Font Size: a A A

Research On Profit Mining In Business Intelligence

Posted on:2009-04-28Degree:DoctorType:Dissertation
Country:ChinaCandidate:X J XuFull Text:PDF
GTID:1119360245463359Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the incessant accumulation of business data, business intelligence is becoming increasingly important. Business intelligence is the process of commercial information collection, management and analysis so decision-makers at all levels can access to corporate of knowledge or insight to make more profitable decisions for their business. In general, business intelligence is composed of data warehousing, online analytical processing, data mining, data backup and restore, and so on. Business intelligence software is related to the realization of software, hardware, consulting services and applications, and its basic architecture, includes data warehousing, online analytical processing and data mining in three parts. It makes business intelligence becomes more and more important that how discovers knowledge from data, obtains the guidance to customers from knowledge, and thereby better serves our customers, expects to get higher profits.As a result, the research of profit mining in business intelligence is of great significance. The major contents could be summarized as follows:(1) This dissertation makes a general summary of research on profit mining for business intelligence, analyzes the derivation background and the course of development. After introducing and analyzing the development of profit mining, the necessary of profit mining tool is presented. Furthermore, the future of profit mining is also discussed in this dissertation. The basic theory and strategies of profit mining are also introduced and analyzed, which are the groundwork of further research works.(2) We analyze the item network by search engine algorithms so to get the solution of item selection problem. Generally speaking, They throw old items which are lost of profit and introduce new items which can bring higher profit. They segment items according to items' profit and ranking items by items' importance. The role of ranking becomes critical so that they can find a discrete subset of items, which can maximize earned profit. How to retain important items so as to find a meaning discrete subset to maximize profit of retailing is item selection problem.Important of items not only depends on some characters of items themselves such as their purpose, quality and price, but also the requirement when customers purchase them. However there are still some objective criterions which can evaluate relative important of items, for example, the connection of a mass of transactions. Here we call the important degree for retailers the authoritative value of items. So authority items means items with high authoritative value. Item selection problem need consider many factors carefully, but a primary element is the cross-selling between items, that means, the profit of an item is not only related itself but also related the selling of the other items. Some items maybe produce low profit but such items can promote to sell items with higher profit. Therefore we use the topology structure of historical transactions to model the relation between items consequently we can find items combination subset with high profit.In this article, we work within the PageRank framework. Our contributions are as follows. First FullRank algorithm is introduced to rank items based on weighted directed graph. Next we prove the convergence of FullRank algorithm and give the method to solve dangling items. Then three interpretations are proposed to explain FullRank algorithm and item authority value. Finally experimental results support our algorithm.(3) We present a new algorithm to solve customer-oriented catalog segmentation problem. Catalog segmentation problem is an important application of data mining in business based on microeconomic view. Basic catalog segmentation is expressed as follows: an enterprise hope to design k catalogs with each size r to send the corresponding interesting customers, so as to maximize the number of items purchased by customers, and to know the products of the enterprise for promotion. Customer-oriented catalog segmentation problem requires that a customer has at least t interesting products in his received catalog. This paper analyzes the problem's complexity and use TFP-Tree to store a customer database. We map a customer database to a tree TFP-Tree and design a new algorithm MaxCover.(4) Based on customer-oriented catalog segmentation problem, we add profit constraint and propose the new problem - double constraints catalog segmentation problem. For the new problem, we analyze its complexity and present an algorithm based on tree.(5) We overview credit scoring problem. Credit scoring problem is to classify credit applicants through analyzing personal information of credit applicants. We deal with the input vector by using link ranking algorithm to improve the classification accuracy and reduce the classification error. Experimental results show that the proposed algorithm can deal effectively with the input vector, and ultimately improve the classification accuracy, achieve the expected effect.(6) Similar with the general data mining software, profit mining software in business intelligence has mainly three aspects: data collection and preprocessing, the experimental results obtained using algorithms, and the actual business activities of the guidance through experimental results. In this paper we construct a professional profit mining tool: ProfitMiningTool. We give the data preprocessing to deal with the basic data, obtain association rule and then focus of the profit mining results.
Keywords/Search Tags:Intelligence
PDF Full Text Request
Related items