Temporal Topics Research Articles

Sparseness-constrained nonnegative tensor factorization for detecting topics at different time scales

Temporal text data, such as news articles or Twitter feeds, often comprises a mixture of long-lasting trends and transient topics. Effective topic modeling strategies should detect both types and clearly locate them in time. We first demonstrate that nonnegative CANDECOMP/PARAFAC decomposition (NCPD) can automatically identify topics of variable persistence. We then introduce sparseness-constrained NCPD (S-NCPD) and its online variant to control the duration of the detected topics more effectively and efficiently, along with theoretical analysis of the proposed algorithms. Through an extensive study on both semi-synthetic and real-world datasets, we find that our S-NCPD and its online variant can identify both short- and long-lasting temporal topics in a quantifiable and controlled manner, which traditional topic modeling methods are unable to achieve. Additionally, the online variant of S-NCPD shows a faster reduction in reconstruction error and results in more coherent topics compared to S-NCPD, thus achieving both computational efficiency and quality of the resulting topics. Our findings indicate that S-NCPD and its online variant are effective tools for detecting and controlling the duration of topics in temporal text data, providing valuable insights into both persistent and transient trends.

A Geometry-Driven Longitudal Topic Model

A simple and scalable framework for longitudinal analysis of Twitter data is developed that combines latent topic models with computational geometric methods. Dimensionality reduction tools from computational geometry are applied to learn the intrinsic manifold on which the latent, temporal topics reside. Then shortest path distances on the manifold are used to link together these topics. The proposed framework permits visualization of the low-dimensional embedding, which provides clear interpretation of the complex, high-dimensional trajectories that may exist among latent topics. Practical application of the proposed framework is demonstrated through its ability to capture and effectively visualize natural progression of latent COVID-19–related topics learned from Twitter data. Interpretability of the trajectories is achieved by comparing to real-world events. In addition, the framework permits study of spatial variation in Twitter behavior for learned topics. The analysis demonstrates that the proposed framework is able to capture granular-level impact of COVID-19 on public discussions. We end by arguing that Twitter data, when analyzed within the proposed framework, can serve as a valuable supplementary data stream for COVID-related studies.

Inferring Global-Scale Temporal Latent Topics From News Reports to Predict Public Health Interventions for COVID-19

Inferring Global-Scale Temporal Latent Topics From News Reports to Predict Public Health Interventions for COVID-19

An alternative topic model based on Common Interest Authors for topic evolution analysis

• Authors are more topically similar when they are from multigraph clusters • Changes in the definition of topics lead to more human-readable topics • Topics from only meta-data of the publication records show higher topical coherence • Use of multiple bibliographic networks is beneficial to author-based topic modeling • Author-based topic modeling allows effective merge/split topic evolution tracking Topic modeling methods aim to extract semantic topics from unstructured documents, and topic evolution is one of its branches seeking to analyze how temporal topics in a set of documents evolve and has shown successful identification of content transitions within static topics over time; yet, the inherent limitations of topic modeling methods inhibit traditional topic evolution methods from highlighting topical correlations between different, dynamic topics. The authors propose an alternative topic modeling method conscious of the topical correlation in the academic domain by introducing the notion of the common interest authors (CIA 1 1 CIA: Common Interest Authors ), defining a topic as a set of shared common research interests of a researcher group. Publication records related to the Human Computer Interaction field were extracted from the Microsoft Academic Graph dataset, with virtual reality as the target field of research. The result showed that the proposed alternative topic modeling is capable of successfully model coherent topics regardless of the topic size with only the meta-data of the document set, indicating that the alternative approach is not only capable of allowing topic correlation analysis during the topic evolution but also able to generate coherent topics at the same time.

