Fuzzy-based conversational recommender for data-intensive science gateway applications

Abstract

Neuro-scientists are increasingly relying on parallel and distributed computing resources for analysis and visualization of their neuron simulations. Although science gateways (SG) have democratized relevant high performance/throughput resources, users require expert knowledge and training to fully utilize the capabilities of a science gateway. In this thesis, we aim to investigate the socio-technical challenges and opportunities to use conversational agents that can be presented as chatbots in guided interfaces of next-generation SG.We explore a novel method to analyze user proficiency through a questionnaire design involving key cyberinfrastructure (CI)/SG stakeholders. Users interact with a context-aware chatbot that is embedded within SG to obtain simulation tools/resources to accomplish their goals. A questionnaire related to CI/SG and neuroscience domains is presented to the user and we use intuitionistic fuzzy logic to handle the fuzziness or vagueness in the user responses and create user profiles. We describe the use of a rule-based Mamdani Inference system which uses the user proficiency to provide guidance to users using an underlying recommender for cloud solution templates. The cloud solution template recommender suggests the best suitable cloud architecture to the users functional (RAM, CPU cores, storage) and nonfunctional (cost, performance, etc.) requirements. We explore the use of KNN algorithm to match user requirements with the catalog of templates to find most relevant templates based on user preferences. Evaluation results show that the questionnaire is consistent and reliable to capture the user proficiency in CI/SG and neuroscience domain. We simulate a series of queries from both expert and novice user and human expert annotation of the responses generated by the chatbot confirmed that our chatbot significantly helped in providing appropriate SG resources and tools based on user proficiency and thereby increasing the diffusion and adoption of CI/SG. We also show that our cloud solution recommender scheme improves the resource provisioning accuracy in a manufacturing science gateway application by up to 21% compared to the existing schemes.