Precipitation events,which follow a life cycle of initiation,development,and decay,represent the fundamental form of precipitation.Comprehensive and accurate detection of these events is crucial for effective water re...Precipitation events,which follow a life cycle of initiation,development,and decay,represent the fundamental form of precipitation.Comprehensive and accurate detection of these events is crucial for effective water resource management and flood control.However,current investigations on their spatio-temporal patterns remain limited,largely because of the lack of systematic detection indices that are specifically designed for precipitation events,which constrains event-scale research.In this study,we defined a set of precipitation event detection indices(PEDI)that consists of five conventional and fourteen extreme indices to characterize precipitation events from the perspectives of intensity,duration,and frequency.Applications of the PEDI revealed the spatial patterns of hourly precipitation events in China and its first-and second-order river basins from 2008 to 2017.Both conventional and extreme precipitation events displayed spatial distribution patterns that gradually decreased in intensity,duration,and frequency from southeast to northwest China.Compared with those in northwest China,the average values of most PEDIs in southeast China were usually 2-10 times greater for first-order river basins and 3-15 times greater for second-order basins.The PEDI could serve as a reference method for investigating precipitation events at global,regional,and basin scales.展开更多
Microseismic monitoring is essential for understanding subsurface dynamics and optimizing oil and gas pro-duction.However,traditional methods for the automatic detection of microseismic events rely heavily on characte...Microseismic monitoring is essential for understanding subsurface dynamics and optimizing oil and gas pro-duction.However,traditional methods for the automatic detection of microseismic events rely heavily on characteristic functions and human intervention,often resulting in suboptimal performance when dealing with complex and noisy data.In this study,we propose a novel approach that leverages deep learning frame to extract multiscale features from microseismic data using a TransUNet neural network.Our model integrates the ad-vantages of Transformer and UNet architectures to achieve high accuracy in multivariate image segmentation and precise picking of P-wave and S-wave first arrivals simultaneously.We validate our approach using both synthetic and field microseismic datasets recorded from gas storage monitoring and roof fracturing in a coal seam.The robustness of the proposed method has been verified in the testing of synthetic data with various levels of Gaussian and real background noises extracted from field data.The comparisons of the proposed method with UNet and SwinUNet in terms of the model architecture and classification performance demonstrate the Tran-sUNet achieves the optimal balance in its architecture and inference speed.With relatively low inference time and network complexity,it operates effectively in high-precision microseismic phase pickings.This advancement holds significant promise for enhancing microseismic monitoring technology in hydraulic fracturing and reser-voir monitoring applications.展开更多
Current Chinese event detection methods commonly use word embedding to capture semantic representation,but these methods find it difficult to capture the dependence relationship between the trigger words and other wor...Current Chinese event detection methods commonly use word embedding to capture semantic representation,but these methods find it difficult to capture the dependence relationship between the trigger words and other words in the same sentence.Based on the simple evaluation,it is known that a dependency parser can effectively capture dependency relationships and improve the accuracy of event categorisation.This study proposes a novel architecture that models a hybrid representation to summarise semantic and structural information from both characters and words.This model can capture rich semantic features for the event detection task by incorporating the semantic representation generated from the dependency parser.The authors evaluate different models on kbp 2017 corpus.The experimental results show that the proposed method can significantly improve performance in Chinese event detection.展开更多
Online social media networks are gaining attention worldwide,with an increasing number of people relying on them to connect,communicate and share their daily pertinent event-related information.Event detection is now ...Online social media networks are gaining attention worldwide,with an increasing number of people relying on them to connect,communicate and share their daily pertinent event-related information.Event detection is now increasingly leveraging online social networks for highlighting events happening around the world via the Internet of People.In this paper,a novel Event Detection model based on Scoring and Word Embedding(ED-SWE)is proposed for discovering key events from a large volume of data streams of tweets and for generating an event summary using keywords and top-k tweets.The proposed ED-SWE model can distill high-quality tweets,reduce the negative impact of the advent of spam,and identify latent events in the data streams automatically.Moreover,a word embedding algorithm is used to learn a real-valued vector representation for a predefined fixed-sized vocabulary from a corpus of Twitter data.In order to further improve the performance of the Expectation-Maximization(EM)iteration algorithm,a novel initialization method based on the authority values of the tweets is also proposed in this paper to detect live events efficiently and precisely.Finally,a novel automatic identification method based on the cosine measure is used to automatically evaluate whether a given topic can form a live event.Experiments conducted on a real-world dataset demonstrate that the ED-SWE model exhibits better efficiency and accuracy than several state-of-art event detection models.展开更多
Event detection(ED)is aimed at detecting event occurrences and categorizing them.This task has been previously solved via recognition and classification of event triggers(ETs),which are defined as the phrase or word m...Event detection(ED)is aimed at detecting event occurrences and categorizing them.This task has been previously solved via recognition and classification of event triggers(ETs),which are defined as the phrase or word most clearly expressing event occurrence.Thus,current approaches require both annotated triggers as well as event types in training data.Nevertheless,triggers are non-essential in ED,and it is time-wasting for annotators to identify the“most clearly”word from a sentence,particularly in longer sentences.To decrease manual effort,we evaluate event detectionwithout triggers.We propose a novel framework that combines Type-aware Attention and Graph Convolutional Networks(TA-GCN)for event detection.Specifically,the task is identified as a multi-label classification problem.We first encode the input sentence using a novel type-aware neural network with attention mechanisms.Then,a Graph Convolutional Networks(GCN)-based multilabel classification model is exploited for event detection.Experimental results demonstrate the effectiveness.展开更多
