Improving Helpdesk Chatbot Performance Using Levenshtein Distance and N-Gram Similarity

The development of NLP-based chatbots faces challenges in understanding user queries and statements, especially when queries contain misspellings or use different synonyms. Previous research has implemented NLP algorithms such as Cosine Similarity to understand user query inputs. However, this algorithm has the weakness of only considering the presence of words in the text, not the frequency or weight of words. This study proposes a solution to overcome the weakness of Cosine Similarity by implementing the N-gram Similarity algorithm. This algorithm focuses on matching word or character sequences (n-grams) between two texts, thus paying more attention to semantic similarity. In addition, this study also proposes the application of the Levenshtein Distance algorithm to recognize and correct spelling errors in user query inputs. This algorithm measures the edit distance between two strings to find the word in the dictionary with the least edit distance. Testing of the Levenshtein Distance algorithm shows satisfactory results with an accuracy rate of 87.5%. This indicates that the algorithm works well in correcting word spelling errors. Overall performance tests indicate a chatbot capability improvement of up to 91.8%