||Temporal data mining techniques have been widely used to extract desirable time-related knowledge from existing databases. However, most of the existing studies only considered different lifespans of items in a set of transactions to find general temporal association rules. In reality, some products in a store may be put on shelf and taken off shelf multiple times, and some biases may exist in the temporal association rules discovered. Besides, it is common for a company to have a chain of retail sites in different locations, and it is thus very critical to flexibly obtain association rules from portions of data with the consideration of on-shelf situations of items in a multi-site environment for providing relevant online decision supports to users. In this thesis, we thus handle the problems of mining temporal association rules with the consideration of on-shelf situations of items in a multi-site environment. |
In the first part of the thesis, we introduce a new research issue named on-shelf association rule mining (OAR) for temporal association rule mining with the consideration of time periods of items. This problem is more difficult than traditional association-rule mining due to this issue without the downward-closure property. Hence, an effective three-phase mining approach is developed to find such rules in a temporal database. In the second part of the thesis, we introduce another new issue named online multi-site on-shelf association rule mining (MOAR) with the consideration of on-shelf locations and time periods of items. Meanwhile, an online mining approach is also developed for generation of online multi-site on-shelf association rules, and an effective strategy is designed to tighten the upper-bounds of supports for candidate itemsets by using the multi-site pattern relation information.
The experimental results on simulation datasets show the effectiveness of the two types of rules (OAR and MOAR) and the performance of the proposed mining approaches under different parameter settings.