What is Pattern Analysis?
Pattern analysis is a data-driven approach to extracting useful patterns from multi-dimensional data with pre-defined templates. It is mainly applicable to data that are sequential in nature. The type of analysis includes Trend, Benchmark, etc. For pattern analysis in Trend, the analysis reports consecutive positive or negative change in a sequence of measures and their change counts. For pattern analysis in Benchmark, the analysis reports the maximum and minimum in a sequence of measures, and their distances from the last benchmark.
The following are two PA applications in action. Win Ratio Monitoring is a PA application for monitoring win/loss activities for Customer Insight and for Home Page in Customer OnDemand (Fig 1); Enterprise Sales Performance Analysis is a PA prototype for analyzing sales performance in the context of Enterprise Performance Management.
Win Ratio Monitoring
As with many data analysis tasks involving sequential information, Win Ratio Monitoring is such a task for examining ups and downs for win/loss activities. As an example, given the following win loss activities (Fig 2), the application automatically summarizes the Trend for the win/loss activities (Fig 3).
Enterprise Sales Performance Analysis
As with many multi-dimensional data analysis tasks, Enterprise Sales Performance is such a task that analyzes sales performance based on various dimensions, key figures and filters. As an example, given the following invoice list (Fig 4), the prototype automatically summarizes the Trend (Fig 5) if it is selected for analysis. Other analysis patterns include Count, Benchmark, and Comparison (Fig 6). A filter can be used to select different context (Fig 7).
Pattern Extraction (PE) is the technique behind Pattern Analysis. Like Information Extraction, PE relies on a set of pre-defined templates to extract useful patterns from data. Some of the common patterns are Trend and Benchmark.
Trend refers to the direction of change for a given measure, or numeric quantity, such as Bond prices. It can be rise, fall or unchanged, at a certain time. For example, “Bond prices rose on Monday”
Count refers to the number of changes for a given dimension, or categorical entity, such as time in terms of year, quarter, month, etc. For example, “The Dow Jones industrials resulted in four days of losses over the past five.” Here, the analysis counts the change in terms of days.
Benchmark refers to a change measured against a specific benchmark, or milestone, on a given measure, such as Home Sales. For example, “New home sales hit bottom in January”, or “Sales were down nearly 33 percent from May last year.”
Comparison refers to relative change for two or more members of a dimension, such as company. For example, “Apple shares added $3.60, or 2.7 percent, to $139.48, while rival smart phone manufacturer Palm Inc. jumped more than 6 percent, rising 87 cents to $13.93”