ACM Transactions on Asian and Low-Resource Language Information Processing最新文献

Recent advances in Natural Language Processing (NLP) have improved the performance of the systems that perform tasks, such as Emotion Detection (ED), Information Retrieval, Translation, etc., in resource-rich languages like English and Chinese. But similar advancements have not been made in Malayalam due to the dearth of annotated datasets. Because of its rich morphology, free word order and agglutinative character, data preparation in Malayalam is highly challenging. In this paper, we employ traditional Machine Learning (ML) techniques such as support vector machines (SVM) and multilayer perceptrons (MLP), and recent deep learning methods such as Recurrent Neural Networks (RNN) and advanced transformer-based methodologies to train an emotion detection system. This work stands out since all the previous attempts to extract emotions from Malayalam text have relied on lexicons, which are inappropriate for handling large amounts of data. By tweaking the hyperparameters, we enhanced the transformer-based model known as MuRIL to obtain an accuracy of 79%, which is then compared with the only state-of-the-art (SOTA) model. We found that the proposed techniques surpass the SOTA system available for detecting emotions in Malayalam reported so far.

Domain adaptation proves to be an effective solution for addressing inadequate translation performance within specific domains. However, the straightforward approach of mixing data from multiple domains to obtain the multi-domain neural machine translation (NMT) model can give rise to the parameter interference between domains problem, resulting in a degradation of overall performance. To address this, we introduce a multi-domain adaptive NMT method aimed at learning domain specific sub-layer latent variable and employ the Gumbel-Softmax reparameterization technique to concurrently train both model parameters and domain specific sub-layer latent variable. This approach facilitates the learning of private domain-specific knowledge while sharing common domain-invariant knowledge, effectively mitigating the parameter interference problem. The experimental results show that our proposed method significantly improved by up to 7.68 and 3.71 BLEU compared with the baseline model in English-German and Chinese-English public multi-domain datasets, respectively.

Code-switching entails mixing multiple languages. It is an increasingly occurring phenomenon in social media texts. Usually, code-mixed texts are written in a single script, even though the languages involved have different scripts. Pre-trained multilingual models primarily utilize the data in the native script of the language. In existing studies, the code-switched texts are utilized as they are. However, using the native script for each language can generate better representations of the text owing to the pre-trained knowledge. Therefore, a cross-language-script knowledge sharing architecture utilizing the cross attention and alignment of the representations of text in individual language scripts was proposed in this study. Experimental results on two different datasets containing Nepali-English and Hindi-English code-switched texts, demonstrate the effectiveness of the proposed method. The interpretation of the model using model explainability technique illustrates the sharing of language-specific knowledge between language-specific representations.

Humour is a crucial aspect of human speech, and it is, therefore, imperative to create a system that can offer such detection. While data regarding humour in English speech is plentiful, the same cannot be said for a low-resource language like Hindi. Through this paper, we introduce two multimodal datasets for humour detection in the Hindi web series. The dataset was collected from over 500 minutes of conversations amongst the characters of the Hindi web series (Kota-Factory) and (Panchayat). Each dialogue is manually annotated as Humour or Non-Humour. Along with presenting a new Hindi language-based Humour detection dataset, we propose an improved framework for detecting humour in Hindi conversations. We start by preprocessing both datasets to obtain uniformity across the dialogues and datasets. The processed dialogues are then passed through the Skip-gram model for generating Hindi word embedding. The generated Hindi word embedding is then passed onto three convolutional neural network (CNN) architectures simultaneously, each having a different filter size for feature extraction. The extracted features are then passed through stacked Long Short-Term Memory (LSTM) layers for further processing and finally classifying the dialogues as Humour or Non-Humour. We conduct intensive experiments on both proposed Hindi datasets and evaluate several standard performance metrics. The performance of our proposed framework was also compared with several baselines and contemporary algorithms for Humour detection. The results demonstrate the effectiveness of our dataset to be used as a standard dataset for Humour detection in the Hindi web series. The proposed model yields an accuracy of 91.79 and 87.32 while an F1 score of 91.64 and 87.04 in percentage for the (Kota-Factory) and (Panchayat) datasets, respectively.

Newton's Second Law of Motion algorithm is crucial to interactive visual effects and interactive behavior in interface design. Designers can only utilize simple algorithm templates in interface design since they lack organized mathematical science, especially programming. Directly using Newton's Second Law of Motion algorithm introduces two interface design issues. First, the created picture has a simplistic impact, laborious interaction, too few interactive parts, and boring visual effects. Second, using this novel approach directly to interface design reduces creativity, originality, and cognitive inertia. This study suggests a Newton's Second Law-based algorithm modification. It provides a novel algorithm application idea and a design strategy based on algorithm change to enable new interface design. Algorithm design gives interface design a new viewpoint and improves content production. In the arithmetic process of Newton's second law of motion algorithm, the introduction of repulsive force, reset force, shape, color and other attributes of interactive objects, and the integration of other algorithms to transform its basic arithmetic logic, which is conducive to the improvement of the visual effect of interaction design. It also improves users' interaction experiences, sentiments, and desire to participate with design work.