Recently, the Internet of Things （loT） has attracted more and more attention. Multimedia sensor network plays an important role in the IoT, and audio event detection in the multimedia sensor net- works is one of the...Recently, the Internet of Things （loT） has attracted more and more attention. Multimedia sensor network plays an important role in the IoT, and audio event detection in the multimedia sensor net- works is one of the most important applications for the Internet of Things. In practice, it is hard to get enough real-world samples to generate the classifi- ers for some special audio events （e.g., car-crash- ing in the smart traffic system）. In this paper, we introduce a TrAdaBoost-based method to solve the above problem. By using the proposed approach, we can train a strong classifier by using only a tiny amount of real-world data and a large number of more easily collected samples （e.g., collected from TV programs）, even when the real-world data is not sufficient to train a model alone. We deploy this ap- proach in a smart traffic system to evaluate its per- formance, and the experiment evaluations demonstrate that our method can achieve satisfying results.展开更多
How to quickly and accurately detect new topics from massive data online becomes a main problem of public opinion monitoring in cyberspace. This paperpresents a new event detection method for the current new event det...How to quickly and accurately detect new topics from massive data online becomes a main problem of public opinion monitoring in cyberspace. This paperpresents a new event detection method for the current new event detection system, based on sorted subtopic matching algorithm and constructs the entire design framework. In this p^per, the subtopics contained in old topics （or news stories） are sorted in descending order according to their importance to the topic（or news stories）, and form a sorted subtopic sequence. In the process of subtopic matching, subtopic scoring matrix is used to determine whether a new story is reporting a new event. Experimental results show that the sorted subtopic matching model improved the accuracy and effectiveness ofthenew event detection system in cyberspace.展开更多
In this paper,we consider unusual event detection problem in a novel viewpoint and provide an algorithm to solve the problem.The actions or events in the scene is usual or not will eventually be reflected on the chang...In this paper,we consider unusual event detection problem in a novel viewpoint and provide an algorithm to solve the problem.The actions or events in the scene is usual or not will eventually be reflected on the changes of some basic features.We summarize these basic event features and propose special representation for each of them.Thus we can model these features in a uniform mode using adaptive Gaussian mixture model.Supervised and unsupervised unusual event detection algorithm can be designed to fit various situations based on this model.The superiority of our model is that it can detect unusual event automatically without to know the determinate model of unusual events.In conclusion,we provide two applications to verify the effectiveness of our model.展开更多
Social media like Twitter who serves as a novel news medium and has become increasingly popular since its establishment. Large scale first-hand user-generated tweets motivate automatic event detection on Twitter. Prev...Social media like Twitter who serves as a novel news medium and has become increasingly popular since its establishment. Large scale first-hand user-generated tweets motivate automatic event detection on Twitter. Previous unsupervised approaches detected events by clustering words. These methods detect events using burstiness,which measures surging frequencies of words at certain time windows. However,event clusters represented by a set of individual words are difficult to understand. This issue is addressed by building a document-level event detection model that directly calculates the burstiness of tweets,leveraging distributed word representations for modeling semantic information,thereby avoiding sparsity. Results show that the document-level model not only offers event summaries that are directly human-readable,but also gives significantly improved accuracies compared to previous methods on unsupervised tweet event detection,which are based on words/segments.展开更多
Supervised models for event detection usually require large-scale human-annotated training data,especially neural models.A data augmentation technique is proposed to improve the performance of event detection by gener...Supervised models for event detection usually require large-scale human-annotated training data,especially neural models.A data augmentation technique is proposed to improve the performance of event detection by generating paraphrase sentences to enrich expressions of the original data.Specifically,based on an existing human-annotated event detection dataset,we first automatically build a paraphrase dataset and label it with a designed event annotation alignment algorithm.To alleviate possible wrong labels in the generated paraphrase dataset,a multi-instance learning(MIL)method is adopted for joint training on both the gold human-annotated data and the generated paraphrase dataset.Experimental results on a widely used dataset ACE2005 show the effectiveness of our approach.展开更多
1 Introduction Sound event detection(SED)aims to identify and locate specific sound event categories and their corresponding timestamps within continuous audio streams.To overcome the limitations posed by the scarcity...1 Introduction Sound event detection(SED)aims to identify and locate specific sound event categories and their corresponding timestamps within continuous audio streams.To overcome the limitations posed by the scarcity of strongly labeled training data,researchers have increasingly turned to semi-supervised learning(SSL)[1],which leverages unlabeled data to augment training and improve detection performance.Among many SSL methods[2-4].展开更多
Semi-supervised sound event detection(SSED)tasks typically leverage a large amount of unlabeled and synthetic data to facilitate model generalization during training,reducing overfitting on a limited set of labeled da...Semi-supervised sound event detection(SSED)tasks typically leverage a large amount of unlabeled and synthetic data to facilitate model generalization during training,reducing overfitting on a limited set of labeled data.However,the generalization training process often encounters challenges from noisy interference introduced by pseudo-labels or domain knowledge gaps.To alleviate noisy interference in class distribution learning,we propose an efficient semi-supervised class distribution learning method through dynamic prompt tuning,named prompting class distribution optimization(PADO).Specifically,when modeling real labeled data,PADO dynamically incorporates independent learnable prompt tokens to explore prior knowledge about the true distribution.Then,the prior knowledge serves as prompt information,dynamically interacting with the posterior noisy-class distribution information.In this case,PADO achieves class distribution optimization while maintaining model generalization,leading to a significant improvement in the efficiency of class distribution learning.Compared with state-of-the-art methods on the SSED datasets from DCASE 2019,2020,and 2021 challenges,PADO achieves significant performance improvements.Furthermore,it is readily extendable to other benchmark models.展开更多
