Text Streams Research Articles

A nonparametric model for online topic discovery with word embeddings

With the explosive growth of short documents generated from streaming textual sources (e.g., Twitter), latent topic discovery has become a critical task for short text stream clustering. However, most online clustering models determine the probability of producing a new topic by manually setting some hyper-parameter/threshold, which becomes barrier to achieve better topic discovery results. Moreover, topics generated by using existing models often involve a wide coverage of the vocabulary which is not suitable for online social media analysis. Therefore, we propose a nonparametric model (NPMM) which exploits auxiliary word embeddings to infer the topic number and employs a “spike and slab” function to alleviate the sparsity problem of topic-word distributions in online short text analyses. NPMM can automatically decide whether a given document belongs to existing topics, measured by the squared Mahalanobis distance. Hence, the proposed model is free from tuning the hyper-parameter to obtain the probability of generating new topics. Additionally, we propose a nonparametric sampling strategy to discover representative terms for each topic. To perform inference, we introduce a one-pass Gibbs sampling algorithm based on Cholesky decomposition of covariance matrices, which can further be sped up using a Metropolis-Hastings step. Our experiments demonstrate that NPMM significantly outperforms the state-of-the-art algorithms.

Brother or ‘Other’? Transformation of strategic narratives in Russian television news during the Ukrainian crisis

This article explores how Russian television news deconstructed the narrative embracing Ukrainians as ‘brothers’ through repositioning them within an imagined social reality wherein Ukrainians assume the guise of a threatening ‘Other’. The research material comprises extracts from Channel One, which is one of Russia’s most significant and popular television channels. The data was collected from the channel’s website over two years from 1 November 2012 to 31 October 2014. The sample of 480 news stories was selected on the basis of tag words and engagement with the stories on social media. Frames comprising visual, auditory and textual streams were analysed to explore the narratives about Ukrainians that were promulgated both before and during the Ukrainian crisis. The data revealed that, prior to the conflict, the media portrayed Ukraine as Russia’s ‘little brother’. Following Euromaidan, Channel One replaced this guise with that of an enemy in order to appeal directly to the cultural and spiritual values of its audience. Applying the conceptual framework of strategic narrative, the study illustrates the dynamics at play in the transformation of one image into another.

An Augmented Intelligence Model to Extract Pragmatic Markers from Large Content Repositories

This paper presents a novel methodology for automatically extracting pragmatic markers from large streams of texts and repositories of documents. Pragmatic markers typically are implications, innuendos, suggestions, contradictions, sarcasms or references that are difficult to define objectively, but that are subjectively evident. Our methodology uses a two-stage augmented learning model applied to a specific use case, extracting from a repository of over 1500 Article IV country reports prepared for government officials by International Monetary Fund (IMF) staff. The model uses principles of evidence theory to train a machine to decipher the textual patterns of suggested actions for government officials and to extract those suggestions from the country reports at scale. We demonstrate the effectiveness of the model with impressive precision and recall metrics that over time outperform even the human trainers.

A WEB-BASED FAST AND RELIABLE TEXT CLASSIFICATION TOOL

INTRODUCTION
 Opinion analysis in the big data analysis context has been a hot topic in science and the business world recently. Social media has become a key data source for opinions generating a large amount of data every day providing content for further analysis.
 In the Big data age, unstructured data classification is one of the key tools for fast and reliable content analysis. I expect significant growth in the demand for content classification services in the nearest future.
 There are many online text classification tools available providing limited functionality -such as automated text classification in predefined categories and sentiment analysis based on a pre-trained machine learning algorithm. The limited functionality does not provide tools such as data mining support and/or a machine learning algorithm training interface.
 There are a limited number of tools available providing the whole sets of tools required for text classification, i.e. this includes all the steps starting from data mining till building a machine learning algorithm and applying it to a data stream from a social network source. My goal is to create a tool able to generate a classified text stream directly from social media with a user friendly set-up interface.
 METHODS AND MATERIALS
 The text classification tool will have a core based modular structure (each module providing certain functionality) so the system can be scaled in terms of technology and functionality.
 The tool will be built on open source libraries and programming languages running on a Linux OS based server. The tool will be based on three key components: frontend, backend and data storage as described below:
 
 backend: Python and Nodejs programming language with machine learning and text filtering libraries: TensorFlow, and Keras,
 for data storage Mysql 5.7/8 will be used,
 frontend will be based on web technologies built using PHP and Javascript.
 
