Reasoning to answer logical queries over a knowledge graph has been paid attention to research recently. Previous studies have only used either distance measurement or machine learning approaches processed logical operators such as conjunction, disjunction, and negation but considering the entity description information of knowledge graph yet. This study proposes a model, named QUERY2TREE, to combine a graph representative learning model and large language model to augment semantic for embedded entities of the knowledge graph such that these entities can handle logical operations more accurately during the reasoning process. Specifically, the QUERY2TREE used the graph neural network (GNN) to embed entities and Gemma model, a family of Google Gemini model, to embed the entity's descriptive information into low-dimensional space. Then, these embedded entities were indexed by K-D tree according to each relation over the knowledge graph. Next, we used nearest neighbor search algorithm of K-D tree to build logical operations such as projection, intersection, union, and negation. Finally, we applied these created logical operations to infer the answer to complex questions. We experimented with our model over benchmark knowledge graphs such as FB15k, FB15k-237, and NELL995. As a result, the QUERY2TREE model improved the accuracy better than the baseline models QUERY2BOX, ConE, CQD-CO, MLP, SILR, GNN-QE, CKGR, LACK, and LACT by up to 69.6

Answering logical questions with a knowledge graph has been a critical research focus because this needs to reason and synthesize information. Previous studies have mainly dealt with logical operations using graph embedding techniques, such as conjunctions, disjunctions, and negation. However, these studies have neither effectively organized the data to retrieve multi-hop reasoning quickly nor combined text description to enhance logical operations’ semantics. Our study introduces a model called QUERY2BERT, which solves two of the above limitations. Specifically, QUERY2BERT first combined the node2vec and the BERT models to embed a knowledge graph with description information of every entity. Then, embedded nodes were indexed with a K-D tree structure. Finally, we used nearest neighbor search on K-D tree to retrieve neighbor-embedded nodes and implemented logical operations like projection, intersection, union, and negation to find answers to complex questions. We tested our model on three benchmark knowledge graph datasets and showed that QUERY2BERT significantly improved accuracy and speed compared to other state-of-the-art models.

An automated query answering system (QAS) is a very useful application in organizations. Therefore, there is a lot of research to build, develop and improve it. In this paper, we present a method to build a multi-hop query answering system (MQAS) based on a large knowledge graph (KG), Bidirectional Encoder Representations from Transformers (BERT), and the indexing structure K-Dimensional Tree (KD-Tree). The large KG provides knowledge for MQAS. BERT is used to transform questions into vectors. KD-Tree helps find the right answers quickly. On the other hand, we also propose a solution for distributed indexing of large vector spaces by building a distributed indexing structure called Distributed KD-Tree (DKD-Tree). In addition, we also present experiments and evaluation results to demonstrate the effectiveness of our solution.

In the digital era, the accuracy and reliability of information are paramount, and fact-checking is essential to achieving this objective. Researching this area presents numerous challenges, particularly for the Vietnamese language, due to the current scarcity of tools and data for Vietnamese fact-checking. To address these challenges and advance fact-checking research both broadly and within the Vietnamese context, this paper introduces a fact-checking model tailored for Vietnamese, named ViKGFC. ViKGFC integrates a Knowledge Graph (KG), inference rules, and the Knowledge graph - Bidirectional Encoder Representations from Transformers (KG-BERT) deep learning model. Its notable capability is the extraction of triples from complex Vietnamese sentences, which significantly enhances information extraction in the Vietnamese language. Additionally, ViKGFC utilizes inference rules to improve the KG’s accuracy and employs matching techniques for rapid verification, thereby aiding in the swift prevention of misinformation spread on media and online platforms. Evaluated on a dataset of 130,190 Vietnamese samples sourced from Wikipedia, ViKGFC attains an outstanding accuracy level, reaching 95%. This proposed method offers an optimistic solution for verifying facts in Vietnamese and could potentially assist in creating fact-checking tools and techniques for other languages. In summary, this study significantly advances data science by introducing a reliable and accurate approach to Vietnamese fact-checking.

