Browsing by Author "Reshma, R."

Now showing 1 - 5 of 5

Alleviating data sparsity and cold start in recommender systems using social behaviour
(2016) Reshma, R.; Ambikesh, G.; Santhi Thilagam, P.
Recommender systems are used to find preferences of people or to predict the ratings with the help of information available from other users. The most widely used collaborative filtering recommender system by the e-commerce sites suffers from both the sparsity and cold-start problem due to insufficient data. Most of the existing systems consider only the ratings of the similar users and they do not give any preferences to the social behavior of users which shall aid the recommendations made to the user to a great extent. In this paper, instead of finding similarity from rating information, we propose a new approach which predicts the ratings of items by considering directed and transitive trust with timestamps and profile similarity from the social network along with the user-rated information. In cases where the trust and the rating details of users from the system is absent, we still make use of the social data of the users like the products liked by the user, user's social profile-education status, location etc.To make recommendation. Experimental analysis proves that our approach can improve the user recommendations at the extreme levels of sparsity in user-rating data. We also show that our approach works considerably well for cold-start users under the circumstances where collaborative filtering approach fails. � 2016 IEEE.
Alleviating data sparsity and cold start in recommender systems using social behaviour
(Institute of Electrical and Electronics Engineers Inc., 2016) Reshma, R.; Ambikesh, G.; Santhi Thilagam, P.S.
Recommender systems are used to find preferences of people or to predict the ratings with the help of information available from other users. The most widely used collaborative filtering recommender system by the e-commerce sites suffers from both the sparsity and cold-start problem due to insufficient data. Most of the existing systems consider only the ratings of the similar users and they do not give any preferences to the social behavior of users which shall aid the recommendations made to the user to a great extent. In this paper, instead of finding similarity from rating information, we propose a new approach which predicts the ratings of items by considering directed and transitive trust with timestamps and profile similarity from the social network along with the user-rated information. In cases where the trust and the rating details of users from the system is absent, we still make use of the social data of the users like the products liked by the user, user's social profile-education status, location etc.To make recommendation. Experimental analysis proves that our approach can improve the user recommendations at the extreme levels of sparsity in user-rating data. We also show that our approach works considerably well for cold-start users under the circumstances where collaborative filtering approach fails. Â© 2016 IEEE.
Benchmarking Shallow and Deep Neural Networks for Contextual Representation of Social Data
(Institute of Electrical and Electronics Engineers Inc., 2021) Reshma, R.; Kamath Sâ€¤, S.; Ananthanarayana, V.S.
Representing the underlying context in text data is a much-explored research domain, where language model construction for the sizeable unstructured corpus is the central premise. To date, several deep language embedding representation techniques have been put forth for context-aware modelling of text data, focusing on word, sentence and document-level representations for specific tasks. In this paper, we experiment with shallow and deep embedding representation techniques for social media text data to predict Atherosclerotic Heart Disease (AHD) mortality rate. We employed Word2Vec, Doc2Vec, and LSTM based embedding techniques for this experimentation and analyzed the performance on standard datasets. Experimental evaluation evidence suggests that Doc2Vec, a shallow network, outperforms deep neural networks by attaining a Pearson correlation value of 0.8199 for tuned hyper-parameters, exceeding Word2Vec and Bi-LSTM models by a margin of 60 per cent. Â© 2021 IEEE.
Efficient parameter tuning of neural foundation models for drug perspective prediction from unstructured socio-medical data
(Elsevier Ltd, 2023) Reshma, R.; Kamath S․, S.K.; V.s, A.
The phenomenal popularity of social media platforms over the past decade has accelerated the development of intelligent applications that leverage social media data for informed decision-making in diverse domains like finance, education, public policy and healthcare management practices. While understanding the colloquial language of users on social media remains a challenging problem, access to users’ medical perspectives that conversationally divulge healthcare-related experiences and insights can help reshape healthcare ecosystems like chronic disease management, pandemics, public health, pharmacovigilance and more. Most existing models are constrained to a particular dataset while neglecting model adaptability across data sources and domains. Model generalization across variable data sizes also has received very little research attention. Conventional foundation models can be fine-tuned by adding additional model heads or by appending contributing network layers, however, there has been very little focus on effective parameter calibration for adapting neural foundation models to a specific task. In this study, an Adaptive Learning mechanism for Socio-Medical data (AL4SM) built on generic foundation neural models with efficient parameter learning is proposed, to categorize users’ perspectives on prescription drug-related experiences and adapt to diverse socio-medical data sources of variable sizes. AL4SM aims to lighten the over-parameterized mechanisms adopted by existing foundational techniques by efficiently learning latent medical information based on optimized parameter calibration and weight reinitialization techniques. Comprehensive cross-domain and cross-data analyses are undertaken to explore specific user perspectives related to prescription effectiveness and side effects. Validation experiments conducted on standard datasets obtained from Drugs.com and Druglib.com revealed that the proposed AL4SM outperformed state-of-the-art models, achieving an improvement of 6.06% in accuracy and 7.62% in F1-score for 3-class and 2% in F1-score for 10-class drug perspective categorization. The cross-data experiments further emphasized the superiority of the proposed model, with improved accuracy of 17% on Drugs.com and 9% on Druglib.com datasets, respectively. © 2023 Elsevier Ltd
Population-centric Profiling with Social Data for Large-scale Epidemiological studies
(Association for Computing Machinery, 2022) Reshma, R.; Kamath Sâ€¤, S.; Ananthanarayana, V.S.
Large-scale social data has the potential for enabling insightful epidemiological studies due to the availability of streaming user-generated data. Research has focused on harnessing the predictive power of deep neural models for applications like a region's health status modelling and behavioural profiling of the population using the location-specific user data for identifying health determinants for designing effective population health management policies and prevention measures. Previous studies have often overlooked the need for semantic representation for extracting behavioural insights from social data. This work presents an approach that attempts to leverage contextual representation models for identifying latent population-centric health determinants. Language variants like Character-level, Word-level and Document-level are experimented with to capture the opinion expressed by the users towards this objective. Experimental results revealed that the Document-level approach achieved the most decisive correlation score of 0.7297, which is nearly 60% more than the score obtained by Character-level and Word-level models. The proposed method also outperformed state-of-the-art works by 30%. Â© 2022 Owner/Author.