Existing Event Detection Methods Research Articles

Reinforced, Incremental and Cross-Lingual Event Detection From Social Messages.

Detecting hot social events (e.g., political scandal, momentous meetings, natural hazards, etc.) from social messages is crucial as it highlights significant happenings to help people understand the real world. On account of the streaming nature of social messages, incremental social event detection models in acquiring, preserving, and updating messages over time have attracted great attention. However, the challenge is that the existing event detection methods towards streaming social messages are generally confronted with ambiguous events features, dispersive text contents, and multiple languages, and hence result in low accuracy and generalization ability. In this paper, we present a novel reinForced, incremental and cross-lingual social Event detection architecture, namely FinEvent, from streaming social messages. Concretely, we first model social messages into heterogeneous graphs integrating both rich meta-semantics and diverse meta-relations, and convert them to weighted multi-relational message graphs. Second, we propose a new reinforced weighted multi-relational graph neural network framework by using a Multi-agent Reinforcement Learning algorithm to select optimal aggregation thresholds across different relations/edges to learn social message embeddings. To solve the long-tail problem in social event detection, a balanced sampling strategy guided Contrastive Learning mechanism is designed for incremental social message representation learning. Third, a new Deep Reinforcement Learning guided density-based spatial clustering model is designed to select the optimal minimum number of samples required to form a cluster and optimal minimum distance between two clusters in social event detection tasks. Finally, we implement incremental social message representation learning based on knowledge preservation on the graph neural network and achieve the transferring cross-lingual social event detection. We conduct extensive experiments to evaluate the FinEvent on Twitter streams, demonstrating a significant and consistent improvement in model quality with 14%-118%, 8%-170%, and 2%-21% increases in performance on offline, online, and cross-lingual social event detection tasks.

Robust event detection for residential load disaggregation

Nonintrusive load monitoring (NILM) can facilate the transition to energy-efficient and low-carbon buildings. Event detection is the first and most critical step in event-based NILM and can improve the efficiency and performance of NILM by decreasing inference times to the number of events with transient features extracted from events. However, existing event detection methods with fixed parameters may fail to achieve high accuracy in the case of uncertain and complicated residential load changes such as high fluctuation, long transition, and near simultaneity in both power and time dimensions. Besides, it is difficult to transfer the fixed parameter to new households with different load profiles. Furthermore, most of these methods prove that they are able to detect events but not able to extract features for load disaggregation. This paper proposes a robust event detection method with adaptive parameters to deal with such issues. Specifically, a window with adaptive margins, multi-window screening, and adaptive threshold method is proposed to detect events in aggregated load data with high sampling rate (>1 Hz). The proposed method captures the transitions by adaptively tuning parameters including window width, margin width, and thresholds. It can also achieve good performance with blind parameter setting so that it is suitable for unknown households or datasets. Furthermore, it captures complete transitions that are indispensable for transient feature extraction. Case studies on a 20 Hz dataset, the 50 Hz LIFTED dataset, and the 60 Hz BLUED dataset show that the proposed method can robustly outperform other state-of-the-art event detection methods. The robust performance of the proposed method is also verified by a cross validation of parameters among different datasets. Lastly, the proposed event detection method is demonstrated to have the merits of improving the performance of load disaggregation.

AutoNanopore: An Automated Adaptive and Robust Method to Locate Translocation Events in Solid-State Nanopore Current Traces.

Solid-state nanopore sequencing has shown impressive performances in several research scenarios but is still challenging, mainly due to the ultrafast speed of DNA translocation and significant noises embedded in raw signals. Hence, event detection, aiming to locate precisely these translocation events, is the fundamental step of data analysis. However, existing event detection methods use either a user-defined global threshold or an adaptive threshold determined by the data, assuming the baseline current to be stable over time. These disadvantages limit their applications in real-world application scenarios, especially considering that the results of different methods are often inconsistent. In this study, we develop an automated adaptive method called AutoNanopore, for fast and accurate event detection in current traces. The method consists of three consecutive steps: current trace segmentation, current amplitude outlier identification by straightforward statistical analyses, and event characterization. Then we propose ideas/metrics on how to quantitatively evaluate the performance of an event detection method, followed by comparing the performance of AutoNanopore against two state-of-the-art methods, OpenNanopore and EventPro. Finally, we examine if one method can detect the overlapping events detected by the other two, demonstrating that AutoNanopore has the highest coverage ratio. Moreover, AutoNanopore also performs well in detecting challenging events: e.g., those with significantly varying baselines.