The aim of non-intrusive appliance load monitoring （NIALM） is to disaggregate the energy consumption of individual electrical appliances from total power consumption utilizing non-intrusive methods. In this paper, a...The aim of non-intrusive appliance load monitoring （NIALM） is to disaggregate the energy consumption of individual electrical appliances from total power consumption utilizing non-intrusive methods. In this paper, a systematic approach to 0N-0FF event detection and clustering analysis for NIALM were presented. From the aggregate power consumption data set, the data are passed through median filtering to reduce noise and prepared for the event detection algorithm. The event detection algorithm is to determine the switching of ON and OFF status of electrical appliances. The goodness- of-fit （GOF） methodology is the event detection algorithm implemented. After event detection, the events detected were paired into ON-0FF pairing appliances. The results from the ON-OFF pairing algorithm were further clustered in groups utilizing the K-means clustering analysis. The K- means clustering were implemented as an unsupervised learning methodology for the clustering analysis. The novelty of this paper is the determination of the time duration an electrical appliance is turned ON through combination of event detection, ON-OFF pairing and K- means clustering. The results of the algorithm implemen- tation were discussed and ideas on future work were also proposed.展开更多
Real-life events are emerging and evolving in social and news streams.Recent methods have succeeded in capturing designed features of monolingual events,but lack of interpretability and multi-lingual considerations.To...Real-life events are emerging and evolving in social and news streams.Recent methods have succeeded in capturing designed features of monolingual events,but lack of interpretability and multi-lingual considerations.To this end,we propose a multi-lingual event mining model,namely MLEM,to automatically detect events and generate evolution graph in multilingual hybrid-length text streams including English,Chinese,French,German,Russian and Japanese.Specially,we merge the same entities and similar phrases and present multiple similarity measures by incremental word2vec model.We propose an 8-tuple to describe event for correlation analysis and evolution graph generation.We evaluate the MLEM model using a massive human-generated dataset containing real world events.Experimental results show that our new model MLEM outperforms the baseline method both in efficiency and effectiveness.展开更多
Taking the advantage of Internet of Things(IoT)enabled measurements,this paper formulates the event detection problem as an information-plus-noise model,and detects events in power systems based on free probability th...Taking the advantage of Internet of Things(IoT)enabled measurements,this paper formulates the event detection problem as an information-plus-noise model,and detects events in power systems based on free probability theory(FPT).Using big data collected from phasor measurement units(PMUs),we construct the event detection matrix to reflect both spatial and temporal characteristics of power gird states.The event detection matrix is further described as an information matrix plus a noise matrix,and the essence of event detection is to extract event information from the event detection matrix.By associating the event detection problem with FPT,the empirical spectral distributions(ESDs)related moments of the sample covariance matrix of the information matrix are computed,to distinguish events from“noises”,including normal fluctuations,background noises,and measurement errors.Based on central limit theory(CLT),the alarm threshold is computed using measurements collected in normal states.Additionally,with the aid of sliding window,this paper builds an event detection architecture to reflect power grid state and detect events online.Case studies with simulated data from Anhui,China,and real PMU data from Guangdong,China,verify the effectiveness of the proposed method.Compared with other data-driven methods,the proposed method is more sensitive and has better adaptability to the normal fluctuations,background noises,and measurement errors in real PMU cases.In addition,it does not require large number of training samples as needed in the training-testing paradigm.展开更多
For accuracy and rapidity of audio event detection in the mass-data audio pro- cessing tasks, a generic method of rapidly recognizing audio event based on 2D-Haar acoustic super feature vector and AdaBoost is proposed...For accuracy and rapidity of audio event detection in the mass-data audio pro- cessing tasks, a generic method of rapidly recognizing audio event based on 2D-Haar acoustic super feature vector and AdaBoost is proposed. Firstly, it combines certain number of con- tinuous audio frames to be an ＂acoustic feature image＂, secondly, uses AdaBoost.MH or fast Random AdaBoost feature selection algorithm to select high representative 2D-Haar pattern combinations to construct super feature vectors; thirdly, analyzes the commonality and differ- ences between subcategories, then extracts common features and reduces different features to obtain a generic audio event template, which can support the accurate identification of multi- ple sub-classes and detect and locate the specific audio event from the audio stream accurately. Experimental results show that the use of 2D-Haar acoustic feature super vector can make recog- nition accuracy 5% higher than ones that MFCC, PLP, LPCC and other traditional acoustic features yielded, and can make tile training processing 7 20 times faster and the recognition processing 5-10 times faster, it can even achieve an average precision of 93.38%, an average recall of 95.03% under the optimal parameter configuration found by grid method. Above all, it can provide an accurate and fast mass-data processing method for audio event detection.展开更多
