Abstract:
"Event planning and outing applications play a crucial role in helping individuals organize and
enjoy various activities. However, the current landscape of these applications reveals
significant gaps in providing relevant information, including medical and travel-related data.
Despite extensive research on the importance of weather and infectious diseases during travel, there is a scarcity of pertinent information available during the activity planning process. The absence of medical information in event planning apps poses risks, particularly as seen by the effects of COVID-19. Moreover, despite advancements in Natural Language Processing (NLP) applications, there remains a limited presence of travel applications within Sri Lankan domains that fully exploit NLP functionalities to create recommendation algorithms facilitating dynamic user requests without requiring a return to the home page.
To address these challenges, this work suggests developing an outing application that harnesses NLP techniques to deliver optimal recommendations and relevant travel information. The user's prompt undergoes keyword extraction through Named Entity Recognition (NER) and semantic analysis to comprehend activity preferences. These preferences are then mapped using Large Language Models (LLMs) employing zero-shot classification techniques. Recommendations are provided using the TOPSIS (Technique for Order of Preference by Similarity to Ideal Solution) model, while existing locations are sourced through the Google Place API.
The base model utilized for zero-shot classification is the BART-Large model trained on the
MultiNLI dataset, achieving an accuracy of 0.6832 on the dataset curated for this study.
Leveraging this LLM, alongside advanced NLP techniques and recommendation algorithms,
the paper introduces a robust system aimed at offering personalized activity recommendations to users."
