Open-answer question with regular expression templates and string completion hinting

Sychev, Oleg

doi:10.1016/j.simpa.2023.100539

Cited by 2 publications

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Knowledge is embedded in the system to identify sentences and determine appropriate answers during the question-and-answer process. The chatbot utilizes various string-matching methods, such as Bruteforce [1], [2], Knuth Morris Pratt (KMP) [3], [4], Boyer Moore (BM) [5], [6], and regular expression [7] to perform word searches and obtain correct and valid answers.…”

Section: Introductionmentioning

confidence: 99%

Generate fuzzy string-matching to build self attention on Indonesian medical-chatbot

Suwarningsih,

Nuryani

2024

IJECE

View full text Add to dashboard Cite

Chatbot is a form of interactive conversation that requires quick and precise answers. The process of identifying answers to users’ questions involves string matching and handling incorrect spelling. Therefore, a system that can independently predict and correct letters is highly necessary. The approach used to address this issue is to enhance the fuzzy string-matching method by incorporating several features for self-attention. The combination of fuzzy string-matching methods employed includes Jaro Winkler distance + Levenshtein Damerau distance and Damerau Levenshtein + Rabin Carp. The reason for using this combination is their ability not only to match strings but also to correct word typing errors. This research contributes by developing a self-attention mechanism through a modified fuzzy string-matching model with enhanced word feature structures. The goal is to utilize this self-attention mechanism in constructing the Indonesian medical bidirectional encoder representations from transformers (IM-BERT). This will serve as a foundation for additional features to provide accurate answers in the Indonesian medical question and answer system, achieving an exact match of 85.7% and an F1-score of 87.6%.

show abstract