Unsupervised event exploration from social text streams

Abstract

Abstract. Social media provides unprecedented opportunities for people to disseminate information and share their opinions and views online. Extracting events from social media platforms such as Twitter could help in understanding what is being discussed. However, event extraction from social text streams poses huge challenges due to the noisy nature of social media posts and dynamic evolution of language. We propose a generic unsupervised framework for exploring events on Twitter which consists of four major steps, filtering, pre-processing, extraction and categorization, and post-processing. Tweets published in a certain time period are aggregated and noisy tweets which do not contain newsworthy events are filtered by the filtering step. The remaining tweets are pre-processed by temporal resolution, part-of-speech tagging and named entity recognition in order to identify the key elements of events. An unsupervised Bayesian model is proposed to automatically extract the structured representations of events in the form of quadruples < entity, keyword, date, location > and further categorize the extracted events into event types. Finally, the categorized events are assigned with the event type labels without human intervention. The proposed framework has been evaluated on over 60 million tweets which were collected for one month in December 2010. A precision of 78.01% is achieved for event extraction using our proposed Bayesian model, outperforming a competitive baseline by nearly 13.6%. Moreover, events are also clustered into coherence groups with the automatically assigned event type labels with an accuracy of 42.57%.

The myth of reproducibility: A review of event tracking evaluations on Twitter

“…In event modeling, researchers could not represent events formally without a structured definition of events, which some found in the ‘four Ws', or Who did What, Where and When (Chen and Li, 2020 ). To expect event tracking algorithms to model events might appear unrealistic, but we could express the ground truth in terms of the “four Ws.” Similarly to Zhou et al ( 2015 , 2017 ), we could expect a document-pivot technique's clusters to describe Who did What, Where and When, or a feature-pivot technique's keywords to capture them.…”

“…Chakrabarti and Punera (2011) devised five labels to describe the state of American football games, plays within those games, and other general comments. Zhou et al (2015Zhou et al ( , 2017 presented slightly more rigid rules based on Who does What, Where and When, but such examples appear scarcely. Most researchers leave the labeling process to the discretion of the annotators, and in discretion, subjectivity prevails.…”

ATM: Adversarial-neural Topic Model

“…• LEM [5], is a bayesian modeling approach for open domain event extraction. It treats an event as a latent variable and models the generation of an event as a joint distribution of its individual event elements (organization , location , person , keyword) 7 . We implement the algorithm with the default configuration.…”

“…Topic models [1,2] underpin many successful applications within the field of Natural Language Processing (NLP). Variants of topic models have been proposed for different tasks including content analysis of e-petitions [3], topic-associated sentiment analysis [4], event extraction from social media [5,6,7] and product aspect mining [8]. However, topic models typically rely on mean-field variational inference [9] or collapsed Gibbs sampling for model learning.…”

Hybrid Onion Layered System for the Analysis of Collective Subjectivity in Social Networks

“…The results show that detection, extraction, and classification of emotions, feelings, and opinions are the main applications coded as private states analysis. This category records 33 techniques, algorithms, or methods for performing [125] (TSSE) Tweet Sentiment Score Estimator [104] (BM) Naive Bayes, Bayesian Logistic [126]; [127] (LSA) Latent Semantic Analysis [128] (LIWC) Linguistic Inquiry and Word Count [129]; [130]; [131] (SANT) Sociological Approach to handling Noisy and short Texts [132] (SC) Sarcasm (TPR) True Positive Ratio [92] (SVM) Support Vector Machine [92] (LRS) Linguistic Rules Sarcasm [133]; [124] (TC) Text Classification (SVM) Support Vector Machine [134]; [135] (ENS) Ensemble Classifiers [135]; [136] (LECM) Latent Event Category Model [137] (BM) Naive Bayes, Bayesian Logistic [137]; [138] (RF) Random Forest [139] (LR) Logistic Regression [140] (SE) Search (FL) Fuzzy Logic [141] (TF-IDF) Term Frecuency [142]; [143] (KB) Knowledge Base (ON) Ontologias [144]; [145]; [95]; [146]; [147] (SI) Social Influence (PN) Proximity Networks [102] (PR) Pagerank [113] (ST) Statistical techniques [100] (BM) Naive Bayes, Bayesian Logistic [111] (DF) Difussion (RM) Rumor Model [117] (BM) Naive Bayes, Bayesian Logistic [127] (ST) Statistical techniques [122] (VAM) Vector Autoregressive Model…”