Drug-drug interactions (DDIs) may bring huge health risks and dangerous effects to a patient’s body when taking two or more drugs at the same time or within a certain period of time. Therefore, the automatic extraction of unknown DDIs has great potential for the development of pharmaceutical agents and the safety of drug use. In this article, we propose a novel recurrent hybrid convolutional neural network (RHCNN) for DDI extraction from biomedical literature. In the embedding layer, the texts mentioning two entities are represented as a sequence of semantic embeddings and position embeddings. In particular, the complete semantic embedding is obtained by the information fusion between a word embedding and its contextual information which is learnt by recurrent structure. After that, the hybrid convolutional neural network is employed to learn the sentence-level features which consist of the local context features from consecutive words and the dependency features between separated words for DDI extraction. Lastly but most significantly, in order to make up for the defects of the traditional cross-entropy loss function when dealing with class imbalanced data, we apply an improved focal loss function to mitigate against this problem when using the DDIExtraction 2013 dataset. In our experiments, we achieve DDI automatic extraction with a micro F-score of 75.48% on the DDIExtraction 2013 dataset, outperforming the state-of-the-art approach by 2.49%.

Knowledge tracing is an important research topic in student modeling. The aim is to model a student’s knowledge state by mining a large number of exercise records. The dynamic key-value memory network (DKVMN) proposed for processing knowledge tracing tasks is considered to be superior to other methods. However, through our research, we have noticed that the DKVMN model ignores both the students’ behavior features collected by the intelligent tutoring system (ITS) and their learning abilities, which, together, can be used to help model a student’s knowledge state. We believe that a student’s learning ability always changes over time. Therefore, this article proposes a new exercise record representation method, which integrates the features of students’ behavior with those of the learning ability, thereby improving the performance of knowledge tracing. Our experiments show that the proposed method can improve the prediction results of DKVMN.

Sentiment analysis is an essential process which is important to many natural language applications. In this paper, we apply two models for Arabic sentiment analysis to the ASTD and ATDFS datasets, in both 2‐class and multiclass forms. Model MC1 is a 2‐layer CNN with global average pooling, followed by a dense layer. MC2 is a 2‐layer CNN with max pooling, followed by a BiGRU and a dense layer. On the difficult ASTD 4‐class task, we achieve 73.17%, compared to 65.58% reported by Attia et al., 2018. For the easier 2‐class task, we achieve 90.06% with MC1 compared to 85.58% reported by Kwaik et al., 2019. We carry out experiments on various data splits, to match those used by other researchers. We also pay close attention to Arabic preprocessing and include novel steps not reported in other works. In an ablation study, we investigate the effect of two steps in particular, the processing of emoticons and the use of a custom stoplist. On the 4‐class task, these can make a difference of up to 4.27% and 5.48%, respectively. On the 2‐class task, the maximum improvements are 2.95% and 3.87%.

Sentiment classification is a fundamental task in many natural language processing applications. Neural networks have achieved great successes on the sentiment classification task in recent years, since recurrent neural networks and long-short-term memory networks have the ability to deal with sequences of different lengths and to capture contextual semantic information. However, the effectiveness of these methods is limited when used to extract contextual information from relatively long texts. Therefore, in our model, we apply bidirectional gated recurrent units to capture contextual information as far as possible when learning word representations, which may effectively reduce the noise compared to other methods. We also propose a novel loss function namely drop loss (DL) which makes the model focus on the hard examples--examples which are easily classified incorrectly--in order to improve the accuracy of the model. We experiment on four commonly used datasets, and the results show that the proposed method has a good performance on four datasets, and needs fewer parameters compared with recent benchmarks, such as CoVe, ULMFiT, embeddings from language models, and bidirectional encoder representations from transformers. Furthermore, we demonstrate that the classification performance of existing shallow network models can be significantly improved by using DL. In particular, the accuracy of the CNN+LSTM model improves 9% on the IMDB-10 dataset.

In this article we present the Amharic Speech Emotion Dataset (ASED), which covers four dialects (Gojjam, Wollo, Shewa, and Gonder) and five different emotions (neutral, fearful, happy, sad, and angry). We believe it is the first Speech Emotion Recognition (SER) dataset for the Amharic language. Sixty-five volunteer participants, all native speakers of Amharic, recorded 2,474 sound samples, 2 to 4 seconds in length. Eight judges (two for each dialect) assigned emotions to the samples with high agreement level (Fleiss kappa = 0.8). The resulting dataset is freely available for download. Next, we developed a four-layer variant of the well-known VGG model, which we call VGGb. Three experiments were then carried out using VGGb for SER, using ASED. First, we investigated which features work best for Amharic, FilterBank, Mel Spectrogram, or Mel-frequency Cepstral Coefficient (MFCC). This was done by training three VGGb SER models on ASED, using FilterBank, Mel Spectrogram, and MFCC features, respectively. Four forms of training were tried, standard cross-validation and three variants based on sentences, dialects, and speaker groups. Thus, a sentence used for training would not be used for testing, and the same for a dialect and speaker group. MFCC features were superior under all four training schemes. MFCC was therefore adopted for Experiment 2, where VGGb and three well-known existing models were compared on ASED: RESNet50, AlexNet, and LSTM. VGGb was found to have very good accuracy (90.73%) as well as the fastest training time. In Experiment 3, the performance of VGGb was compared when trained on two existing SER datasets—RAVDESS (English) and EMO-DB (German)—as well as on ASED (Amharic). Results are comparable across these languages, with ASED being the highest. This suggests that VGGb can be successfully applied to other languages. We hope that ASED will encourage researchers to explore the Amharic language and to experiment with other models for Amharic SER.