Building a Business Intelligence System for AUB (second edition)
Authors
Ly, Trien Huy ; Mogensen, Jacob ; Skouboe, Kristian Reesen
Term
4. term
Education
Publication year
2004
Abstract
Rapporten beskriver udviklingen af et business intelligence (BI) system til AUB. Kernen er et datalager, en central database der samler data til analyse. Systemet understøtter tre hovedopgaver: at finde associationsregler (mønstre som 'brugere, der låner X, låner også Y'), at lave statistikker og at give boganbefalinger. Vi kortlægger først AUB's behov og designer datalageret derefter. Til mønstergenkendelse implementerer vi vores egen algoritme, LIQ, og laver en hybrid af LIQ og den kendte Apriori-metode for at kombinere deres styrker og øge hastigheden. Til anbefalinger bygger vi en tjeneste, der kombinerer en item-baseret tilgang (finde lignende bøger) med en indholdsbaseret tilgang (bruge bøgernes beskrivelser) for at forbedre kvaliteten. Vi beskriver implementeringerne, viser resultater fra ydelsestests og stiller et webværktøj til rådighed, som giver adgang til statistik, associationsregler og anbefalinger.
This report presents the development of a business intelligence (BI) system for AUB. At its core is a data warehouse, a central database that gathers data for analysis. The system supports three main tasks: discovering association rules (patterns such as 'users who borrow X also borrow Y'), producing statistics, and generating book recommendations. We first identify AUB's needs and design the data warehouse accordingly. For pattern discovery, we implement our own algorithm, LIQ, and create a hybrid of LIQ and the well-known Apriori method to combine their strengths and improve speed. For recommendations, we build a service that combines an item-based approach (finding similar books) with a content-based approach (using book descriptions) to improve recommendation quality. We describe the implementations, report performance tests, and provide a web tool that gives access to the statistics, association rule mining, and recommendation features.
[This abstract was generated with the help of AI]
Documents