AAI_2025_Capstone_Chronicles_Combined

‭ResolveAI‬

‭●‬ ‭answer (text):‬ ‭Includes the agent’s response. Entries lacking an answer were dropped to‬ ‭ensure completeness.‬ ‭●‬ ‭type (categorical/string):‬ ‭Classifies the ticket into‬‭categories such as Incident, Request,‬ ‭Problem, or Change.‬ ‭●‬ ‭queue (categorical/string):‬ ‭Identifies the support‬‭department handling the ticket (e.g.,‬ ‭Billing, Technical Support, Customer Service, Product Support).‬ ‭●‬ ‭priority (categorical/string):‬ ‭Indicates the urgency‬‭level, typically labeled as Low,‬ ‭Medium, or High.‬ ‭●‬ ‭language (categorical/string):‬ ‭Specifies the language in which the ticket was submitted.‬ ‭●‬ ‭tag_1 to tag_8:‬ ‭Represent additional descriptors; these columns were consolidated into a‬ ‭single list-based format for streamlined analysis.‬ ‭The data cleaning and preprocessing begins with an initial review of the raw text fields‬ ‭like subject, body, and answer, to identify inconsistencies and noise such as irregular casing,‬ ‭extraneous punctuation, and special characters (including emojis and formatting artifacts). The‬ ‭text is then converted to lowercase to standardize the content, and unwanted characters are‬ ‭removed to ensure that the analysis focuses solely on meaningful words. Subsequently,‬ ‭tokenization is applied to break the text into individual words or tokens. As part of this process,‬ ‭common stopwords (words that typically do not add substantial meaning) are removed to reduce‬ ‭noise and improve the efficiency of the subsequent modeling.‬ ‭After these cleaning steps, the text is transformed into sequences of integers using a‬ ‭custom tokenizer, and the sequences are padded to a consistent maximum length to create a‬

‭6‬

54