K. Krishnakumari

Bio: K. Krishnakumari is an academic researcher from A. V. C. College of Engineering. The author has contributed to research in topics: Domain (software engineering) & Artificial neural network. The author has an hindex of 1, co-authored 2 publications receiving 7 citations.

Hyperparameter tuning in convolutional neural networks for domain adaptation in sentiment classification (HTCNN-DASC)

Abstract: In-domain adaptation (DA), the knowledge trained in one domain, is used to test an unknown domain. Existing approaches use limited efforts on DA in sentiment classification (SC) using neural networks. The challenging task here is the dissimilarity in the semantic behavior across domains. In this paper, convolutional neural networks (CNNs) learn the knowledge of a particular domain using Doc2Vec feature representation which provides good performance for DA in SC for the target domain. Our empirical analysis with one-layer CNN exhibits significant change in the accuracy by tuning the hyperparameters involved with the CNN. This paper derives into a suitable CNN architecture accompanying hyperparameters which favor DA between different domains. Our empirical analysis with multi-domain dataset demonstrates that with suitable hyperparameters, CNN works well for DASC problems. The comparative study shows that CNN with Doc2Vec model provides a strong capability of learning large data representation semantically with other state-of-the-art methods for the DASC.

An enhanced sentiment dictionary for domain adaptation with multi-domain dataset in Tamil language (ESD-DA)

Abstract: Mostly sentiment analysis employs dictionary approaches for recognizing the polarity of terms in a review. However, in sentiment analysis between different domains called domain adaptation (DA), the sentiment lexicon disappoints that leads to the feature mismatch problem. Now, many e-commerce sites try to process reviews in their native languages. In this paper, we propose an enhanced dictionary in our native language (Tamil) that aims at building contextual relationships among the terms of multi-domain datasets that tries to minimize the feature mismatch problem. The proposed dictionary employs both labeled and unlabeled data from the source domain and unlabeled data from the target domain. More precisely, the initial dictionary explores pointwise mutual information for calculating contextual weight then the final dictionary estimates the rank score based on the importance of terms among all the reviews. This work intends to classify reviews of multiple target domains in Tamil by using the unified dictionary with a large number of vocabularies. This extendible dictionary significantly improves the accuracy of DA with the other baseline methods and handles many words in multiple domains with ease.

Sentiment analysis with deep neural networks: comparative study and performance assessment

Abstract: The current decade has witnessed the remarkable developments in the field of artificial intelligence, and the revolution of deep learning has transformed the whole artificial intelligence industry. Eventually, deep learning techniques have become essential components of any model in today’s computational world. Nevertheless, deep learning techniques promise a high degree of automation with generalized rule extraction for both text and sentiment classification tasks. This article aims to provide an empirical study on various deep neural networks (DNN) used for sentiment classification and its applications. In the preliminary step, the research carries out a study on several contemporary DNN models and their underlying theories. Furthermore, the performances of different DNN models discussed in the literature are estimated through the experiments conducted over sentiment datasets. Following this study, the effect of fine-tuning various hyperparameters on each model’s performance is also examined. Towards a better comprehension of the empirical results, few simple techniques from data visualization have been employed. This empirical study ensures deep learning practitioners with insights into ways to adapt stable DNN techniques for many sentiment analysis tasks.

Sentiment Analysis of Customer Reviews of Food Delivery Services Using Deep Learning and Explainable Artificial Intelligence: Systematic Review

Abstract: During the COVID-19 crisis, customers’ preference in having food delivered to their doorstep instead of waiting in a restaurant has propelled the growth of food delivery services (FDSs). With all restaurants going online and bringing FDSs onboard, such as UberEATS, Menulog or Deliveroo, customer reviews on online platforms have become an important source of information about the company’s performance. FDS organisations aim to gather complaints from customer feedback and effectively use the data to determine the areas for improvement to enhance customer satisfaction. This work aimed to review machine learning (ML) and deep learning (DL) models and explainable artificial intelligence (XAI) methods to predict customer sentiments in the FDS domain. A literature review revealed the wide usage of lexicon-based and ML techniques for predicting sentiments through customer reviews in FDS. However, limited studies applying DL techniques were found due to the lack of the model interpretability and explainability of the decisions made. The key findings of this systematic review are as follows: 77% of the models are non-interpretable in nature, and organisations can argue for the explainability and trust in the system. DL models in other domains perform well in terms of accuracy but lack explainability, which can be achieved with XAI implementation. Future research should focus on implementing DL models for sentiment analysis in the FDS domain and incorporating XAI techniques to bring out the explainability of the models.

Fine-Tuning Textrank for Legal Document Summarization: A Bayesian Optimization Based Approach

Abstract: Automatic text summarization techniques have a very high applicability in the legal domain, due to the complex and lengthy nature of legal documents. Most of the classical text summarization algorithms, which are also used in the legal domain, have certain hyperparameters, which if optimized properly, can further improve these algorithms. The choices of these hyperparameters have a big effect on the performance of such algorithms, yet this step of hyperparameter tuning is often overlooked while applying these algorithms in practice. In this work, a Bayesian Optimization based approach is proposed to optimize one of the classical summarization algorithms, Textrank, over this space of choices, by optimizing a ROUGE score mixture based objective function. The process of fine tuning and further evaluation is performed with the help of a publicly available dataset. From the experimental evaluation, it has been observed that the hyperparameter tuned Textrank is able to outperform baseline one-hot vector based Textrank and word2vec based Textrank models, with respect to ROUGE-1, ROUGE-2 and ROUGE-L metrics. The experimental analysis suggests that if proper hyperparameter tuning is performed, even a simple algorithm like Textrank can also perform significantly in the legal document summarization task.