Aspect-based Sentiment Analysis (ABSA), also known as fine-grained sentiment analysis, aims to predict the sentiment polarity of specific aspect words in the sentence. Some studies have explored the semantic correlation between words in sentences through attention-based methods. Other studies have learned syntactic knowledge by using graph convolution networks to introduce dependency relations. These methods have achieved satisfactory results in the ABSA tasks. However, due to the complexity of language, effectively capturing semantic and syntactic knowledge remains a challenging research question. Therefore, we propose an Adaptive Dual Graph Convolution Fusion Network (AD-GCFN) for aspect-based sentiment analysis. This model uses two graph convolution networks: one for the semantic layer to learn semantic correlations by an attention mechanism, and the other for the syntactic layer to learn syntactic structure by dependency parsing. To reduce the noise caused by the attention mechanism, we designed a module that dynamically updates the graph structure information for adaptively aggregating node information. To effectively fuse semantic and syntactic information, we propose a cross-fusion module that uses the double random similarity matrix to obtain the syntactic features in the semantic space and the semantic features in the syntactic space, respectively. Additionally, we employ two regularizers to further improve the ability to capture semantic correlations. The orthogonal regularizer encourages the semantic layer to learn word semantics without overlap, while the differential regularizer encourages the semantic and syntactic layers to learn different parts. Finally, the experimental results on three benchmark datasets show that the AD-GCFN model is superior to the contrast models in terms of accuracy and macro-F1.

Named Entity Recognition (NER) in low-resource settings aims to identify and categorize entities in a sentence with limited labeled data. Although prompt-based methods have succeeded in low-resource perspectives, challenges persist in effectively harnessing information and optimizing computational efficiency. In this work, we present a novel prompt-based method to enhance low-resource NER without exhaustive template tuning. First, we construct knowledge-enriched prompts by integrating representative entities and background information to provide informative supervision tailored to each entity type. Then, we introduce an efficient reverse generative framework inspired by QA, which avoids redundant computations. Finally, We reduce costs by generating entities from their types while retaining model reasoning capacity. Experiment results demonstrate that our method outperforms other baselines on three datasets under few-shot settings.

To improve the scalability of resources and ensure the effective sharing and utilization of online English resources, an online English resource integration algorithm based on high-dimensional mixed-attribute data mining is proposed. First, an integration structure based on high-dimensional mixed-attribute data mining is constructed. According to this structure, the characteristics of online English resources are extracted, and historical data mining is carried out in combination with the spatial distribution characteristics of resources. In this way, the spatial mapping function of features is established, and the optimal clustering center is designed according to the clustering and fusion structure of online English resources. At this node, the clustering and fusion of online English resources are carried out. According to the fusion results, the distribution structure model of online English resources is constructed, and the optimization research of the integration algorithm of online English resources is carried out. The experimental results show that the integration optimization efficiency of the proposed algorithm is 89%, and the packet loss rate is 0.19%. It has good integration performance, and can realize the integration of multi-channel and various forms of online English resources.

In recommender systems, providing reasonable explanations can enhance users’ comprehension of recommended results. Template-based explainable recommendation heavily relies on pre-defined templates, constraining the expressiveness of generated sentences and resulting in low-quality explanations. Recently, a novel approach was introduced, utilizing embedding representations of items and comments to address the issue of user IDs and item IDs not residing in the same semantic space as words, thus attributing linguistic meaning to IDs. However, these models often fail to fully exploit collaborative information within the data. In personalized recommendation and explanation processes, understanding the user’s emotional feedback and feature preferences is paramount. To address this, we propose a personalized explainable recommendation model based on self-attention collaboration. Initially, the model employs an attention network to amalgamate the user’s historical interaction feature preferences with their user ID information, while simultaneously integrating all feature information of the item with its item ID to enhance semantic ID representation. Subsequently, the model incorporates the user’s comment feature rhetoric and sentiment feedback to generate more personalized recommendation explanations utilizing a self-attention network. Experimental evaluations conducted on two datasets of varying scales demonstrate the superiority of our model over current state-of-the-art approaches, validating its effectiveness.

Due to the complexity of Chinese and the differences between Chinese and English, the application of Chinese text in the digital field has a certain complexity. Taking Chinese text in Open Relation Extraction (ORE) as the research object, the complexity of Chinese text is analyzed. An extraction system of word vectors based on construction grammar theory and Deep Learning (DL) is constructed to achieve smooth extraction of Chinese text. The work of this paper mainly includes the following aspects. To study the application of DL in the complexity analysis of Chinese text based on construction grammar, firstly, the connotation of construction grammar and its role in Chinese text analysis are explored. Secondly, from the perspective of the ORE of word vectors in language analysis, an ORE model based on word vectors is implemented. Moreover, an extraction method based on the distance of word vectors is proposed. The test results show that the F1 value of the proposed algorithm is 67% on the public WEB-500 and NYT-500 datasets, which is superior to other similar text extraction algorithms. When the recall rate is more than 30%, the accuracy of the proposed method is higher than several other latest language analysis systems. This indicates that the proposed Chinese text extraction system based on the DL algorithm and construction grammar theory has advantages in complexity analysis and can provide a new research idea for Chinese text analysis.