Semantics and Usage Statistics for Multi-dimensional Query Expansion

Abstract

Abstract. As the amount and complexity of data keeps increasing in data warehouses, their exploration for analytical purposes may be hindered. Recommender systems have grown very popular on the Web with sites like Amazon, Netflix, etc. These systems proved successful to help users explore available content related to what they are currently looking at. Recent systems consider the use of recommendation techniques to suggest data warehouse queries and help an analyst pursue its exploration. In this paper, we present a personalized query expansion component which suggests measures and dimensions to iteratively build consistent queries over a data warehouse. Our approach leverages (a) semantics defined in multi-dimensional domain models, (b) collaborative usage statistics derived from existing repositories of Business Intelligence documents like dashboards and reports and (c) preferences defined in a user profile. We finally present results obtained with a prototype implementation of an interactive query designer.

A-BI+: A framework for Augmented Business Intelligence

“…Over the years, scholars have highlighted the importance of exploiting contextual information to provide focused recommendations with the nature of contexts being quite heterogeneous (a summarized description in provided in Table 1), for instance space and time [27], query logs [18,19], statistics on results [23,24] or databases [21], user interests [16], and social data [20]. Given such heterogeneity, other contributions (e.g., [30,31,20]) address the integration of contextual data to provide a common ground (e.g., a global schema [31] or an application programming interface [20]) enabling recommendation from…”

“…MD RT C D [16] OLAP query User profile [17] OLAP session OLAP session log [18] SQL query Query logs [19] OLAP query Dashboards, reports [20] SPARQL query Web documents [21] SQL query Database statistics [22] SQL query Result feedback [23] SQL query Result statistics [24] SQL query Result statistics [25] Web query Clicks, query log [26] Web query Clicks, query log [27] Web query Location [28] Web query Query logs [29] OLAP query Query logs A-BI + none Physical env., log multiple data sources. The previous context types have been widely adopted in several applications where the recommendation process is activated by an explicit user-defined input statement (e.g., query or keywords).…”

“…The previous context types have been widely adopted in several applications where the recommendation process is activated by an explicit user-defined input statement (e.g., query or keywords). Examples of applications are web query categorization [28,17], recommendation [16,25], and diversification [26]; query completion [18,19]; localized web keywords suggestion [27]; and interactive exploration of databases [22]. The main differences between A-BI + and these works are: (1) the multidimensional nature of the data handled and returned, (2) the nature of the context as well as the type of user/system interaction that triggers the recommendation, and the presence of (3) real-time, (4) cardinality and (5) diversification constraints.…”

BINLI: An Ontology-Based Natural Language Interface for Multidimensional Data Analysis

“…After NLP techniques and the application of some rules, the system seeks a mapping between core terms in user question and OLAP schema objects. This is achieved by semantic similarity computation and semantic reasoning [15,16]. In the end of this phase, if a word or an expression admits several meanings, all these semantic paths are considered, resulting in a list of several possible interpretations for the question.…”