 EXPECTED RESULTS
 The expected result of my work is a web-based text classification tool for opinion analysis using data streams from social media. The tool will provide a user friendly interface for data collection, algorithm selection, machine learning algorithm setup and training.
 Multiple text classification algorithms will be available as listed below:
 
 Linear SVM
 Random Forest
 Multinomial Naive Bayes
 Bernoulli Naive Bayes
 Ridge Regressio
 Perceptron
 Passive Aggressive Classifier
 Deep machine learning algorithm.
 
 System users will be able to identify the most effective algorithm for their text classification task and compare them based on their accuracy.
 The architecture of the text classification tool will be based on a frontend interface and backend services. The frontend interface will provide all the tools the system user will be interacting with the system. This includes setting up data collection streams from multiple social networks and allocating them to pre-specified channels based on keywords.
 Data from each channel can be classified and assigned to a pre-defined cluster. The tool will provide a training interface for machine learning algorithms.
 This text classification tool is currently in active development for a client with planned testing and implementation in April 2019.

User interest community detection on social media using collaborative filtering

Community detection in microblogging environment has become an important tool to understand the emerging events. Most existing community detection methods only use network topology of users to identify optimal communities. These methods ignore the structural information of the posts and the semantic information of users’ interests. To overcome these challenges, this paper uses User Interest Community Detection model to analyze text streams from microblogging sites for detecting users’ interest communities. We propose HITS Latent Dirichlet Allocation model based on modified Hypertext Induced Topic Search and Latent Dirichlet Allocation to distil emerging interests and high-influence users by reducing negative impact of non-related users and its interests. Moreover, we propose HITS Label Propagation Algorithm method based on Label Propagation Algorithm and Collaborative Filtering to segregate the community interests of users more accurately and efficiently. Our experimental results demonstrate the effectiveness of our model on users’ interest community detection and in addressing the data sparsity problem of the posts.

Collaboratively Tracking Interests for User Clustering in Streams of Short Texts

In this paper, we aim at tackling the problem of user clustering in the context of their published short text streams. Clustering users by short text streams is more challenging than in the case of long documents associated with them as it is difficult to track users' dynamic interests in streaming sparse data. To obtain better user clustering performance, we propose two user collaborative interest tracking models that aim at tracking changes of each user's dynamic topic distributions in collaboration with their followees' dynamic topic distributions, based both on the content of current short texts and the previously estimated distributions. Our models can be either short-term or long-term dependency topic models. Short-term dependency model collaboratively tracks users' interests based on users' topic distributions at the previous time period only, whereas long-term dependency model collaboratively tracks users' interests based on users' topic distributions at multiple time periods in the past. We also propose two collapsed Gibbs sampling algorithms for collaboratively inferring users' dynamic interests for their clustering in our short-term and long-term dependency topic models, respectively. We evaluate our proposed models via a benchmark dataset consisting of Twitter users and their tweets. Experimental results validate the effectiveness of our proposed models that integrate both users' and their collaborative interests for user clustering by short text streams.

Semantic-Aware Visual Abstraction of Large-Scale Social Media Data With Geo-Tags

With the rapid growth of geo-tagged social media data, it has become feasible to explore topics across different areas through text mining and geographical visualization. However, the visual elements of social media data always overlap with each other in the map view, which largely disturbs visual perception of semantic features and their geographical distribution. Thus, it is of great significance to reduce the visual clutter of large-scale social media data, and enhance the visibility of semantic features across local areas. In this paper, we utilize a doc2vec model to transform geo-tagged social media data into high-dimensional vectors, and the semantic correlation can be easily characterized in the dimensionality reduction space. Aiming at the reduction of visual clutter of geographical visualization, a dual-objective blue noise sampling model is proposed to select a subset of social media data, by means of which both the semantic correlation and spatial distribution of large scale social media data are well retained. A rich set of visual designs are implemented enabling users to evaluate the sampled results from multiple perspectives and explore the changes of semantic features across areas, such as heatmap, word cloud and text stream. The effectiveness and validity of the proposed visualization system are further demonstrated through case studies and expert reviews.

