26th International Association for Management of Technology Conference - IAMOT 2017

The amount of available data has been increasing at a very high speed in the arising digital world. As ever, people wants to get the most possible profit of data. In this sense, in recent years, technologies of Business Intelligence (BI) have been receiving special attention because we need to convert data into useful information and new knowledge. Data Mining (DM) constitutes the core of any BI process. However, current tools for DM are too specialized to be directly used by decision-makers. Thus, the intervention of DM specialists results imperative. This human intermediation implies not only high costs, but also more time to make more accurate decisions based on knowledge discovered. One of the most important tasks in DM is the search of relationships between variables or attributes. Existing algorithms for this purpose search for all the relationships that meet user desired minimum support and confidence degrees. However, this technique fails in generating too many association rules as "interesting rules" when they are not really of interest. Rules are said to be interesting only because they fulfill the specified minimum support and confidence. Therefore, a process of filtering the association rules which are really-useful must be carry out by humans. The challenge addressed here is the popularization of the so-called BI approach to improve decision making, avoiding over-cost of specialized human intermediation. In this work, we introduce a user-friendly mechanism for querying and validating association rules. This mechanism is based on fuzziness and combination of different metrics of strength. Fuzzy sets theory provides a formal tool for handling vague terms of natural language. Fuzziness is often present in human communication, hence results very useful to provide intelligent interfaces that process it. The use of linguistic terms overcomes the difficulty to comprehend numerical responses given by an automated system. Nowadays, some database querying languages, as SQLf, are featured for fuzziness. We extend SQLf for allowing linguistic truth values as “very true”. We design a general-purpose interface for interactive data exploration. The system will answer not only with numerical values indicating the validity of the considered association rule, but also gives linguistic truth value of the proposition defined by the rule. Validation is achieved using fuzzy quantifiers and different metrics as the Lift, Leverage or the Confidence for quantifying the strength of the association rules. With the proposed mechanism, we have built a graphical user interface in a way that association rules might be easily validated by non-experts in DM, fuzzy sets based systems or database languages. The use of linguistic truth values will be a key for better comprehension of the results of an association rule validation. Really-useful rules will be more easily reached because measure of rule importance strength involves different metrics, not just support and confidence. Novel Innovation process in the digital world may take advantage of existing large databases using BI. Development of transformation strategies, management of production and new business models require decision makings that would benefit with association rules mining. Mechanisms as we propose will allow decision-makers to deal in an easier way with complexity of DM systems without the demand of a BI processional work skills.