Enhancing convolution-based sentiment extractor via dubbed N-gram embedding-related drug vocabulary

Abstract

Everyday patients’ narratives on social media can reveal crucial public health issues. Mining those online narratives, which remained so far unconsidered, may mirror further hidden patient health status. Deep learning-based sentiment analysis (SA) approaches broadly focus on grammar directions such as semantic direction or only center on extract sentiment words. They provide both richer representation capabilities and better performance but do not consider the related medication concepts. As a result, the inaccurate recognition of related drug entities may seriously fail to retrieve the relevant sentiment expressed, leading to a lower recall than desired. Thus, the frequent use of informal medical language, non-standard format, wrongly spelled, and abbreviation forms, as well as typos in social media messages, has to be taken into consideration. In other words, the core of efficiently quantifying the sentimental aspects for related medication texts hardly involves a degree of medical language comprehension. In this paper, we seek to improve the importance of considering related drug entities that keep appearing in new Unicode Versions, ranging from drugs’ names, disease symptoms, drug misuse to potentially adverse effects. We propose N-gram-based convolution vocabulary scheme, which is dedicated mainly to featuring text under medical setting and clarifying related sentiment at the same level. This vectorization results in highly sentiment extraction, which produces medical concept normalization under distributed dependency. This architecture’s layers are a shared neural network between the medical featuring channel and the bidirectional sentiment information detector channel. Fewer approaches are proposed in this matter, we evaluate the effectiveness and transferability of this study across five benchmarking datasets and various online medication-related posts (Twitter posts, Parkinson’s disease forum’s discussions), which were significantly better than all other baselines.