Dynamic Feature Selection for Clustering High Dimensional Data Streams

Change in a data stream can occur at the concept level and at the feature level. Change at the feature level can occur if new, additional features appear in the stream or if the importance and relevance of a feature changes as the stream progresses. This type of change has not received as much attention as concept-level change. Furthermore, a lot of the methods proposed for clustering streams (density-based, graph-based, and grid-based) rely on some form of distance as a similarity metric and this is problematic in high-dimensional data where the curse of dimensionality renders distance measurements and any concept of “density” difficult. To address these two challenges we propose combining them and framing the problem as a feature selection problem, specifically a dynamic feature selection problem. We propose a dynamic feature mask for clustering high dimensional data streams. Redundant features are masked and clustering is performed along unmasked, relevant features. If a feature's perceived importance changes, the mask is updated accordingly; previously unimportant features are unmasked and features which lose relevance become masked. The proposed method is algorithm-independent and can be used with any of the existing density-based clustering algorithms which typically do not have a mechanism for dealing with feature drift and struggle with high-dimensional data. We evaluate the proposed method on four density-based clustering algorithms across four high-dimensional streams; two text streams and two image streams. In each case, the proposed dynamic feature mask improves clustering performance and reduces the processing time required by the underlying algorithm. Furthermore, change at the feature level can be observed and tracked.

AMASP (ASCII Master Slave Protocol): a Lightweight MODBUS Based Customizable Communication Protocol for General Applications

In this paper, we describe AMASP, a proposal for an open standard communication protocol / model whose objective is providing a novel and lightweight solution to the problem of communication between small computers acting in particular on embedded systems, but not limited to them. The protocol establishes addresses for computers or peripherals, allowing them to send messages directly to these addresses. It is based on message interchanging, being the messages composed by ASCII (American Standard Code for Information Interchange) text or binary streams, according to the user customization. AMASP is very customizable using four different packets to send the messages and manage a master / slave communication model. The protocol is designed to work where a direct communication link is established or a bus architecture is available - by serial, USB connections, etc - and it supports to many error checking algorithms that can be used in unsafe links. Due to the focus on simplicity, low overhead and metadata, it is a connectionless protocol and does not support routing.

DEEP-HEAR: A Multimodal Subtitle Positioning System Dedicated to Deaf and Hearing-Impaired People

In this paper, we introduce the DEEP-HEAR framework, a multimodal dynamic subtitle positioning system designed to increase the accessibility of deaf and hearing impaired people (HIP) to multimedia documents. The proposed system exploits both computer vision algorithms and deep convolutional neural networks specifically designed and tuned in order to detect and recognize the identity of the active speaker. The main contributions of the paper concern: a novel method dedicated to recognizing various characters existent in the video stream. A video temporal segmentation algorithm that divides the video sequence into semantic units, based on face tracks and visual consistency. Finally, the core of our approach concerns a novel active speaker recognition method relying on the multimodal information fusion from the text, audio, and video streams. The experimental results carried out on a large scale dataset of more than 30 videos, validate the proposed methodology with average accuracy and recognition rates superior to 90%. Moreover, the method shows robustness to important object/camera motion and face pose variation, yielding gains of more than 8% in precision and recall rates when compared with state-of-the-art techniques. The subjective evaluation of the proposed dynamic subtitle positioning system demonstrates the effectiveness of our approach.

Fish Everywhere by Britta Teckentrup

Reviewed by: Fish Everywhere by Britta Teckentrup Elizabeth Bush Teckentrup, Britta Fish Everywhere; written and illus. by Britta Teckentrup. Big Picture/Candlewick, 2019 [32p] ISBN 978-1-5362-0625-8 $16.99 Reviewed from galleys R 4-9 yrs There's a literary sweet spot for kids who delight in sea life, in amassing sharable details, and in immersive (pun fully intended) experiences, and Teckentrup hits that mark with a lushly illustrated picture book made for the oxymoronic serious browser. Each double-page spread in this British import features a briny or freshwater environment or fish-related topic; elaborate pictorial detail and chatty text and captioning will keep lap sitters and floor sprawlers rapt. Embedded within most of the watercolor and cut-paper scenes are viewing challenges: for example, determine which creatures are not fish; locate an Atlantic cod at various sea levels; pick out distinctively colored male, female, and juvenile parrot fish in a reef; go back through the book to find the lone bristlemouth—a daunting mission unless one follows a significant clue. The need to occasionally flip the book ninety degrees [End Page 408] or to follow winding streams of text assures yet another level of engagement. This is a great choice for kids heading off to lake, river, ocean, or aquarium adventures, or those just itching to trade their jeans for waders or a wetsuit. Copyright © 2019 The Board of Trustees of the University of Illinois

