Multilingual Clustering of Streaming News

Miranda, Sebastião Átila Fonseca; Znotiņš, Artūrs; Cohen, Shay B.; Bārzdiņš, Guntis

doi:10.48550/arxiv.1809.00540

Cited by 1 publication

(8 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recent approaches to TDT have explored both sparse and dense features. Miranda et al (2018) proposes an online clustering method that represents documents with TF-IDF features, and demonstrates high performance on a benchmark news article data set. Building on this work, Staykovski et al (2019) compares sparse TF-IDF features with dense Doc2Vec representations, showing a sizeable improvement on the standard data set according to the BCubed evaluation metric.…”

Section: Related Workmentioning

confidence: 99%

“…Some online approaches combine the time element implicitly by sorting documents in chronological order, dividing them with time slicing, and processing each slice (Allan et al 1998;Yang, Pierce, and Carbonell 1998;Dai, He, and Sun 2010;Hu et al 2017). Other work uses decay functions to extract sparse time features (Yang, Pierce, and Carbonell 1998;Brants, Chen, and Farahat 2003;Li et al 2005;He et al 2010;Ribeiro, Ferret, and Tannier 2017;Miranda et al 2018;Saravanakumar et al 2021). None of the previous work has used temporal embeddings to represent time for the TDT task.…”

Section: Related Workmentioning

confidence: 99%

“…We follow the previous work (Miranda et al 2018;Saravanakumar et al 2021) to adopt a variant of the streaming K-means algorithm (Figure 3) with a few key changes. This system consists of three main components: (1) a document encoder, (2) a document-cluster weighted similarity model, (3) a cluster creation model.…”

Section: Online Tdt Pipelinementioning

confidence: 99%

“…We use two data sets for our experiments, which we refer to as News2013 and TDT-1. News2013 refers to the English portion of the multilingual news data set produced in Miranda et al (2018), which in turn derives from Rupnik et al (2016). The News2013 data set contains a train set from 2013-12-18 to 2014-11-02, and a temporally disjoint test set from 2014-11-02 to 2015-08-25.…”

Section: Experiments Data Setsmentioning

confidence: 99%

“…The unit of time can be a crucial factor in the model's performance. Our work follows the previous work (Miranda et al 2018;Saravanakumar et al 2021) and uses 1 day as the time granularity for the News2013 data set. We also experimented with hourly, bidaily, weekly (7-day), and monthly (30-day) granularity on this data set (Table 4).…”

Section: Time Granularitymentioning

confidence: 99%

See 4 more Smart Citations

Topic Detection and Tracking with Time-Aware Document Embeddings

Jiang¹,

Beeferman²,

Weiquan³

et al. 2021

Preprint

View full text Add to dashboard Cite

The time at which a message is communicated is a vital piece of metadata in many real-world natural language processing tasks such as Topic Detection and Tracking (TDT). TDT systems aim to cluster a corpus of news articles by event, and in that context, stories that describe the same event are likely to have been written at around the same time. Prior work on time modeling for TDT takes this into account, but does not well capture how time interacts with the semantic nature of the event. For example, stories about a tropical storm are likely to be written within a short time interval, while stories about a movie release may appear over weeks or months. In our work, we design a neural method that fuses temporal and textual information into a single representation of news documents for event detection. We fine-tune these time-aware document embeddings with a triplet loss architecture, integrate the model into downstream TDT systems, and evaluate the systems on two benchmark TDT data sets in English. In the retrospective setting, we apply clustering algorithms to the time-aware embeddings and show substantial improvements over baselines on the News2013 data set. In the online streaming setting, we add our document encoder to an existing state-of-the-art TDT pipeline and demonstrate that it can benefit the overall performance. We conduct ablation studies on the time representation and fusion algorithm strategies, showing that our proposed model outperforms alternative strategies. Finally, we probe the model to examine how it handles recurring events more effectively than previous TDT systems.

show abstract

Section: Related Workmentioning

confidence: 99%