What Happens Next? Future Subevent Prediction Using Contextual Hierarchical LSTM

Hu, Linmei; Li, Juanzi; Nie, Liqiang; Liu, Yuanyuan; Shao, Chao

doi:10.1609/aaai.v31i1.11001

Cited by 33 publications

(6 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…When these datasets are utilized for forecasting, they are usually represented as time series approach [22] [23], with each data point having a timestamp. An alternative approach is script-learning, in which a model is asked to predict the association between a series of events and a "future" event after being given the sequence and the "future" event [24] [25]. They demand that questions and response options be translated into their format and that text data be converted into event triples, which lessens the expressiveness of natural text.…”

Section: Forecastingmentioning

confidence: 99%

A technique to forecast Pakistan’s news using deep hybrid learning model

Ihsan,

Khurshid,

Shoaib

et al. 2024

Int. j. inf. tecnol.

View full text Add to dashboard Cite

Forecasting future events is a challenging task that can have a signi cant impact on decision-making and policy-making. In this research, we focus on forecasting news related to Pakistan. Despite the importance of accurate predictions in this eld, there currently exists no dataset for forecasting Pakistani news, speci cally with regard to politics. In order to address this gap, our research aims to create the rst Pakistani news dataset for forecasting Pakistan news that is mostly related to the politics of Pakistan. This dataset was collected from various sources, including Pakistani news websites and social media platforms, as well as frequently asked questions about Pakistani politics. We develop a forecasting model using this dataset and evaluate the effectiveness of cutting-edge deep hybrid learning techniques incorporating neural networks, random forest, Word2vec, NLP, and Naive Bayes. To the best of our understanding, no research has been done on the application of a deep hybrid learning model-a blend of deep learning and machine learning-for news forecasting. The accuracy of the forecasting model is 97%. According to our ndings, the model's performance is adequate when compared to that of other forecasting models. Our research not only lls the gap in the current literature but also presents a new challenge for large language models and has the potential to bring signi cant practical advantages in the eld of forecasting.

show abstract

Section: Forecastingmentioning

confidence: 99%

A technique to forecast Pakistan’s news using deep hybrid learning model

Ihsan,

Khurshid,

Shoaib

et al. 2024

Int. j. inf. tecnol.

View full text Add to dashboard Cite

show abstract

“…These three works all model narrative chains, that is, event sequences in which a single entity (the protagonist) participates in every event. Hu et al (2017) also apply a RNN approach, applying a new hierarchical LSTM model in order predict events by generating discriptive word sequences.…”

Section: Related Workmentioning

confidence: 99%

“…As illustrated in Figure 1, due to the overlap of words, parameterized additive models (Granroth-Wilding and Clark 2016; Modi 2016) and RNN-based models (Pichotta and Mooney 2016;Hu et al 2017) are limited in their transformations. Additive models combine the words in these phrases by the passing the concatenation or addition of their word embeddings to a parameterized function (usually a feed forward neural network) that maps the summed vector into event embedding space.…”

Section: Introductionmentioning

confidence: 99%

Event Representations With Tensor-Based Compositions

Weber

Balasubramanian

Chambers

2018

AAAI

View full text Add to dashboard Cite

Robust and flexible event representations are important to many core areas in language understanding. Scripts were proposed early on as a way of representing sequences of events for such understanding, and has recently attracted renewed attention. However, obtaining effective representations for modeling script-like event sequences is challenging. It requires representations that can capture event-level and scenario-level semantics. We propose a new tensor-based composition method for creating event representations. The method captures more subtle semantic interactions between an event and its entities and yields representations that are effective at multiple event-related tasks. With the continuous representations, we also devise a simple schema generation method which produces better schemas compared to a prior discrete representation based method. Our analysis shows that the tensors capture distinct usages of a predicate even when there are only subtle differences in their surface realizations.

show abstract

“…1, these works can be categorized into three major groups, i.e., intraevent-based, interevent-based, scenario-based models. Speciically, built on some popular neural networks, intraevent-based approaches are designed to model multiplicative interactions among intraevent elements, e.g., skip-gram network [16,23] and tensor networks [46]. Interevent-based approaches explore a range of complex and diverse interevent relations between events to reine the generated representations, e.g., time-order-based [45], event-graph-based [26], event-segment-based [30] and discourse-relations-based [24] relations.…”

Section: Introductionmentioning

confidence: 99%

Sequence-aware Knowledge Distillation for a Lightweight Event Representation

Cai

Ling

Chen

2022

ACM Trans. Inf. Syst.

View full text Add to dashboard Cite

Event representation targets to model the event-reasoning process as a machine-readable format. Previous studies on event representation mostly concentrate on a sole modeling perspective and have not well investigated the scenario-level knowledge, which can cause information loss. To cope with this dilemma, we propose a unified fine-tuning architecture-based approach ( UniFA-S ) that integrates all levels of trainings, including the scenario-level knowledge. However, another challenge for existing models is the ever-increasing computation overheads, restricting the deployment ability on limited resources devices. Hence, in this paper, we aim to compress the cumbersome model UniFA-S into a lighter and easy-to-deploy one without much performance damage. To this end, we propose a sequence-aware knowledge distillation model (SaKD) that employs a dynamic self-distillation on the decouple-compress-couple framework for compressing UniFA-S , which cannot only realize the model compression but also retain the integrity of individual components. We also design two fitting strategies to address the less-supervised issue at the distillation stage. Comprehensive experiments on representation-and-inference ability-based tasks validate the effectiveness of SaKD. Compared to UniFA-S , SaKD realizes a more portable event representation model at the cost of only 1.0% performance drop in terms of accuracy or Spearman’s correlation that is far less than other knowledge distillation models.

show abstract

What Happens Next? Future Subevent Prediction Using Contextual Hierarchical LSTM

Cited by 33 publications

References 22 publications

A technique to forecast Pakistan’s news using deep hybrid learning model

A technique to forecast Pakistan’s news using deep hybrid learning model

Event Representations With Tensor-Based Compositions

Sequence-aware Knowledge Distillation for a Lightweight Event Representation

Contact Info

Product

Resources

About