Non-intrusive load monitoring(NILM) can provide appliance-level power consumption information without deploying submeters for each load, in which load event detection is one of the crucial steps. However, the existing...Non-intrusive load monitoring(NILM) can provide appliance-level power consumption information without deploying submeters for each load, in which load event detection is one of the crucial steps. However, the existing event detection methods do not efficiently detect both the starting time of an event(STE) and the ending time of an event(ETE), and their adaptability to scenarios with different sampling rates is limited. To address these problems, in this paper, an event detection method based on robust random cut forest(RRCF) algorithm, which is an unsupervised learning method for detecting anomalous data points within a dataset, is proposed. First, the meanpooling preprocessing is applied to the aggregated load power series with a high sampling rate to minimize fluctuations. Then, the power differential series is obtained, and the anomaly score of each data point is calculated using the RRCF algorithm for preliminary detection. If an event has been preliminarily detected, misidentification caused by fluctuation will be further eliminated by using an adaptive power difference threshold approach. Finally, linear fitting is used to finely and accurately adjust the STE and ETE. The proposed method does not require any pretraining of the detection model and has been validated with both the BLUED dataset(with high and low sampling rates) and the REDD dataset(with low sampling rate). The experimental results demonstrate that the proposed method not only meets real-time requirements, but also exhibits strong adaptability across multiple scenarios. The precision is greater than 92% in distinct sampling rate scenarios, and the F1 score of phase B on the BLUED dataset reaches 94% in the scenario with a high sampling rate. These results indicate that the proposed method outperforms other state-of-the-art methods.展开更多
Event detection(ED)seeks to recognize event triggers and classify them into the predefined event types.Chinese ED is formulated as a character-level task owing to the uncertain word boundaries.Prior methods try to inc...Event detection(ED)seeks to recognize event triggers and classify them into the predefined event types.Chinese ED is formulated as a character-level task owing to the uncertain word boundaries.Prior methods try to incorpo-rate word-level information into characters to enhance their semantics.However,they experience two problems.First,they fail to incorporate word-level information into each character the word encompasses,causing the insufficient word-charac-ter interaction problem.Second,they struggle to distinguish events of similar types with limited annotated instances,which is called the event confusing problem.This paper proposes a novel model named Label-Aware Heterogeneous Graph Attention Network(L-HGAT)to address these two problems.Specifically,we first build a heterogeneous graph of two node types and three edge types to maximally preserve word-character interactions,and then deploy a heterogeneous graph attention network to enhance the semantic propagation between characters and words.Furthermore,we design a pushing-away game to enlarge the predicting gap between the ground-truth event type and its confusing counterpart for each character.Experimental results show that our L-HGAT model consistently achieves superior performance over prior competitive methods.展开更多
This paper proposes five indicators to evaluate the effectiveness and viability for adverse glycemic events detection based on predicted blood glucose(BG)values.False negative rate(FNR)and false positive rate(FPR)are ...This paper proposes five indicators to evaluate the effectiveness and viability for adverse glycemic events detection based on predicted blood glucose(BG)values.False negative rate(FNR)and false positive rate(FPR)are defined to evaluate whether it can detect adverse glycemic events(AGEs)based on the predicted value.The temporal overlap(TO)and time difference(TD)are proposed to evaluate whether the predicted model can capture the accurate time duration of AGEs.The sum of squared percent(SSP)measures comprehensive similarity between prediction values and true values.We examined 328 patients with type 2 diabetes,containing real continuous glucose monitoring data with 5-minute time intervals.Autoregressive integrated moving average model has lower FNR and FPR.The gated recurrent unit has better temporal behavior where the mean TO with standard deviation is calculated as 0.84±0.18,and the mean TD with standard deviation is(4.39±4.01)min.Neural models have better effects on SSP(for hypoglycemia,long-short tern memory possesses 0.149 and 0.246).These five indicators are able to evaluate whether we can detect abnormal BG levels and reveal the temporal behavior of AGEs effectively.The proposed neural predictive models have more promising application in AGE detection.展开更多
The occurrence of microseismic is not random but is related to the physical properties of the underground medium.Due to the low intensity and the influence of noise,microseismic eventually lead to poor signal-to-noise...The occurrence of microseismic is not random but is related to the physical properties of the underground medium.Due to the low intensity and the influence of noise,microseismic eventually lead to poor signal-to-noise ratio.We proposed a method for automatic detection of microseismic events by adoption of multiscale top-hat transformation.The method is based on the difference between the signal and noise in the multiscale top-hat transform section and achieves the detection on a specific section.The microseismic data are decomposed into different scales by multiscale morphology top-hat transformation firstly.Then the potential microseismic events could be detected by picking up the peak value in the multiscale top-hat section,and the characteristic profile obtains the start point with a specific threshold value.Finally,the synthetic data experiences demonstrate the advantages of this method under strong and weak noisy conditions,and the filed data example also shows its reliability and adaptability.展开更多
基金National Key Research and Development Program of China,No.2023YFC3206605,No.2021YFC3201102National Natural Science Foundation of China,No.41971035。
文摘Precipitation events,which follow a life cycle of initiation,development,and decay,represent the fundamental form of precipitation.Comprehensive and accurate detection of these events is crucial for effective water resource management and flood control.However,current investigations on their spatio-temporal patterns remain limited,largely because of the lack of systematic detection indices that are specifically designed for precipitation events,which constrains event-scale research.In this study,we defined a set of precipitation event detection indices(PEDI)that consists of five conventional and fourteen extreme indices to characterize precipitation events from the perspectives of intensity,duration,and frequency.Applications of the PEDI revealed the spatial patterns of hourly precipitation events in China and its first-and second-order river basins from 2008 to 2017.Both conventional and extreme precipitation events displayed spatial distribution patterns that gradually decreased in intensity,duration,and frequency from southeast to northwest China.Compared with those in northwest China,the average values of most PEDIs in southeast China were usually 2-10 times greater for first-order river basins and 3-15 times greater for second-order basins.The PEDI could serve as a reference method for investigating precipitation events at global,regional,and basin scales.
