OAR@UM Collection:
https://www.um.edu.mt/library/oar/handle/123456789/8369
2024-03-28T08:42:22ZMorphological analysis for the Maltese language : the challenges of a hybrid system
https://www.um.edu.mt/library/oar/handle/123456789/119978
Title: Morphological analysis for the Maltese language : the challenges of a hybrid system
Authors: Borg, Claudia; Gatt, Albert
Abstract: Maltese is a morphologically rich language with a hybrid morphological system which features both concatenative and non-concatenative processes. This paper analyses the impact of this hybridity on the performance of machine learning techniques for morphological labelling and clustering. In particular, we analyse a dataset of morphologically related word clusters to evaluate the difference in results for concatenative and non-concatenative clusters. We also describe research carried out in morphological labelling, with a particular focus on the verb category. Two evaluations were carried out, one using an unseen dataset, and another one using a gold standard dataset which was manually labelled. The gold standard dataset was split into concatenative and non-concatenative to analyse the difference in results between the two morphological systems.2017-01-01T00:00:00ZCross-lingual transfer from related languages : treating low-resource Maltese as multilingual code-switching
https://www.um.edu.mt/library/oar/handle/123456789/119975
Title: Cross-lingual transfer from related languages : treating low-resource Maltese as multilingual code-switching
Authors: Micallef, Kurt; Habash, Nizar; Borg, Claudia; Eryani, Fadhl; Bouamor, Houda
Abstract: Although multilingual language models exhibit impressive cross-lingual transfer capabilities on unseen languages, the performance on downstream tasks is impacted when there is a script disparity with the languages used in the multilingual model's pre-training data. Using transliteration offers a straightforward yet effective means to align the script of a resource-rich language with a target language, thereby enhancing cross-lingual transfer capabilities. However, for mixed languages, this approach is suboptimal, since only a subset of the language benefits from the cross-lingual transfer while the remainder is impeded. In this work, we focus on Maltese, a Semitic language, with substantial influences from Arabic, Italian, and English, and notably written in Latin script. We present a novel dataset annotated with word-level etymology. We use this dataset to train a classifier that enables us to make informed decisions regarding the appropriate processing of each token in the Maltese language. We contrast indiscriminate transliteration or translation to mixing processing pipelines that only transliterate words of Arabic origin, thereby resulting in text with a mixture of scripts. We fine-tune the processed data on four downstream tasks and show that conditional transliteration based on word etymology yields the best results, surpassing fine-tuning with raw Maltese or Maltese processed with non-selective pipelines.2024-03-01T00:00:00ZEnhancing stock price prediction models by using concept drift detectors
https://www.um.edu.mt/library/oar/handle/123456789/119818
Title: Enhancing stock price prediction models by using concept drift detectors
Authors: Sammut, Charlton; Abela, Charlie; Vella, Vince
Abstract: Stock price movement prediction is faced with the problem that the distribution of certain underlying variables change over time. This phenomenon is defined as concept drift. Due to this phenomenon, stock price prediction models tend to give less accurate results, since the data distribution that the model has been trained on is no longer in-line with the current data distribution. In this paper an Adversarial Attentive Long Short-Term Memory (Adv-ALSTM) model is used together with a Hoeffding’s inequality based Drift Detection Method with moving Average-test (HDDMA) concept drift detector in order to make price movement predictions on 50 different stocks. Every time the HDDMA concept drift detector detects a concept drift, the model undergoes one of four possible retraining methods. The conducted experiments highlight the effectiveness of each of the proposed retraining methods, as well as how each of the methods mitigate the negative effects of concept drift in different ways. The best observed results were a 2.5% increase in accuracy and a 135.38% increase in Matthews Correlation Coefficient (MCC) when compared to the vanilla Adv-ALSTM model. These results validate the effectiveness of the proposed retraining methods, when applied to a model that has been trained on a financial dataset.2022-11-01T00:00:00ZDeep reinforcement learning of autonomous control actions to improve bus-service regularity
https://www.um.edu.mt/library/oar/handle/123456789/117943
Title: Deep reinforcement learning of autonomous control actions to improve bus-service regularity
Authors: Bajada, Josef; Grech, Joseph; Bajada, Therese
Abstract: Bus Bunching is caused by irregularities in demand across the bus route, together with other factors such as traffic. The effect of this problem is that buses operating on the same route start to catch up with each other, severely impacting the regularity and the quality of the service. Control actions such as Bus Holding and Stop Skipping can be used to regulate the service and adjust the headway between two buses. Traditionally, this phenomenon is mitigated either reactively online through simple rule-based control, or preemptively through analytical scheduling solutions, such as mathematical optimization. Over time, both approaches degrade to an irregular service. In this work, we investigate the use of Deep Reinforcement Learning algorithms to train a policy that determines which actions should take place at specific control points to regularise the bus service. While prior studies are typically restricted to one control action, we consider both Bus Holding and Stop Skipping. We replicate benchmarks found in the latest literature, and also introduce traffic to increase the realism of the simulation. Furthermore, we also consider scenarios where the service is already unstable and buses are already bunched together, a first of this kind of study. We compare the performance of the RL-based policies with a no-control policy and a rule-based policy. The learnt policies not only keep a significantly lower headway variance and mean waiting time, but also recover from unstable scenarios and restore service regularity.2023-01-01T00:00:00Z