Burst Hotspots Dynamic Detection and Tracking on Large-Scale Text Stream

The online social network has become an important communication tool for people and forms a virtual society interacting with the real world. The numerous events rapidly spread through social networks and may become hotspots in a short period of time. Especially, the negative events vibrate national security and social stability, potentially causing a series of social problems. Therefore, detection and tracking burst hotspots on social networks are of great significance. However, this problem is non-trivial because of the challenges of massive noise, sparsity, high-dimensionality, and dynamic changing. To address these challenges, this paper proposes a distributed method of burst hotspots dynamic detection and tracking based on Map/Reduce. To judge the relevance of the current text and previous texts, the keyword similarity matrix is calculated by using Word2Vec and context information. The keyword weights are modified according to the association model to further reduce noise. The proposed method is not only robust to the sparsity of short text but also overcomes the curse of dimensionality. Finally, it can dynamically detect and track burst hotspots from large-scale short text stream on the Hadoop platform. The experiments have shown that the proposed method outperforms state-of-the-art algorithms.

Dynamic topic modeling via self-aggregation for short text streams

Social networks such as Twitter, Facebook, and Sina microblogs have emerged as major sources for discovering and sharing the latest topics. Because social network topics change quickly, developing an effective method to model such topics is urgently needed. However, topic modeling is challenging due to the sparsity problem and the dynamic change of topics in microblog streams. In this study, we propose dynamic topic modeling via a self-aggregation method (SADTM) that can capture the time-varying aspect of topic distributions and resolve the sparsity problem. The SADTM aggregates the observable and unordered short texts as the aggregated document without leveraging an external context to overcome the sparsity problem of short text. Furthermore, we exploit word pairs instead of words for each microblog to generate more word co-occurrence patterns. The SADTM models temporal dynamics by using the topic distribution at previous time steps in microblog streams to infer the current topic from sequential data. Extensive experiments on a real-world Sina microblog dataset demonstrate that our SADTM algorithm outperforms several state-of-the-art methods in topic coherence and cluster quality. Additionally, when applied in a search scene, our SADTM significantly outperforms all baseline methods in terms of the quality of the search results.

Online Biterm Topic Model based short text stream classification using short text expansion and concept drifting detection

Short text stream classification suffers from enormous challenges, due to the sparsity, high dimension and rapid variability of the short text stream. In this paper, we present a short text stream classification approach refined from online Biterm Topic Model (BTM) using short text expansion and concept drifting detection. Specifically, in our method, we firstly extend short text streams from an external resource to make up for the sparsity of data, and use online BTM to select representative topics instead of the word vector to represent the feature of short texts. Secondly, we propose a concept drift detection method based on the topic model to detect the hidden concept drifts in short text streams. Thirdly, we build an ensemble model using several data chunks and update with the newest data chunk and results of the concept drift detection. Finally, extensive experimental results demonstrate that compared to well-known baselines, our approach achieves a better performance in the classification and concept drifting detection.

An Evolutionary Context-aware Sequential Model for topic evolution of text stream

Social media acts as the platform for users to acquire information and spreads out breaking news. The overwhelming amount of fast-growing information makes it a challenge to track the subsequences of the breaking news or events and find the corresponding user opinions towards special aspects. Tracking the evolution of an event and predicting its subsequent trends play an important role in social media. In this paper, we propose an Evolutionary Context-aware Sequential model (ECSM) to track the evolutionary trends of the streaming text and investigate their focused context-aware topics. We integrate two novel layers into the Recurrent Chinese Restaurant Process (RCRP), respectively one context-aware topic layer and one Long Short Term Memory (LSTM) based sequential layer. The context-aware topic layer can help capture the global context-aware semantic coherences and the sequential layer is exploited to learn the local dynamics and semantic dependencies during the dynamic evolutionary process. Experimental results on real datasets show that our method significantly outperforms the state-of-the-art approaches.