基金supported by a National Natural Science Foundation of China(Grant number 41974150 and 42174158)Natural Science Basic Research Program of Shaanxi(2023-JC-YB-220).
文摘Microseismic monitoring is essential for understanding subsurface dynamics and optimizing oil and gas pro-duction.However,traditional methods for the automatic detection of microseismic events rely heavily on characteristic functions and human intervention,often resulting in suboptimal performance when dealing with complex and noisy data.In this study,we propose a novel approach that leverages deep learning frame to extract multiscale features from microseismic data using a TransUNet neural network.Our model integrates the ad-vantages of Transformer and UNet architectures to achieve high accuracy in multivariate image segmentation and precise picking of P-wave and S-wave first arrivals simultaneously.We validate our approach using both synthetic and field microseismic datasets recorded from gas storage monitoring and roof fracturing in a coal seam.The robustness of the proposed method has been verified in the testing of synthetic data with various levels of Gaussian and real background noises extracted from field data.The comparisons of the proposed method with UNet and SwinUNet in terms of the model architecture and classification performance demonstrate the Tran-sUNet achieves the optimal balance in its architecture and inference speed.With relatively low inference time and network complexity,it operates effectively in high-precision microseismic phase pickings.This advancement holds significant promise for enhancing microseismic monitoring technology in hydraulic fracturing and reser-voir monitoring applications.
基金973 Program,Grant/Award Number:2014CB340504The State Key Program of National Natural Science of China,Grant/Award Number:61533018+3 种基金National Natural Science Foundation of China,Grant/Award Number:61402220The Philosophy and Social Science Foundation of Hunan Province,Grant/Award Number:16YBA323Natural Science Foundation of Hunan Province,Grant/Award Number:2020JJ4525Scientific Research Fund of Hunan Provincial Education Department,Grant/Award Number:18B279,19A439。
文摘Current Chinese event detection methods commonly use word embedding to capture semantic representation,but these methods find it difficult to capture the dependence relationship between the trigger words and other words in the same sentence.Based on the simple evaluation,it is known that a dependency parser can effectively capture dependency relationships and improve the accuracy of event categorisation.This study proposes a novel architecture that models a hybrid representation to summarise semantic and structural information from both characters and words.This model can capture rich semantic features for the event detection task by incorporating the semantic representation generated from the dependency parser.The authors evaluate different models on kbp 2017 corpus.The experimental results show that the proposed method can significantly improve performance in Chinese event detection.
基金The work reported in this paper has been supported by UK-Jiangsu 20-20 World Class University Initiative programme.
文摘Online social media networks are gaining attention worldwide,with an increasing number of people relying on them to connect,communicate and share their daily pertinent event-related information.Event detection is now increasingly leveraging online social networks for highlighting events happening around the world via the Internet of People.In this paper,a novel Event Detection model based on Scoring and Word Embedding(ED-SWE)is proposed for discovering key events from a large volume of data streams of tweets and for generating an event summary using keywords and top-k tweets.The proposed ED-SWE model can distill high-quality tweets,reduce the negative impact of the advent of spam,and identify latent events in the data streams automatically.Moreover,a word embedding algorithm is used to learn a real-valued vector representation for a predefined fixed-sized vocabulary from a corpus of Twitter data.In order to further improve the performance of the Expectation-Maximization(EM)iteration algorithm,a novel initialization method based on the authority values of the tweets is also proposed in this paper to detect live events efficiently and precisely.Finally,a novel automatic identification method based on the cosine measure is used to automatically evaluate whether a given topic can form a live event.Experiments conducted on a real-world dataset demonstrate that the ED-SWE model exhibits better efficiency and accuracy than several state-of-art event detection models.
基金supported by the Hunan Provincial Natural Science Foundation of China(Grant No.2020JJ4624)the National Social Science Fund of China(Grant No.20&ZD047)+1 种基金the Scientific Research Fund of Hunan Provincial Education Department(Grant No.19A020)the National University of Defense Technology Research Project ZK20-46 and the Young Elite Scientists Sponsorship Program 2021-JCJQ-QT-050.