Reasoning to answer complex questions is a difficult task and is receiving research attention in natural language processing today. The purpose of this study is to propose a model, QUERY2RULE, that automatically generates logic rules representing the reasoning process and learn how to find answer entities from logical reasoning rules in the graph embedding space. Specifically, for generating automatic logical reasoning rules, we improved the REBEL model by combining LSTM to create automatic logical reasoning rules for each complex question. To learn how to find answers to logic rules, we used the node2vec algorithm to convert the knowledge graph to the embedding space. We added descriptive information about each entity to the graph for more semantics. Then, the K-D tree technique indexed embedded entities for fast retrieval. Finally, we combined the question embedding and the logical reasoning rule embedding and gave them input to the decoder layer to rank the highest candidate entities as the answer. We tested the proposed model on benchmark datasets such as FreeBaseQA, WebQSP, CWQ, and FB15k-237. Experimental results showed that the QUERY2RULE model achieved better than previous models when comparing classification, link prediction, and multi-hop reasoning tasks.

This paper introduces the ViFoodNLI dataset, a natural language inference (NLI) dataset for Vietnamese. While recent efforts have been made to build high-quality NLI datasets for Vietnamese and some Cross-Lingual NLI Corpus (with support for Vietnamese) for multiple domains, our dataset specifically focuses on the field of local cuisine. The main reason for choosing this field is that cuisine is a significant component of Vietnamese culture, and thus the dataset encompasses many characteristics of the Vietnamese language. By collecting information on culinary topics from reliable news sources, we have developed various methods and logics such as knowledge graphs and Generative AI to create high-quality pairs of premise and hypothesis sentences. Through rigorous testing, the dataset has achieved significant results, creating momentum for future research and practical applications.

In this study, we propose a method to automatically create Vietnamese Natural Language Inference (NLI) datasets from Knowledge Graph (KG). The approach leverages information of Knowledge Graph (KG) and employs thesaurus and antonym dictionary expansion techniques to generate premise-hypothesis sentence pairs with labels of entailment, contradiction, and neutral for natural language inference (NLI). The researchers also conducted a process of validating and improving the quality of the generated dataset. The experimental results demonstrate that this method of automatically creating Vietnamese NLI datasets from KG achieves reliable and effective performance. The generated dataset not only expands the scale of existing Vietnamese NLI datasets but also provides valuable resources for training and evaluating Vietnamese NLI models. This method holds the potential for wide-ranging applications in the development of Vietnamese NLP applications and related research.

Tại Việt Nam, có thể nói ngành chăm sóc sắc đẹp là một trong những lĩnh vực kinh doanh có mức độ cạnh tranh cao. Việc tìm hiểu những sản phẩm nào đang được quan tâm tìm kiếm phổ biến trên Google và nắm được số liệu dự đoán tìm kiếm tương lai trên Google giúp cho các nhà đầu tư, những người phụ trách phòng kinh doanh, tiếp thị những thông tin hữu ích để có thể nghiên cứu đưa ra các chiến lược tiếp thị kinh doanh cạnh tranh với đối thủ hoặc các nhà đầu tư cân nhắc đưa ra quyết định đầu tư của mình. Bài báo này thực hiện các phương pháp thu thập, tiền xử lý dữ liệu, phân tích và trích xuất thông tin nội dung của các trang web được phổ biến trên Google. Cuối cùng là dự đoán số liệu tìm kiếm trong tương lai trên Google bằng các thuật toán học máy. Kết quả thực nghiệm đã cho biết các sản phẩm nổi bật và đề xuất mô hình phù hợp dự đoán số liệu tìm kiếm tương lai trên Google.

There are many attempts to implement deep neural network (DNN) distributed training frameworks. In these attempts, Apache Spark was used to develop the frameworks. Each framework has its advantages and disadvantages and needs further improvements. In the process of using Apache Spark to implement distributed training systems, we ran into some obstacles that significantly affect the performance of the systems and programming thinking. This is the reason why we developed our own distributed training framework, called Distributed Deep Learning Framework (DDLF), which is completely independent of Apache Spark. Our proposed framework can overcome the obstacles and is highly scalable. DDLF helps to develop applications that train DNN in a distributed environment (referred to as distributed training) in a simple, natural, and flexible way. In this paper, we will analyze the obstacles when implementing a distributed training system on Apache Spark and present solutions to overcome them in DDLF. We also present the features of DDLF and how to implement a distributed DNN training application on this framework. In addition, we conduct experiments by training a Convolutional Neural Network (CNN) model with datasets MNIST and CIFAR-10 in Apache Spark cluster and DDLF cluster to demonstrate the flexibility and effectiveness of DDLF.

