recommended to avoid maternal-fetal complications.

The power and potentials of Flexible Query Answering Systems: A critical and comprehensive analysis

The popularity of chatbots, such as ChatGPT, has brought research attention to question answering systems, capable to generate natural language answers to user’s natural language queries. However, also in other kinds of systems, flexibility of querying, including but also going beyond the use of natural language, is an important feature. With this consideration in mind the paper presents a critical and comprehensive analysis of recent developments, trends and challenges of Flexible Query Answering Systems (FQASs). Flexible query answering is a multidisciplinary research field that is not limited to question answering in natural language, but comprises other query forms and interaction modalities, which aim to provide powerful means and techniques for better reflecting human preferences and intentions to retrieve relevant information. It adopts methods at the crossroad of several disciplines among which Information Retrieval (IR), databases, knowledge based systems, knowledge and data engineering, Natural Language Processing (NLP) and the semantic web may be mentioned. The analysis principles are inspired by the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) framework, characterized by a top-down process, starting with relevant keywords for the topic of interest to retrieve relevant articles from meta-sources And complementing these articles with other relevant articles from seed sources Identified by a bottom-up process. to mine the retrieved publication data a network analysis is performed Which allows to present in a synthetic way intrinsic topics of the selected publications. issues dealt with are related to query answering methods Both model-based and data-driven (the latter based on either machine learning or deep learning) And to their needs for explainability and fairness to deal with big data Notably by taking into account data veracity. conclusions point out trends and challenges to help better shaping the future of the FQAS field.

Physics-informed neural network combined with characteristic-based split for solving Navier–Stokes equations

This paper presents a novel approach for solving the shallow-water transport equation using a physics-informed neural network (PINN) combined with characteristic-based split (CBS). Our method separates the variables, so it is no need to consider the weight coefficients between different loss functions. As not all partial derivatives are involved in the gradient backpropagation, our method can save half of the memory occupation without losing accuracy, and resulting in a significant reduction in computation time compared to traditional PINN. We solving the progress of the dispersion of hot water under known flow fields. Furthermore, we propose a boundary condition that accounts for the second-order partial derivative term, which is more appropriate for solving the diffusion equation with open domains than the commonly used assumption of zero boundary values. Our numerical results demonstrate that this boundary condition leads to improved convergence of the network. In addition, we introduce a parameter estimation method to estimate the diffusion coefficient of hot water flow, which requires information from the field at only two different times. We observe that excessive participation of variables in gradient backpropagation can lead to neural networks getting trapped in local optima. We use PINN combined with CBS method to solve 3-D incompressible flow. As the number of variables involved in gradient backpropagation increases, the accuracy of the solution decreases, which can partially support our viewpoint. The source codes for the numerical examples in this work are available at https://github.com/double110/PINN-cbs-.git.

MetaQA: Enhancing human-centered data search using Generative Pre-trained Transformer (GPT) language model and artificial intelligence.

Accessing and utilizing geospatial data from various sources is essential for developing scientific research to address complex scientific and societal challenges that require interdisciplinary knowledge. The traditional keyword-based geosearch approach is insufficient due to the uncertainty inherent within spatial information and how it is presented in the data-sharing platform. For instance, the Gulf of Mexico Coastal Ocean Observing System (GCOOS) data search platform stores geoinformation and metadata in a complex tabular. Users can search for data by entering keywords or selecting data from a drop-down manual from the user interface. However, the search results provide limited information about the data product, where detailed descriptions, potential use, and relationship with other data products are still missing. Language models (LMs) have demonstrated great potential in tasks like question answering, sentiment analysis, text classification, and machine translation. However, they struggle when dealing with metadata represented in tabular format. To overcome these challenges, we developed Meta Question Answering System (MetaQA), a novel spatial data search model. MetaQA integrates end-to-end AI models with a generative pre-trained transformer (GPT) to enhance geosearch services. Using GCOOS metadata as a case study, we tested the effectiveness of MetaQA. The results revealed that MetaQA outperforms state-of-the-art question-answering models in handling tabular metadata, underlining its potential for user-inspired geosearch services.

Improved meta-analysis pipeline ameliorates distinctive gene regulators of diabetic vasculopathy in human endothelial cell (hECs) RNA-Seq data.

