Automated Extraction and Classification of Slot Machine Requirements from Gaming Regulations

Abstract

Analyzing stakeholder needs and transforming them into requirements is an important early step in the systems engineering lifecycle [1]. In regulated industries, important technical requirements can be found in state and federal laws and regulations. Casino gaming is one such industry. This paper analyzes South Dakota and Nevada slot machine regulations and applies automated natural language processing to extract and analyze technical requirements derived from them. First, each parts of speech (POS) in the regulations is identified. From this, the important adjective and noun keywords and keyword combinations are extracted using the Rapid Automatic Keyword Extraction (RAKE) algorithm [2]. Next, slot machine requirements are extracted from the gaming laws, many of which lack a “shall” in them. To perform this, a 12-rule pattern matching algorithm that applies phrase substitutions and identifies leader-subordinate paragraph headings is applied to the slot machine gaming rules. This approach successfully extracts nearly all of the slot machine technical and operations requirements, though fails to separate compound requirements accounting for approximately 3% of the total. Then, after stemming and stopping the regulations, a Naïve Bayes model for identifying functional requirements is constructed from the South Dakota regulations and applied to the Nevada regulations. This model is able to predict the Nevada functional product requirements from amongst the full set of extracted requirements with 87.5% accuracy. Finally, using a modified version of the Dice similarity metric where the word counts are weighted by the term frequency-inverse document frequency (TF-TDF) scores, the South Dakota requirements most similar to each of the Nevada requirements is determined. The paired South Dakota and Nevada requirements are then assessed using systems engineering expertise for equivalency and relatedness. Using the geometric mean of sensitivity and specificity as a scoring metric, the pairing algorithm optimum performance is 96.1% accurate in identifying equivalent requirements between the two sets of regulations, and 82.0% accurate in identifying related requirements.

Predicting Verification Methods from Natural Language Requirements

“…NLP was used to identify nonfunctional requirements from natural language documents in [15]. A similar approach for slot machine development, the technical domain of this paper, is found in [16].…”

“…A 12-rule heuristic algorithm that had over 99% accuracy in extracting technical slot machine product requirements from Nevada and South Dakota gaming regulations is described in [16]. That ruleset was adopted for this study but extended to 14 rules so as to take into account phrasing variations across the different English-speaking countries (Canada, New Zealand, Australia and United States).…”