文摘Event detection(ED)is aimed at detecting event occurrences and categorizing them.This task has been previously solved via recognition and classification of event triggers(ETs),which are defined as the phrase or word most clearly expressing event occurrence.Thus,current approaches require both annotated triggers as well as event types in training data.Nevertheless,triggers are non-essential in ED,and it is time-wasting for annotators to identify the“most clearly”word from a sentence,particularly in longer sentences.To decrease manual effort,we evaluate event detectionwithout triggers.We propose a novel framework that combines Type-aware Attention and Graph Convolutional Networks(TA-GCN)for event detection.Specifically,the task is identified as a multi-label classification problem.We first encode the input sentence using a novel type-aware neural network with attention mechanisms.Then,a Graph Convolutional Networks(GCN)-based multilabel classification model is exploited for event detection.Experimental results demonstrate the effectiveness.
基金supported by the National Natural Science Foundation of China(No.60821001)the National Basic Research Program of China(No.2007CB311203)
文摘Recently, the Internet of Things （loT） has attracted more and more attention. Multimedia sensor network plays an important role in the IoT, and audio event detection in the multimedia sensor net- works is one of the most important applications for the Internet of Things. In practice, it is hard to get enough real-world samples to generate the classifi- ers for some special audio events （e.g., car-crash- ing in the smart traffic system）. In this paper, we introduce a TrAdaBoost-based method to solve the above problem. By using the proposed approach, we can train a strong classifier by using only a tiny amount of real-world data and a large number of more easily collected samples （e.g., collected from TV programs）, even when the real-world data is not sufficient to train a model alone. We deploy this ap- proach in a smart traffic system to evaluate its per- formance, and the experiment evaluations demonstrate that our method can achieve satisfying results.
基金Funded by the Planning Project of National Language Committee in the "12th 5-year Plan"(No.YB125-49)the Foundation for Key Program of Ministry of Education,China(No.212167)the Fundamental Research Funds for the Central Universities(No.SWJTU12CX096)
文摘How to quickly and accurately detect new topics from massive data online becomes a main problem of public opinion monitoring in cyberspace. This paperpresents a new event detection method for the current new event detection system, based on sorted subtopic matching algorithm and constructs the entire design framework. In this p^per, the subtopics contained in old topics （or news stories） are sorted in descending order according to their importance to the topic（or news stories）, and form a sorted subtopic sequence. In the process of subtopic matching, subtopic scoring matrix is used to determine whether a new story is reporting a new event. Experimental results show that the sorted subtopic matching model improved the accuracy and effectiveness ofthenew event detection system in cyberspace.
基金the National Natural Science Foundation of China (No. 60805001)
文摘In this paper,we consider unusual event detection problem in a novel viewpoint and provide an algorithm to solve the problem.The actions or events in the scene is usual or not will eventually be reflected on the changes of some basic features.We summarize these basic event features and propose special representation for each of them.Thus we can model these features in a uniform mode using adaptive Gaussian mixture model.Supervised and unsupervised unusual event detection algorithm can be designed to fit various situations based on this model.The superiority of our model is that it can detect unusual event automatically without to know the determinate model of unusual events.In conclusion,we provide two applications to verify the effectiveness of our model.
基金Supported by the National High Technology Research and Development Programme of China(No.2015AA015405)
文摘Social media like Twitter who serves as a novel news medium and has become increasingly popular since its establishment. Large scale first-hand user-generated tweets motivate automatic event detection on Twitter. Previous unsupervised approaches detected events by clustering words. These methods detect events using burstiness,which measures surging frequencies of words at certain time windows. However,event clusters represented by a set of individual words are difficult to understand. This issue is addressed by building a document-level event detection model that directly calculates the burstiness of tweets,leveraging distributed word representations for modeling semantic information,thereby avoiding sparsity. Results show that the document-level model not only offers event summaries that are directly human-readable,but also gives significantly improved accuracies compared to previous methods on unsupervised tweet event detection,which are based on words/segments.
基金National Natural Science Foundation of China(No.62006039)。
文摘Supervised models for event detection usually require large-scale human-annotated training data,especially neural models.A data augmentation technique is proposed to improve the performance of event detection by generating paraphrase sentences to enrich expressions of the original data.Specifically,based on an existing human-annotated event detection dataset,we first automatically build a paraphrase dataset and label it with a designed event annotation alignment algorithm.To alleviate possible wrong labels in the generated paraphrase dataset,a multi-instance learning(MIL)method is adopted for joint training on both the gold human-annotated data and the generated paraphrase dataset.Experimental results on a widely used dataset ACE2005 show the effectiveness of our approach.
基金supported by the Zhejiang Provincial Key R&D Program(Nos.2024C01108,2023C01030,2023C01034)the Hangzhou Key R&D Program(Nos.2023SZD0046,2024SZD1A03)the Ningbo Key R&D Program(No.2024Z114).
文摘1 Introduction Sound event detection(SED)aims to identify and locate specific sound event categories and their corresponding timestamps within continuous audio streams.To overcome the limitations posed by the scarcity of strongly labeled training data,researchers have increasingly turned to semi-supervised learning(SSL)[1],which leverages unlabeled data to augment training and improve detection performance.Among many SSL methods[2-4].