Text Classification Using the N-Gram Graph Representation Model Over High Frequency Data Streams

A prominent challenge in our information age is the classification over high frequency data streams. In this research, we propose an innovative and high-accurate text stream classification model that is designed in an elastic distributed way and is capable to service text load with fluctuated frequency. In this classification model, text is represented as N-Gram Graphs and the classification process takes place using text preprocessing, graph similarity and feature classification techniques following the supervised machine learning approach. The work involves the analysis of many variations of the proposed model and its parameters, such as various representations of text as N-Gram Graphs, graph comparisons metrics and classification methods in order to conclude to the most accurate setup. To deal with the scalability, the availability and the timely response in case of high frequency text we employ the Beam programming model. Using the Beam programming model the classification process occurs as a sequence of distinct tasks and facilitates the distributed implementation of the most computational demanding tasks of the inference stage. The proposed model and the various parameters that constitute it are evaluated experimentally and the high frequency stream emulated using two public datasets (20NewsGroup and Reuters-21578) that are commonly used in the literature for text classification

A Knowledge-Based Semisupervised Hierarchical Online Topic Detection Framework.

Topic models have achieved big success in recent years. To detect topics in a text stream, various online topic models have been proposed in the literature. The limitations of these works include that: 1) most of them run with fixed topic numbers and 2) the overlaps between the topics may enlarge in the evolving process. Hierarchical topic model is a candidate solution to these problems since it can reveal many useful relationships between the topics. These relationships can help to find high quality topics and reduce topic overlaps. In this paper, a knowledge-based semisupervised hierarchical online topic detection framework is proposed. The proposed framework can detect topics in an online hierarchical way. In addition, it has been proven that introducing external knowledge can improve the performance of text mining. Therefore, the knowledge from external knowledge sources and human experts are also integrated in the proposed framework. Experiments are conducted to evaluate the proposed framework with different metrics. The results show that compared with the baseline methods, our framework can achieve better performance with competitive time efficiency.

Spatio-temporal top-k term search over sliding window

In part due to the proliferation of GPS-equipped mobile devices, massive volumes of geo-tagged streaming text messages are becoming available on social media. It is of great interest to discover most frequent nearby terms from such tremendous stream data. In this paper, we present novel indexing, updating, and query processing techniques that are capable of discovering top-k most frequent nearby terms over a sliding window. Specifically, given a query location and a set of geo-tagged messages within a sliding window, we study the problem of searching for the top-k terms by considering term frequency, spatial proximity, and term freshness. We develop a novel and efficient mechanism to solve the problem, including a quad-tree based indexing structure, indexing update technique, and a best-first based searching algorithm. An empirical study is conducted to show that our proposed techniques are efficient and fit for users’ requirements through varying a number of parameters.

Introducing time series chains: a new primitive for time series data mining

Time series motifs were introduced in 2002 and have since become a fundamental tool for time series analytics, finding diverse uses in dozens of domains. In this work, we introduce Time Series Chains, which are related to, but distinct from, time series motifs. Informally, time series chains are a temporally ordered set of subsequence patterns, such that each pattern is similar to the pattern that preceded it, but the first and last patterns can be arbitrarily dissimilar. In the discrete space, this is similar to extracting the text chain “data, date, cate, cade, code” from text stream. The first and last words have nothing in common, yet they are connected by a chain of words with a small mutual difference. Time series chains can capture the evolution of systems, and help predict the future. As such, they potentially have implications for prognostics. In this work, we introduce two robust definitions of time series chains and scalable algorithms that allow us to discover them in massive complex datasets.

Lightly supervised alignment of subtitles on multi-genre broadcasts

This paper describes a system for performing alignment of subtitles to audio on multigenre broadcasts using a lightly supervised approach. Accurate alignment of subtitles plays a substantial role in the daily work of media companies and currently still requires large human effort. Here, a comprehensive approach to performing this task in an automated way using lightly supervised alignment is proposed. The paper explores the different alternatives to speech segmentation, lightly supervised speech recognition and alignment of text streams. The proposed system uses lightly supervised decoding to improve the alignment accuracy by performing language model adaptation using the target subtitles. The system thus built achieves the third best reported result in the alignment of broadcast subtitles in the Multi–Genre Broadcast (MGB) challenge, with an F1 score of 88.8%. This system is available for research and other non–commercial purposes through webASR, the University of Sheffield’s cloud–based speech technology web service. Taking as inputs an audio file and untimed subtitles, webASR can produce timed subtitles in multiple formats, including TTML, WebVTT and SRT.