People will increasingly get expedited and diverse means of accessing news as societies progress. Furthermore, there is a noticeable increase in the prevalence of incorrect and misleading information. Our research is motivated by the significant concerns regarding the detrimental impacts of disinformation on the general public, political stability, and trust in the media. The scarcity of Vietnamese-language datasets can be attributed to the predominant focus of false news detection studies on datasets only in English. Detection investigations of fake news have predominantly relied on supervised machine learning algorithms, which possess notable limitations when confronted with unclassified news articles that are either authentic or untrue. The utilization of Knowledge Graphs (KG) and Graph Convolutional Networks (GCN) holds promise in addressing the constraints of supervised machine learning algorithms. To address these problems, we propose an approach that integrates KG)into the procedure for detecting fake news. We utilize the Vietnamese Fake News Detection dataset (VFND-vietnamese-fake-news), comprising authentic and deceptive news articles from reputable Vietnamese newspapers such as vnexpress, tuoitre, and have been collected from 2018 to 2023. News articles are only labeled as real or fake after experiencing independent verification. The Glove embedding (Global Vectors for Word Representation) is employed to establish a knowledge network for the given dataset. This knowledge graph’s construction is accomplished using the Word Mover’s Distance (WMD) algorithm in conjunction with the K-nearest neighbor approach; GCN approach and the input KG train models to discern between real and fake news. With labeling half of the input dataset, the experimental findings indicate a notable level of accuracy, reaching up to 85%. Our research holds significant importance in identifying fake news, particularly within the context of the Vietnamese language.

Detecting fake news problems has been a topic of research noticed in recent years. However, at present, fake news detection research has only been conducted with English data sets, no research has been conducted with Vietnamese datasets. Therefore, it is essentially important and urgent to do research on fake news detection with the Vietnamese dataset. In this research, we proposed a detecting fake news method using a knowledge graph (KG) combined with a semi-supervised learning Graph Convolutional Networks (GCN) to predict whether the news is real or fake. Our research is implemented by collecting data including real and fake news from the datasets of online newspapers in Vietnamese including vnexpress, tuoitre, etc., and VFND-vietnamese-fake-news. In addition, we will not label news that has not been verified for authenticity. We build a knowledge graph for the dataset by embedding words into the dataset using the Glove library and building a knowledge graph using the Word Mover’s Distance algorithm (WMD) combined with the K-nearest-neighbor algorithm (KNN). Detecting real or fake news models with input knowledge graph built and GCN algorithm and finally perform the tests with datasets. With our research, we have built a knowledge graph for the Vietnamese data set of real and fake news collected from 2018 to 2023 combined with the VFND-vietnamese-fake-news dataset. Our proposed method can resolve to detect fake news problems by the GCN algorithm. Research’s results have a high precision of up to 85

This paper focuses on constructing a dataset consisting of both fake news and factual news in the Vietnamese language. We employ Deep Learning models, namely Long Short-Term Memory, bidirectional Long Short-Term Memory, and Convolutional Neural Network - bidirectional Long Short-Term Memory, to identify Vietnamese fake news. The performance evaluation of the models includes assessing the prediction ratio Area Under The Curve of each model and providing insights into their computational efficiency. Additionally, these three models evaluate the contribution of deep learning techniques for fake news detection and emphasize the potential for exploring interconnections between neural networks in addressing automatic Vietnamese fake news detection.

In the era of digital information, ensuring the accuracy and reliability of information is crucial, making fact-checking a vital process. Currently, English fact-checking has thrived due to various language processing tools and ample datasets. However, the same cannot be said for Vietnamese fact-checking, which faces significant challenges due to the lack of such resources. To address these challenges, we propose a model for checking Vietnamese facts by synthesizing three popular technologies: Knowledge Graph (KG), Datalog, and KG-BERT. The KG serves as the foundation for the fact-checking process, containing a dataset of Vietnamese information. Datalog, a logical programming language, is used with inference rules to complete the knowledge within the Vietnamese KG. KG-BERT, a Deep Learning (DL) model, is then trained on this KG to rapidly and accurately classify information that needs fact-checking. Furthermore, to put Vietnamese complex sentences into the fact-checking model, we present a solution for extracting triples from these sentences. This approach also contributes significantly to the ease of constructing foundational datasets for the Vietnamese KG. To evaluate the model's performance, we create a Vietnamese dataset comprising 130,190 samples to populate the KG. Using Datalog, we enrich this graph with additional knowledge. The KG is then utilized to train the KG-BERT model, achieving an impressive accuracy of 95%. Our proposed solution shows great promise for fact-checking Vietnamese information and has the potential to contribute to the development of fact-checking tools and techniques for other languages. Overall, this research makes a significant contribution to the field of data science by providing an accurate solution for fact-checking information in Vietnamese language contexts.