基金supported by the National Natural Science Foundation of China(Nos.62176106 and U1836220)the Special Scientific Research Project of School of Emergency Management of Jiangsu University(No.KY-A-01)+2 种基金the Project of Faculty of Agricultural Engineering of Jiangsu University(No.NGXB20240101)the Post-graduate Research&Practice Innovation Program of Jiangsu Province(Nos.KYCX22_3668 and KYCX21_3373)the Jiangsu Key Research and Development Plan(No.BE2020036)。
文摘Semi-supervised sound event detection(SSED)tasks typically leverage a large amount of unlabeled and synthetic data to facilitate model generalization during training,reducing overfitting on a limited set of labeled data.However,the generalization training process often encounters challenges from noisy interference introduced by pseudo-labels or domain knowledge gaps.To alleviate noisy interference in class distribution learning,we propose an efficient semi-supervised class distribution learning method through dynamic prompt tuning,named prompting class distribution optimization(PADO).Specifically,when modeling real labeled data,PADO dynamically incorporates independent learnable prompt tokens to explore prior knowledge about the true distribution.Then,the prior knowledge serves as prompt information,dynamically interacting with the posterior noisy-class distribution information.In this case,PADO achieves class distribution optimization while maintaining model generalization,leading to a significant improvement in the efficiency of class distribution learning.Compared with state-of-the-art methods on the SSED datasets from DCASE 2019,2020,and 2021 challenges,PADO achieves significant performance improvements.Furthermore,it is readily extendable to other benchmark models.
文摘The aim of non-intrusive appliance load monitoring （NIALM） is to disaggregate the energy consumption of individual electrical appliances from total power consumption utilizing non-intrusive methods. In this paper, a systematic approach to 0N-0FF event detection and clustering analysis for NIALM were presented. From the aggregate power consumption data set, the data are passed through median filtering to reduce noise and prepared for the event detection algorithm. The event detection algorithm is to determine the switching of ON and OFF status of electrical appliances. The goodness- of-fit （GOF） methodology is the event detection algorithm implemented. After event detection, the events detected were paired into ON-0FF pairing appliances. The results from the ON-OFF pairing algorithm were further clustered in groups utilizing the K-means clustering analysis. The K- means clustering were implemented as an unsupervised learning methodology for the clustering analysis. The novelty of this paper is the determination of the time duration an electrical appliance is turned ON through combination of event detection, ON-OFF pairing and K- means clustering. The results of the algorithm implemen- tation were discussed and ideas on future work were also proposed.
基金This work was supported by NSFC program(Grant Nos.61872022,61421003,U1636123)SKLSDE-2018ZX-16 and partly by the Beijing Advanced Innovation Center for Big Data and Brain Computing.
文摘Real-life events are emerging and evolving in social and news streams.Recent methods have succeeded in capturing designed features of monolingual events,but lack of interpretability and multi-lingual considerations.To this end,we propose a multi-lingual event mining model,namely MLEM,to automatically detect events and generate evolution graph in multilingual hybrid-length text streams including English,Chinese,French,German,Russian and Japanese.Specially,we merge the same entities and similar phrases and present multiple similarity measures by incremental word2vec model.We propose an 8-tuple to describe event for correlation analysis and evolution graph generation.We evaluate the MLEM model using a massive human-generated dataset containing real world events.Experimental results show that our new model MLEM outperforms the baseline method both in efficiency and effectiveness.
基金supported by the National Key Research and Development Program of China(No.2021YFB2401302)。
文摘Taking the advantage of Internet of Things(IoT)enabled measurements,this paper formulates the event detection problem as an information-plus-noise model,and detects events in power systems based on free probability theory(FPT).Using big data collected from phasor measurement units(PMUs),we construct the event detection matrix to reflect both spatial and temporal characteristics of power gird states.The event detection matrix is further described as an information matrix plus a noise matrix,and the essence of event detection is to extract event information from the event detection matrix.By associating the event detection problem with FPT,the empirical spectral distributions(ESDs)related moments of the sample covariance matrix of the information matrix are computed,to distinguish events from“noises”,including normal fluctuations,background noises,and measurement errors.Based on central limit theory(CLT),the alarm threshold is computed using measurements collected in normal states.Additionally,with the aid of sliding window,this paper builds an event detection architecture to reflect power grid state and detect events online.Case studies with simulated data from Anhui,China,and real PMU data from Guangdong,China,verify the effectiveness of the proposed method.Compared with other data-driven methods,the proposed method is more sensitive and has better adaptability to the normal fluctuations,background noises,and measurement errors in real PMU cases.In addition,it does not require large number of training samples as needed in the training-testing paradigm.
文摘For accuracy and rapidity of audio event detection in the mass-data audio pro- cessing tasks, a generic method of rapidly recognizing audio event based on 2D-Haar acoustic super feature vector and AdaBoost is proposed. Firstly, it combines certain number of con- tinuous audio frames to be an ＂acoustic feature image＂, secondly, uses AdaBoost.MH or fast Random AdaBoost feature selection algorithm to select high representative 2D-Haar pattern combinations to construct super feature vectors; thirdly, analyzes the commonality and differ- ences between subcategories, then extracts common features and reduces different features to obtain a generic audio event template, which can support the accurate identification of multi- ple sub-classes and detect and locate the specific audio event from the audio stream accurately. Experimental results show that the use of 2D-Haar acoustic feature super vector can make recog- nition accuracy 5% higher than ones that MFCC, PLP, LPCC and other traditional acoustic features yielded, and can make tile training processing 7 20 times faster and the recognition processing 5-10 times faster, it can even achieve an average precision of 93.38%, an average recall of 95.03% under the optimal parameter configuration found by grid method. Above all, it can provide an accurate and fast mass-data processing method for audio event detection.