Short text filtering based on gradient boosting decision tree

With the rapid development of the Internet, event detection from massive data has become a research hotspot. However, the existing event detection methods for social networks rarely consider the noise data in short text data in detail. Therefore, there is a lot of noise in the input of event detection, resulting in a large number of false-positive events in the event detection results, which affects the efficiency and accuracy of event detection. In this paper, three types of features are proposed to mine text features, and the gradient boosting decision tree algorithm is used. Experiments show that the algorithm has good filtering performance and interpretability for short texts in social networks.

Event Detection in Twitter Stream Using Weighted Dynamic Heartbeat Graph Approach [Application Notes

Tweets about everyday events are published on Twitter. Detecting such events is a challenging task due to the diverse and noisy contents of Twitter. In this paper, we propose a novel approach named Weighted Dynamic Heartbeat Graph (WDHG) to detect events from the Twitter stream. Once an event is detected in a Twitter stream, WDHG suppresses it in later stages, in order to detect new emerging events. This unique characteristic makes the proposed approach sensitive to capture emerging events efficiently. Experiments are performed on three real-life benchmark datasets: FA Cup Final 2012, Super Tuesday 2012, and the US Elections 2012. Results show considerable improvement over existing event detection methods in most cases.

Enhanced Heartbeat Graph for emerging event detection on Twitter using time series networks

With increasing popularity of social media, Twitter has become one of the leading platforms to report events in real-time. Detecting events from Twitter stream requires complex techniques. Event-related trending topics consist of a group of words which successfully detect and identify events. Event detection techniques must be scalable and robust, so that they can deal with the huge volume and noise associated with social media. Existing event detection methods mostly rely on burstiness, mainly the frequency of words and their co-occurrences. However, burstiness sometimes dominates other relevant details in the data which could be equally significant. Besides, the topological and temporal relationships in the data are often ignored. In this work, we propose a novel graph-based approach, called the Enhanced Heartbeat Graph (EHG), which detects events efficiently. EHG suppresses dominating topics in the subsequent data stream, after their first detection. Experimental results on three real-world datasets (i.e., Football Association Challenge Cup Final, Super Tuesday, and the US Election 2012) show superior performance of the proposed approach in comparison to the state-of-the-art techniques.

Dynamic windowing mechanism to combine sentiment and N-gram analysis in detecting events from social media

Social sensing is a new paradigm that inherits the main ideas of sensor networks and considers the users as new sensor types. For instance, by the time the users find out that an event has happened, they start to share the related posts and express their feelings through the social networks. Consequently, these networks are becoming a powerful news media in a wide range of topics. Existing event detection methods mostly focus on either the keyword burst or sentiment of posts, and ignore some natural aspects of social networks such as the dynamic rate of arriving posts. In this paper, we devised Dynamic Social Event Detection approach that exploits a new dynamic windowing method. Besides, we add a mechanism to combine the sentiment of posts with the keywords burst in the dynamic windows. The combination of sentiment analysis and the frequently used keywords enhances our approach to detect events with a different level of user engagement. To analyze the behavior of the devised approach, we use a wide range of metrics including histogram of window sizes, sentiment oscillations of posts, topic recall, keyword precision, and keyword recall on two benchmarked datasets. One of the significant outcomes of the devised method is the topic recall of 100% for FA Cup dataset.