A triple consists of a head entity, relation, and tail entity. Triple is a key component of the knowledge graph. Extracting the triples from a sentence in the Vietnamese corpus is an interesting problem in building knowledge graphs and generating questions for the question answering systems. The previous works mostly have been done by using rules. Little research has applied machine learning models to extract triples from a given sentence. In this paper, we proposed a model named BERT-TRIPLE that is based on the BERT model to extract triples from the Vietnamese corpus. In particular, we performed two phases that were annotation and training. In the annotation phase, our method annotates word tokens in sentences. In the training phase, we fine-tuned the BERT model to train the labeled dataset for predicting triples in Vietnamese sentences. After collecting the triples of Vietnamese text, we generate questions for Question Answering systems. We experiment to prove the precision of our proposed method on the Vietnamese Wikipedia corpus. Our model improved F1 score better than VnCoreNLP and underthesea models with 12% and 17% respectively. Also, our model was more efficient than the Text-to-Text Transfer Transformer (T5) model in the question generation task.

Fact checking is the assessment of the correctness of a claim in a particular context. This is usually done by experts, by drawing on evidences such as speeches, arguments, legal information and previously published figures or known facts, with inference to draw conclusions about the claim. In the world, there have been many researches on the topic of fact checking, mostly based on datasets supporting the English language, using natural language processing techniques to analyze the sentence components in the claim and the evidences, from which there is a basis for comparison to draw conclusions. In Vietnam, researches on checking facts on Vietnamese documents have not been done much, due to the supporting tools for analyzing Vietnamese sentence structures and semantics is still limited and the performance is not high. Therefore, we propose a model to help checking facts with Vietnamese content by combining knowledge graph (KG) with Bidirectional Encoder Representation from Transformer (BERT) deep learning technique. These are advanced techniques that are being used by many researchers. Experimental results on the Vietnamese dataset of 129045 triples extracted from Wikipedia show that our model has a high accuracy up to 96%, and is capable of supporting inference while checking facts.

One of the challenges in constructing Knowledge Graphs from text is verifying the correctness of the produced results. Each language has its unique characteristics, so a Knowledge Graphs construction system may perform better on certain languages and worse on others. In order to detect the most suitable Knowledge Graph construction systems for Vietnamese, in this paper, we propose a method to classify triples extracted from such systems into two categories: Existent and Non-existent. Vietnamese is a low-resource language with limited natural language processing tools and datasets. By combining BERT with a self-constructed Vietnamese Knowledge Graph, we build a classification model to verify the existence of triples in paragraphs. Our results suggest that BERT can learn contextual relations between words from a large amount of text, even for a low-resource language like Vietnamese. BERT’s adaptive capability to detect meaningful triples is also shown and discussed. The outcome of this paper could potentially be used to build more sophisticated systems to solve Knowledge Graph construction and Triple Classification tasks in low resource languages.

Discovering relationships between vertices in a secured information network is an important task in information network analysis. In HIN, meta-path, or a sequence of vertex types and edge types connecting two vertices. Path instance of a meta-path is path in HIN that satisfies the meta-path. The length of meta-path is the number of relations (edges) in this meta-path. Meaningful meta-path is a meta-path with at least one path instance. Recent works on meta-path discovery mainly focus on in-memory algorithms that fit in only one computer. In this chapter, we propose distributed algorithms to discover all shortest meaningful meta-paths between two vertices of a large HIN using Apache Spark. Shortest meaningful meta-path is a meaningful meta-path with shortest length. We employ a scalable implementation of the Distributed Breadth-First Search (D-BFS) algorithm as a baseline approach. Finding all possible shortest paths in a large HIN can be time consuming. Therefore, we propose a novel algorithm called shortest meaningful meta-path based search (S-MPS). S-MPS first searches all shortest meta-path candidates between vertices in the graph of the network schema of HIN. We conduct experiments on DBLP data set to prove the efficiency of our proposed S-MPS algorithm over D-BFS.