The automatic construction of large-scale knowledge graphs has received much attention from both academia and industry in the past few years. Notable graph systems include Google Knowledge Graph, DBPedia, YAGO, NELL, Probase many others. organizes information a structured way by explicitly describing relations among entities. Since entity identification relation extraction are highly depending on language itself, data sources largely determine processed, extracted, ultimately how formed,...

Relation Extraction (RE) is a vital step to complete Knowledge Graph (KG) by extracting entity relations from texts. However, it usually suffers the long-tail issue. This paper proposes novel approach learn relation prototypes unlabeled texts, facilitate RE transferring knowledge types with sufficient training data. We as an implicit factor between entities, which reflects meanings of and their proximities. construct co-occurrence graph capture both first-order second-order proximities for...

Relation extraction (RE) aims at extracting the relation between two entities from text corpora. It is a crucial task for Knowledge Graph (KG) construction. Most existing methods predict an entity pair by learning training sentences, which contain targeted pair. In contrast to distant supervision approaches that suffer insufficient corpora extract relations, our proposal of mining implicit mutual massive unlabeled transfers semantic information pairs into RE model, more expressive and...

Dysarthric speech poses significant challenges for individuals with dysarthria, impacting their ability to communicate socially. Despite the widespread use of Automatic Speech Recognition (ASR), accurately recognizing dysarthric remains a formidable task, largely due limited availability data. To address this gap, we developed Chinese Dysarthria Database (CDSD), most extensive collection dysarthria data date, featuring 133 hours recordings from 44 speakers. Our benchmarks reveal best...

Named Entity Recognition (NER) on Clinical Electronic Medical Records (CEMR) is a fundamental step in extracting disease knowledge by identifying specific entity terms such as diseases, symptoms, etc. However, the state-of-the-art NER methods based Long Short-Term Memory (LSTM) fail to exploit GPU parallelism fully under massive medical records. Although novel method Iterated Dilated CNNs (ID-CNNs) can accelerate network computing, it tends ignore word-order feature and semantic information...

Relation Extraction (RE) is a vital step to complete Knowledge Graph (KG) by extracting entity relations from texts.However, it usually suffers the long-tail issue. The training data mainly concentrates on few types of relations, leading lackof sufficient annotations for remaining relations. In this paper, we propose general approach learn relation prototypesfrom unlabeled texts, facilitate extraction transferring knowledge with trainingdata. We prototypes as an implicit factor between...

Automatic text generation techniques with either extractive-based or generative-based methods are becoming increasingly popular and widely used in industry. In contrast to existing approaches that ignore the coherence, our proposed approach integrates both keyword coverage coherence into optimization framework. this paper, we employ semantics-based syntax-based metrics evaluate coherence. Extensive experiments on a real corpus demonstrate method outperforms baselines overall regrading ROUGE...

A new method of building intelligent sentiment lexicon based on LDA and word clustering is put forward in this paper. In order to make seed words more representative universal, uses topic model build the term vectors select words. The improved SO-PMI algorithm has been used calculate emotional tendency each word. addition, domain lexicon's automatic extension update designed deal with dynamic corpus data. Experiments show that proposed can higher accuracy, reflect change words' real time. It...

We study the problem of multimodal fusion in this paper. Recent exchanging-based methods have been proposed for vision-vision fusion, which aim to exchange embeddings learned from one modality other. However, most them project inputs multimodalities into different low-dimensional spaces and cannot be applied sequential input data. To solve these issues, paper, we propose a novel model MuSE text-vision based on Transformer. first use two encoders separately map spaces. Then employ decoders...

We present the Chinese Dysarthria Speech Database (CDSD) as a valuable resource for dysarthria research. This database comprises speech data from 24 participants with dysarthria. Among these participants, one recorded an additional 10 hours of data, while each hour, resulting in 34 material. To accommodate varying cognitive levels, our text pool primarily consists content AISHELL-1 dataset and speeches by primary secondary school students. When read texts, they must use mobile device or ZOOM...

This paper presents a topic-enhanced recurrent autoencoder model to improve the accuracy of sentiment classification short texts. First, concept is proposed tackle problems in recursive including 'increasing computation complexity' and 'ignoring natural word order'. Then, enhanced with topic information generated by joint sentiment-topic (JST) model. Besides, order identify negations ironies texts, lexicon utilised add feature dimensions for sentence representations. Experiments are...