Enormous gene expression data generated through next-generation sequencing (NGS) technologies are accessible to the scientific community via public repositories. The data harboured in these repositories are foundational for data integrative studies enabling large-scale data analysis whose potential is yet to be fully realized. Prudent integration of individual gene expression data i.e. RNA-Seq datasets is remarkably challenging as it encompasses an assortment and series of data analysis steps that requires to be accomplished before arriving at meaningful insights on biological interrogations. These insights are at all times latent within the data and are not usually revealed from the modest individual data analysis owing to the limited number of biological samples in individual studies. Nevertheless, a sensibly designed meta-analysis of select individual studies would not only maximize the sample size of the analysis but also significantly improves the statistical power of analysis thereby revealing the latent insights. In the present study, a custom-built meta-analysis pipeline is presented for the integration of multiple datasets from different origins. As a case study, we have tested with the integration of two relevant datasets pertaining to diabetic vasculopathy retrieved from the open source domain. We report the meta-analysis ameliorated distinctive and latent gene regulators of diabetic vasculopathy and uncovered a total of 975 i.e. 930 up-regulated and 45 down-regulated gene signatures. Further investigation revealed a subset of 14 DEGs including CTLA4, CALR, G0S2, CALCR, OMA1, and DNAJC3 as latent i.e. novel as these signatures have not been reported earlier. Moreover, downstream investigations including enrichment analysis, and protein-protein interaction (PPI) network analysis of DEGs revealed durable disease association signifying their potential as novel transcriptomic biomarkers of diabetic vasculopathy. While the meta-analysis of individual whole transcriptomic datasets for diabetic vasculopathy is exclusive to our comprehension, however, the novel meta-analysis pipeline could very well be extended to study the mechanistic links of DEGs in other disease conditions.

NIR-Prompt: A Multi-task Generalized Neural Information Retrieval Training Framework

Information retrieval aims to find information that meets users’ needs from the corpus. Different needs correspond to different IR tasks such as document retrieval, open-domain question answering, retrieval-based dialogue, and so on, while they share the same schema to estimate the relationship between texts. It indicates that a good IR model can generalize to different tasks and domains. However, previous studies indicate that state-of-the-art neural information retrieval (NIR) models, e.g., pre-trained language models (PLMs) are hard to generalize. It is mainly because the end-to-end fine-tuning paradigm makes the model overemphasize task-specific signals and domain biases but loses the ability to capture generalized essential signals. To address this problem, we propose a novel NIR training framework named NIR-Prompt for retrieval and reranking stages based on the idea of decoupling signal capturing and combination. NIR-Prompt exploits Essential Matching Module (EMM) to capture the essential matching signals and gets the description of tasks by Matching Description Module (MDM). The description is used as task-adaptation information to combine the essential matching signals to adapt to different tasks. Experiments under in-domain multi-task, out-of-domain multi-task, and new task adaptation settings show that NIR-Prompt can improve the generalization of PLMs in NIR for both retrieval and reranking stages compared with baselines.

Novel approach to integrate various feature extraction techniques for the Visual Question Answering System with skeletal images in the healthcare sector

In the realm of medical science, one of the most challenging concepts to grasp is the Medical Imaging Query Response System. The comprehension and classification of the diverse representations of the human body require a significant degree of effort and expertise. Furthermore, it is imperative for users within the healthcare sector to rigorously validate the system. In the domain of human health, a plethora of imaging techniques, including MRI, CT, ultrasound, X-ray, PET-CT, and others, play a pivotal role in the identification of medical issues. These technologies are instrumental in supporting both patient engagement and clinical decision-making. However, the utilization of models, techniques, and datasets for processing textual and visual information introduces complexities that can at times impede the provision of pertinent clinical solutions. The overarching objective of the proposed approach is to conduct a comprehensive comparative analysis of various feature extraction methodologies for both visual and textual information within the Visual Question Answering (VQA) system, focusing on human skeletal images. This endeavor is aimed at enhancing the VQA system's performance with newer datasets and addressing any limitations inherent in existing models. In addition, this research initiative seeks to enable researchers to identify and optimize novel methods that enhance the accuracy of the VQA system. The models under scrutiny in this analysis encompass various methods of feature extraction that help to improve the model and quality of the healthcare industry. The researcher will find the proper methodology for different datasets. To gauge the efficacy of each model in delivering the desired outcomes, an array of metrics will be employed, including classification measurement accuracy, F-classification, C-true positive rate (CTPR), C-precision, C-recall, C-sensitivity, and C-false negative rate (FNR). These metrics are designed to enhance the accuracy of any dataset and optimize the performance of both visual and textual components to ensure accurate responses to the posed queries.