LEARNING WORD RELATEDNESS OVER TIME FOR TEMPORAL RANKING

Queries and ranking with temporal aspects gain significant attention in field of Information Retrieval. While searching for articles published over time, the relevant documents usually occur in certain temporal patterns. Given a query that is implicitly time sensitive, we develop a temporal ranking using the important times of query by drawing from the distribution of query trend relatedness over time. We also combine the model with Dual Embedding Space Model (DESM) in the temporal model according to document timestamp. We apply our model using three temporal word embeddings algorithms to learn relatedness of words from news archive in Bahasa Indonesia: (1) QT-W2V-Rank using Word2Vec (2) QT-OW2V-Rank using OrthoTrans-Word2Vec (3) QT-DBE-Rank using Dynamic Bernoulli Embeddings. The highest score was achieved with static word embeddings learned separately over time, called QT-W2V-Rank, which is 66% in average precision and 68% in early precision. Furthermore, studies of different characteristics of temporal topics showed that QT-W2V-Rank is also more effective in capturing temporal patterns such as spikes, periodicity, and seasonality than the baselines.

Discovery and tracking of temporal topics of interest based on belief-function and aging theories

Content-based microblogging networks like Twitter have a significant and fast evolution of the information amount. In addition to textual content and member activities, analyzing structural information concerning user neighborhood and communities is an opportunity for several applications such as user interest modeling. In this context, this work proposes a new approach for temporal topics of interest derivation and tracking in the Twitter micro-blogging network. The challenge of this approach is to take into consideration the different dimensions of the user’s interest such as personal, social and temporal aspect. We propose a new definition of an ego network that incorporates friendship reliability and dependency between members. The belief function theory that allows combining evidence from dependent or independent sources in an uncertain context is introduced. This theory has been widely utilized in various decision-making systems. To model the topics of interest evolution, we introduce the aging theory suggested mainly for event life-cycle detection and modeling in a micro-blogging network. We present an experimental study performed on the actual Twitter users, validating our approach. This model is compared to other methods using the same database.

A dynamic human activity‐driven model for mixed land use evaluation using social media data

AbstractMassive amounts of social media data have enabled the evaluation of mixed land use from the human activities perspective. While most studies have focused on approaches to better measure mixed land use, the dynamic activity patterns used to distinguish land use types have seldom been discussed. In light of this, in this study, a mixed land use evaluation model is constructed, which is driven by dynamic human activities hidden in social media data. Specifically, temporal topics related to human activities are first extracted from Twitter messages. On this basis, these topics are clustered with successive measurements of both temporal and semantic similarities, and then, they are further transformed into dynamic features. In addition, the dynamic feature‐based Shannon entropy is employed for mixed land use evaluation. For verification, the building classes from OpenStreetMap are referenced to investigate their correlations with dynamic features in mixed land uses. Finally, a strong correlation demonstrates the effectiveness and power of the proposed method in the evaluation of mixed land uses.

Microscopic and Macroscopic Spatio-Temporal Topic Models for Check-in Data

Twitter, together with other online social networks, such as Facebook, and Gowalla have begun to collect hundreds of millions of check-ins. Check-in data captures the spatial and temporal information of user movements and interests. To model and analyze the spatio-temporal aspect of check-in data and discover temporal topics and regions, we first propose a spatio-temporal topic model, i.e., Upstream Spatio-Temporal Topic Model (USTTM). USTTM can discover temporal topics and regions, i.e., a user’s choice of region and topic is affected by time in this model. We use continuous time to model check-in data, rather than discretized time, avoiding the loss of information through discretization. In addition, USTTM captures the property that user’s interests and activity space will change over time, and users have different region and topic distributions at different times in USTTM. However, both USTTM and other related models capture “microscopic patterns” within a single city, where users share POIs, and cannot discover “macroscopic” patterns in a global area, where users check-in to different POIs. Therefore, we also propose a macroscopic spatio-temporal topic model, MSTTM, employing words of tweets that are shared between cities to learn the topics of user interests. We perform an experimental evaluation on Twitter and Gowalla data sets from New York City and on a Twitter US data set. In our qualitative analysis, we perform experiments with USTTM to discover temporal topics, e.g., how topic “tourist destinations” changes over time, and to demonstrate that MSTTM indeed discovers macroscopic, generic topics. In our quantitative analysis, we evaluate the effectiveness of USTTM in terms of perplexity, accuracy of POI recommendation, and accuracy of user and time prediction. Our results show that the proposed USTTM achieves better performance than the state-of-the-art models, confirming that it is more natural to model time as an upstream variable affecting the other variables. Finally, the performance of the macroscopic model MSTTM is evaluated on a Twitter US dataset, demonstrating a substantial improvement of POI recommendation accuracy compared to the microscopic models.