Event Detection and User Interest Discovering in Social Media Data Streams

Social media plays an increasingly important role in people's life. Microblogging is a form of social media which allows people to share and disseminate real-life events. Broadcasting events in microblogging networks can be an effective method of creating awareness, divulging important information, and so on. However, many existing approaches at dissecting the information content primarily discuss the event detection model and ignore the user interest which can be discovered during event evolution. This leads to difficulty in tracking the most important events as they evolve including identifying the influential spreaders. There is further complication given that the influential spreaders interests will also change during event evolution. The influential spreaders play a key role in event evolution and this has been largely ignored in traditional event detection methods. To this end, we propose a user-interest model-based event evolution model, named the hot event evolution model. This model not only considers the user interest distribution but also uses the short text data in the social network to model the posts and the recommend methods to discover the user interests. This can resolve the problem of data sparsity, as exemplified by many existing event detection methods, and improve the accuracy of event detection. A hot event automatic filtering algorithm is initially applied to remove the influence of general events, improving the quality and efficiency of mining the event. Then, an automatic topic clustering algorithm is applied to arrange the short texts into clusters with similar topics. An improved user-interest model is proposed to combine the short texts of each cluster into a long text document simplifying the determination of the overall topic in relation to the interest distribution of each user during the evolution of important events. Finally, a novel cosine measurebased event similarity detection method is used to assess correlation between events, thereby detecting the process of event evolution. The experimental results on a real Twitter data set demonstrate the efficiency and accuracy of our proposed model for both event detection and user interest discovery during the evolution of hot events.

An event detection method for social networks based on hybrid link prediction and quantum swarm intelligent

How can we discover and estimate major events in complex social networks? Even with ever enlarging networks and data scale? Event detection and evaluation in social networks provide an effective solution, which has become the critical basis for many real applications, such as crisis management and decision making. The existing event detection methods mainly focus on text analysis that is limited in social media content and graphic feature statistic that needs calculate vast variables. Can we find an efficient way for generalized social networks with limited topology information? In this paper, a novel hybrid quantum swarm intelligence indexing method (HQSII) from the perspective of link prediction is proposed for the first time, which includes an optimal weight algorithm (OWA) and a fluctuation detection algorithm (FDA). The innovations behind HQSII lie in three aspects: (1) The mixed index that can universally describe the social network evolutions is proposed firstly, which explores the cooperation of different independent similarity indexes. OWA is further proposed to determine the optimal mixed index that achieves higher link prediction accuracy and better network evolution description than other independent and mixed similarity indexes. (2) To better avoid the interferences of routine network evolution fluctuations, the otherness of micro node evolutions is considered into link prediction. FDA is further proposed to quantify the abnormal fluctuations caused by events. (3) Based on OWA and FDA, HQSII is proposed for all the generalized social networks, which detects events by discovering abnormal fluctuations and evaluates events by analyzing fluctuation trends. Extensive experiments on theoretical and real-world social networks show that HQSII can accurately detect events and quantitatively evaluate event impacts in social networks with single and multiple events.

Detection of events in seismic time series by time–frequency methods

The detection of events in seismic time series has been a subject of great interest during the last 30 years. Most of the works in this area were based on detecting special patterns or clusters in seismic data. The authors present here a event detection method based on a time–frequency analysis through the Wigner distribution (WD). The proposed method consists on defining an appropriate entropic measure through a suitable time–frequency distribution, acting as probability distribution function. It is known from previous studies in the field that the information entailed by time–frequency representations (TFR) of time signals can be explored by means of different Rényi entropy measures. The non-positivity character of the WD implies that the classical Shannon entropy cannot be used, and therefore it has been replaced by a generalised measure such as the Rényi entropy. However, owing to the existence of multiple TFR normalisations, the so-called quantum normalisation has been empirically selected here for this particular application. This method is based on the identification of the events as those temporal clusters having the highest amount of information (entropy). The method is described and applied to different earthquake signals and volcanic tremors, using both real and synthetic data. The results are compared to other existing event detection methods.

An adaptive threshold framework for event detection using HMM-based life profiles

When an event occurs, it attracts attention of information sources to publish related documents along its lifespan. The task of event detection is to automatically identify events and their related documents from a document stream, which is a set of chronologically ordered documents collected from various information sources. Generally, each event has a distinct activeness development so that its status changes continuously during its lifespan. When an event is active, there are a lot of related documents from various information sources. In contrast when it is inactive, there are very few documents, but they are focused. Previous works on event detection did not consider the characteristics of the event's activeness, and used rigid thresholds for event detection. We propose a concept called life profile, modeled by a hidden Markov model, to model the activeness trends of events. In addition, a general event detection framework, LIPED, which utilizes the learned life profiles and the burst-and-diverse characteristic to adjust the event detection thresholds adaptively, can be incorporated into existing event detection methods. Based on the official TDT corpus and contest rules, the evaluation results show that existing detection methods that incorporate LIPED achieve better performance in the cost and F1 metrics, than without.