Machine learning methods for financial document analysis have been focusing mainly on the textual part. However, the numerical parts of these documents are also rich in information content. In order to further analyze the financial text, we should as
say the numeric information in depth. In light of this, the purpose of this research is to identify the linking between the target cashtag and the target numeral in financial tweets, which is more challenging than analyzing news and official documents. In this research, we developed a multi model fusion approach which integrates Bidirectional Encoder Representations from Transformers (BERT) and Convolutional Neural Network (CNN). We also encode dependency information behind text into the model to derive semantic latent features. The experimental results show that our model can achieve remarkable performance and outperform comparisons.
Modern approaches to Constituency Parsing are mono-lingual supervised approaches which require large amount of labelled data to be trained on, thus limiting their utility to only a handful of high-resource languages. To address this issue of data-spa
rsity for low-resource languages we propose Universal Recurrent Neural Network Grammars (UniRNNG) which is a multi-lingual variant of the popular Recurrent Neural Network Grammars (RNNG) model for constituency parsing. UniRNNG involves Cross-lingual Transfer Learning for Constituency Parsing task. The architecture of UniRNNG is inspired by Principle and Parameter theory proposed by Noam Chomsky. UniRNNG utilises the linguistic typology knowledge available as feature-values within WALS database, to generalize over multiple languages. Once trained on sufficiently diverse polyglot corpus UniRNNG can be applied to any natural language thus making it Language-agnostic constituency parser. Experiments reveal that our proposed UniRNNG outperform state-of-the-art baseline approaches for most of the target languages, for which these are tested.
This paper presents a global summarization method for live sport commentaries for which we have a human-written summary available. This method is based on a neural generative summarizer. The amount of data available for training is limited compared t
o corpora commonly used by neural summarizers. We propose to help the summarizer to learn from a limited amount of data by limiting the entropy of the input texts. This step is performed by a classification into categories derived by a detailed analysis of the human-written summaries. We show that the filtering helps the summarization system to overcome the lack of resources. However, several improving points have emerged from this preliminary study, that we discuss and plan to implement in future work.
This paper investigates how the ordering of tone relative to the segmental string influences the calculation of phonotactic probability. Trigram and recurrent neural network models were trained on syllable lexicons of four Asian syllable-tone languag
es (Mandarin, Thai, Vietnamese, and Cantonese) in which tone was treated as a segment occurring in different positions in the string. For trigram models, the optimal permutation interacted with language, while neural network models were relatively unaffected by tone position in all languages. In addition to providing a baseline for future evaluation, these results suggest that phonotactic probability is robust to choices of how tone is ordered with respect to other elements in the syllable.
This paper describes the model built for the SIGTYP 2021 Shared Task aimed at identifying 18 typologically different languages from speech recordings. Mel-frequency cepstral coefficients derived from audio files are transformed into spectrograms, whi
ch are then fed into a ResNet-50-based CNN architecture. The final model achieved validation and test accuracies of 0.73 and 0.53, respectively.
Explaining neural network models is important for increasing their trustworthiness in real-world applications. Most existing methods generate post-hoc explanations for neural network models by identifying individual feature attributions or detecting
interactions between adjacent features. However, for models with text pairs as inputs (e.g., paraphrase identification), existing methods are not sufficient to capture feature interactions between two texts and their simple extension of computing all word-pair interactions between two texts is computationally inefficient. In this work, we propose the Group Mask (GMASK) method to implicitly detect word correlations by grouping correlated words from the input text pair together and measure their contribution to the corresponding NLP tasks as a whole. The proposed method is evaluated with two different model architectures (decomposable attention model and BERT) across four datasets, including natural language inference and paraphrase identification tasks. Experiments show the effectiveness of GMASK in providing faithful explanations to these models.
In this paper, we define an abstract task called structural realization that generates words given a prefix of words and a partial representation of a parse tree. We also present a method for solving instances of this task using a Gated Graph Neural
Network (GGNN). We evaluate it with standard accuracy measures, as well as with respect to perplexity, in which its comparison to previous work on language modelling serves to quantify the information added to a lexical selection task by the presence of syntactic knowledge. That the addition of parse-tree-internal nodes to this neural model should improve the model, with respect both to accuracy and to more conventional measures such as perplexity, may seem unsurprising, but previous attempts have not met with nearly as much success. We have also learned that transverse links through the parse tree compromise the model's accuracy at generating adjectival and nominal parts of speech.
Siamese Neural Networks have been widely used to perform similarity classification in multi-class settings. Their architecture can be used to group the clinical trials belonging to the same drug-development pathway along the several clinical trial ph
ases. Here we present an approach for the unmet need of drug-development pathway reconstruction, based on an Enhanced hybrid Siamese-Deep Neural Network (EnSidNet). The proposed model demonstrates significant improvement above baselines in a 1-shot evaluation setting and in a classical similarity setting. EnSidNet can be an essential tool in a semi-supervised learning environment: by selecting clinical trials highly likely to belong to the same drug-development pathway it is possible to speed up the labelling process of human experts, allowing the check of a consistent volume of data, further used in the model's training dataset.
The deep learning algorithm has recently achieved a lot of success, especially in the field of computer vision. This research aims to describe the classification method applied to the dataset of multiple types of images (Synthetic Aperture Radar (SAR
) images and non-SAR images). In such a classification, transfer learning was used followed by fine-tuning methods. Besides, pre-trained architectures were used on the known image database ImageNet. The model VGG16 was indeed used as a feature extractor and a new classifier was trained based on extracted features.The input data mainly focused on the dataset consist of five classes including the SAR images class (houses) and the non-SAR images classes (Cats, Dogs, Horses, and Humans). The Convolutional Neural Network (CNN) has been chosen as a better option for the training process because it produces a high accuracy. The final accuracy has reached 91.18% in five different classes. The results are discussed in terms of the probability of accuracy for each class in the image classification in percentage. Cats class got 99.6 %, while houses class got 100 %.Other types of classes were with an average score of 90 % and above.
A reliable and continuous supply of electrical energy is necessary for the functioning
of today’s complex society. Because of the increasing consumption and the extension of
existing electrical transmission networks and these power systems are oper
ated closer and
closer to their limits accordingly the possibilities of overloading, equipment failures and
blackout are also increasing, furthermore, we have an additional obstacle which is that
electrical energy cannot be stored efficiently, so, electrical energy should be generated only
when it's needed.
Due to the fact that world is facing a lack of oil reserves and the difficulties related to
have alternative sources to generate electrical power, then, electrical load forecasting is
considered as a crucial factor in electrical power system either from economical or
technical point of view on both planning and operating levels.
This research introduces a short term electrical load forecasting system by using
artificial neural networks with a simulation in Matlab environment in addition to an
interface for the system and all that is depending on previous load data and weather
parameters in Tartous province.