Usefulness of open domain model for identifying missing software requirements concepts

SummaryDetecting missing requirements during software development is crucial to avoid unexpected consequences. However, this task is challenging due to limited domain knowledge of requirements analysts and the dynamic nature of software requirements. Previous studies have shown that requirement‐oriented domain models can help identify omissions in requirements, but they are often incomplete for many domains. Meanwhile, domain models constructed from other artifacts are available online. This raises the question: Can these domain models be useful in identifying missing functional information in requirement specifications? To address this question, we conducted a study to measure the overlap between entities in domain models and requirements. We analyzed the occurrence of overlapped entities, considering four distribution characteristics: the type of entities in the domain model, the distribution of mapped entities in the domain model, the family belonging of the mapped entities in the domain model, and the distribution of mapped entities in the requirements. Based on our findings, we proposed recommendations for missing requirements. Additionally, we performed experiments, including the use of the proposed metric “ancestors of the highest level with the most mapped entities” (AHME). The results showed significant improvements with gains of 146% and 223% in the two domains, highlighting the benefits of these distribution characteristics.

Span prompt dense passage retrieval for Chinese open domain question answering

Dense passage retrieval is a popular method in information retrieval recently, especially in open domain question answering. It aims to retrieve related articles from massive passages to answer the question. Retriever can increase retrieval speed with less loss of accuracy compared to other methods. However, the pretrained language models used in recent research are often ineffective in semantic embedding, which will reduce accuracy. In addition, we find that contrastive learning will diverge the representation space, and Siamese models with independent parameters on both sides will decrease generalization performance. Therefore, we propose span prompt dense passage retrieval (SPDPR) based on span mask prompt tuning and parameter sharing in Chinese open-domain dense retrieval. This model can generate more efficient representation embeddings and effectively counteract the separation tendency between positive samples. We evaluate the effectiveness of SPDPR in DYKzh, as well as two Chinese datasets. SPDPR surpasses all SOTAs implemented in DYKzh and achieves a competitive result in other datasets.

Representation-centric approach for classification of Consumer Health Questions

Question Classification is the very first and pivotal step in Question Answering(QA) systems. It maps a given question into a predefined class. Classification of Consumer Health Questions contributes to answering medical queries of non-expert users. It plays a principal role in the performance of Consumer Health Question Answering systems. There are several approaches to the classification of questions. However, they do not consider the nature of Consumer Health Questions. In this study, we perform experiments focusing on the question representation methods to classify Consumer Health Questions accurately. These experiments provide an intuition of practical considerations for Consumer Health Question classification. The questions were represented through certain manually-designed features, word embedding, sentence embedding, and finetuning-based techniques. We create question classifiers that involve designing models for representing and categorizing questions. Based on the result, the fine-tuned Bidirectional Encoder Representations from Transformers(BERT)-large-based model with an accuracy of 86.00% on Genetic and Rare Diseases (GARD) and 80.40% on Yahoo! Answers-based datasets outperform previous works. In addition, the method based on A Robustly Optimized BERT Pretraining Approach(RoBERTa)-large model with 86.20% accuracy surpasses fine-tuned BERT-large model on the GARD dataset. Examining the outcomes of the models gives insight into decisive considerations for representing and classifying Consumer Health Questions.

CLRN: A reasoning network for multi-relation question answering over Cross-lingual Knowledge Graphs

Cross-lingual Knowledge Graphs-based Question Answering (CLKGQA) requires the question answering (QA) system to combine the knowledge graphs (KGs) in different languages to obtain answers to input questions. In previous works, the common idea is to merge Cross-lingual Knowledge Graphs (CLKGs) into a single KG through aligned entity pairs and then treat it as a traditional KG-based QA. However, as demonstrated by Tan et al. (2023), existing Entity Alignment (EA) models cannot generate highly accurate aligned entity pairs for CLKGs. Therefore, two issues need to be addressed in the CLKGQA task: (1) Remove the dependency of the QA model on the fused KG; (2) Improve the performance of the EA model in obtaining aligned entity pairs from locally isomorphic CLKGs. To solve the above two issues, this paper presents Cross-lingual Reasoning Network (CLRN), a novel multi-hop QA model that allows switching knowledge graphs at any stage of the multi-hop reasoning. Furthermore, we establish an iterative framework that combines CLRN and EA model, in which CLRN is used for extracting potential alignment triple pairs from CLKGs during the QA process. The extracted triple pairs provide pseudo-aligned entities, and the additional aligned entity pairs are used to mine missing relations between entities in CLKGs. These pseudo-aligned entity pairs and relations improve the performance of the EA model, resulting in higher accuracy in QA. Extensive experiments demonstrate the effectiveness of the proposed model, which outperforms the baseline approaches. Through iterative enhancement, the performance of the EA model has also been improved by > 1.0 % in [email protected] and [email protected], and the improvement is statistically significant in the confidence interval of p<0.01. Moreover, our work discusses the correlation between QA and EA from the side of QA, which has reference value for the follow-up exploration of related communities. We have open-sourced our dataset and code, which is available at the URL https://github.com/tan92hl/Cross-lingual-Reasoning-Network-for-CLKGQA.