基金This work was supported by the Zhejiang Provincial Natural Science Foundation of China(No.LGG22F030008)Key R&D Program of Zhejiang Province(No.2021C01113).
文摘Non-intrusive load monitoring(NILM) can provide appliance-level power consumption information without deploying submeters for each load, in which load event detection is one of the crucial steps. However, the existing event detection methods do not efficiently detect both the starting time of an event(STE) and the ending time of an event(ETE), and their adaptability to scenarios with different sampling rates is limited. To address these problems, in this paper, an event detection method based on robust random cut forest(RRCF) algorithm, which is an unsupervised learning method for detecting anomalous data points within a dataset, is proposed. First, the meanpooling preprocessing is applied to the aggregated load power series with a high sampling rate to minimize fluctuations. Then, the power differential series is obtained, and the anomaly score of each data point is calculated using the RRCF algorithm for preliminary detection. If an event has been preliminarily detected, misidentification caused by fluctuation will be further eliminated by using an adaptive power difference threshold approach. Finally, linear fitting is used to finely and accurately adjust the STE and ETE. The proposed method does not require any pretraining of the detection model and has been validated with both the BLUED dataset(with high and low sampling rates) and the REDD dataset(with low sampling rate). The experimental results demonstrate that the proposed method not only meets real-time requirements, but also exhibits strong adaptability across multiple scenarios. The precision is greater than 92% in distinct sampling rate scenarios, and the F1 score of phase B on the BLUED dataset reaches 94% in the scenario with a high sampling rate. These results indicate that the proposed method outperforms other state-of-the-art methods.
基金This work was supported by the National Key Research and Development Program of China under Grant No.2021YFB3100600the Youth Innovation Promotion Association of Chinese Academy of Sciences under Grant No.2021153the State Key Program of National Natural Science Foundation of China under Grant No.U2336202.
文摘Event detection(ED)seeks to recognize event triggers and classify them into the predefined event types.Chinese ED is formulated as a character-level task owing to the uncertain word boundaries.Prior methods try to incorpo-rate word-level information into characters to enhance their semantics.However,they experience two problems.First,they fail to incorporate word-level information into each character the word encompasses,causing the insufficient word-charac-ter interaction problem.Second,they struggle to distinguish events of similar types with limited annotated instances,which is called the event confusing problem.This paper proposes a novel model named Label-Aware Heterogeneous Graph Attention Network(L-HGAT)to address these two problems.Specifically,we first build a heterogeneous graph of two node types and three edge types to maximally preserve word-character interactions,and then deploy a heterogeneous graph attention network to enhance the semantic propagation between characters and words.Furthermore,we design a pushing-away game to enlarge the predicting gap between the ground-truth event type and its confusing counterpart for each character.Experimental results show that our L-HGAT model consistently achieves superior performance over prior competitive methods.
文摘This paper proposes five indicators to evaluate the effectiveness and viability for adverse glycemic events detection based on predicted blood glucose(BG)values.False negative rate(FNR)and false positive rate(FPR)are defined to evaluate whether it can detect adverse glycemic events(AGEs)based on the predicted value.The temporal overlap(TO)and time difference(TD)are proposed to evaluate whether the predicted model can capture the accurate time duration of AGEs.The sum of squared percent(SSP)measures comprehensive similarity between prediction values and true values.We examined 328 patients with type 2 diabetes,containing real continuous glucose monitoring data with 5-minute time intervals.Autoregressive integrated moving average model has lower FNR and FPR.The gated recurrent unit has better temporal behavior where the mean TO with standard deviation is calculated as 0.84±0.18,and the mean TD with standard deviation is(4.39±4.01)min.Neural models have better effects on SSP(for hypoglycemia,long-short tern memory possesses 0.149 and 0.246).These five indicators are able to evaluate whether we can detect abnormal BG levels and reveal the temporal behavior of AGEs effectively.The proposed neural predictive models have more promising application in AGE detection.
基金supported in part by the National Natural Science Foundation of China under Grant 41904098Fundamental Research Funds for the Central Universities,under Grant 2462018YJRC020 and Grant 2462020YXZZ006。
文摘The occurrence of microseismic is not random but is related to the physical properties of the underground medium.Due to the low intensity and the influence of noise,microseismic eventually lead to poor signal-to-noise ratio.We proposed a method for automatic detection of microseismic events by adoption of multiscale top-hat transformation.The method is based on the difference between the signal and noise in the multiscale top-hat transform section and achieves the detection on a specific section.The microseismic data are decomposed into different scales by multiscale morphology top-hat transformation firstly.Then the potential microseismic events could be detected by picking up the peak value in the multiscale top-hat section,and the characteristic profile obtains the start point with a specific threshold value.Finally,the synthetic data experiences demonstrate the advantages of this method under strong and weak noisy conditions,and the filed data example also shows its reliability and adaptability.
