A framework for representing navigational patterns as full temporal objects

Abstract

Navigational patterns have applications in several areas including: web personalization, recommendation, userprofiling and clustering, etc. Most existing works on navigational pattern-discovery give little consideration to the effects of time (or temporal trends) on navigational patterns. Some recent works have proposed frameworks for partial temporal representation of navigational patterns. This paper proposes a framework that models navigational patterns as full temporal objects that may be represented as time series. Such a representation allows a rich array of analysis techniques to be applied to the data. The proposed framework also enhances the understanding and interpretation of discovered patterns, and provides a rich environment for integrating the analysis of navigational patterns with data from the underlying organizational environments and other external factors. Such integrated analysis is very helpful in understanding navigational patterns (e.g., E-commerce sites may integrate the trend analysis of navigational patterns with other market data and economic indicators). To achieve full temporal representation, this paper proposes a navigational pattern-discovery technique that is not based on pre-defined thresholds. This is a shift from existing techniques that are driven by pre-defined thresholds that can only support partial temporal representation of navigational patterns.

Maintaining Knowledge-Bases of Navigational Patterns from Streams of Navigational Sequences

“…However, unlike most algorithms, our technique is not based on predefined support (or another threshold). We reported an algorithm for mining contiguous navigational patterns based on an adapted generalized suffix tree (GST) that does not require support thresholds earlier [8]. However, the adapted GST algorithm cannot be used efficiently with the window maintenance scheme discussed in Section 3.2.…”

“…Leaf nodes are handled differently than internal nodes in the adapted GST technique. (Udechukwu et al [8] provide a full discussion on the adapted GST algorithm.) There are two internal nodes in Figure 3, with labels "LQ" and "Q", respectively.…”

“…Adapted GST with leaf-node relevance for the navigational sequences "LQR", "LQR", "LQ" and "QR" with "$" appended to each sequence as a stop symbol [8] We utilize a memory-based DBMS Similarly, the navigational patterns from batch B2 are mined and added to the table. The batches are mined independently and added to the table until the window size is reached (in this case, until batch B4 is processed).…”