Unsupervised adaptive microblog filtering for broad dynamic topics

Information filtering has been a major task of study in the field of information retrieval (IR) for a long time, focusing on filtering well-formed documents such as news articles. Recently, more interest was directed towards applying filtering tasks to user-generated content such as microblogs. Several earlier studies investigated microblog filtering for focused topics. Another vital filtering scenario in microblogs targets the detection of posts that are relevant to long-standing broad and dynamic topics, i.e., topics spanning several subtopics that change over time. This type of filtering in microblogs is essential for many applications such as social studies on large events and news tracking of temporal topics. In this paper, we introduce an adaptive microblog filtering task that focuses on tracking topics of broad and dynamic nature. We propose an entirely-unsupervised approach that adapts to new aspects of the topic to retrieve relevant microblogs. We evaluated our filtering approach using 6 broad topics, each tested on 4 different time periods over 4 months. Experimental results showed that, on average, our approach achieved 84% increase in recall relative to the baseline approach, while maintaining an acceptable precision that showed a drop of about 8%. Our filtering method is currently implemented on TweetMogaz, a news portal generated from tweets. The website compiles the stream of Arabic tweets and detects the relevant tweets to different regions in the Middle East to be presented in the form of comprehensive reports that include top stories and news in each region.

User Group Oriented Temporal Dynamics Exploration

Temporal online content becomes the zeitgeist to reflect our interests and changes. Active users are essential participants and promoters behind it. Temporal dynamics becomes a viable way to investigate users. However, most current work only use global temporal trend and fail to distinguish such fine-grained patterns across groups. Different users have diverse interest and exhibit distinct behaviors, and temporal dynamics tend to be different. This paper proposes GrosToT (Group Specific Topics-over-Time), a unified probabilistic model to infer latent user groups and temporal topics at the same time. It models group-specific temporal topic variation from social content. By leveraging the comprehensive group-specific temporal patterns, GrosToT significantly outperforms state-of-the-art dynamics modeling methods. Our proposed approach shows advantage not only in temporal dynamics but also group content modeling. The dynamics over different groups vary, reflecting the groups' intention. GrosToT uncovers the interplay between group interest and temporal dynamics. Specifically, groups' attention to their medium-interested topics are event-driven, showing rich bursts; while its engagement in group's dominating topics are interest-driven, remaining stable over time.

Classifying Temporal Topics with Similar Patterns on Twitter

Twitter is a popular microblogging service that enables the users to send and read short text messages. These messages are becoming source to analyze topic trends and identify relations among temporal topics. In this paper, we propose a method to classify the temporal topics on Twitter as a problem of grouping the similar patterns. To provide a starting point for a classification under the same topics, we identify the content word weighting scheme based on Latent Dirichlet Allocation (LDA). And we formulate how the temporal topics in the time window can be classified like peaky topics, constant topics, and periodic topics. We provide different real case studies which show the validity of the proposed method. Evaluations show that the proposed method is useful as a classifying model in the analysis of the temporal topics.

Extended Probabilistic Latent Semantic Analysis Model For Topics In Time-Stamped Images

Abstract This paper considers the problem of modelling the topics in a sequence of images with known time stamp. Detecting and tracking of temporal data is an important task in multiple applications, such as finding hot research point from scientific literature, news article series analysis, email surveillance, search query log mining, etc. In contrast to existing works mainly focusing on text document collections, this paper considers mining temporal topic trends from image data set. An extension of the Probabilistic Latent Semantic Analysis(PLSA) model, which includes an additional variable associated with the time stamp to better model the temporal topics, is presented to extract topics among images and tract how topics change over time. Experiments show the effectiveness of this method.

Unsupervised Segmentation Methods of TV Contents

We present a generic algorithm to address various temporal segmentation topics of audiovisual contents such as speaker diarization, shot, or program segmentation. Based on a GLR approach, involving the ΔBIC criterion, this algorithm requires the value of only a few parameters to produce segmentation results at a desired scale and on most typical low-level features used in the field of content-based indexing. Results obtained on various corpora are of the same quality level than the ones obtained by other dedicated and state-of-the-art methods.