您现在的位置:网站首页 >> 国际合作 >> 内容详细
基于社交网络的采购欺诈检测研究
作 者: 来 源: 合作办 发布时间: 2015-03-23 点击次数: 6589

基于社交网络的采购欺诈检测研究

 

Project Description:

  1. Project Title: Big Data and Social Network Analytics Based Procurement Fraud Detection
  2. Project Overview:

Procurement fraud creates significant loss and is difficult to detect. IBM T.J. Watson Research Center initiated Procurement Risk Analytics (PRA) project last year, and GPS Ningbo team is responsible to implement and support. This project employs various analytics from advanced machine learning to simple arithmetic queries, statistical tests, outlier detection methods, text analytics built using SPSS, Java and SQL.However we believe there are still significant analytics opportunities exist on the social network to limit exposure to loss.

 

The Procurement fraud Analytics on the social network SUR project proposes to include Zhejiang University to focus on research and implementation of procurement fraud analytics based on social network analytics .This project will apply multiple data mining technologies to assess closeness of connections, qualification of suppliers and identify collusion likeliness. To better make the project in IBM, a strong team from IBM Research team, China Development Lab (CDL), will work closely with universities

  1. Description:

Project Goals

  1. To propose a procurement fraud detection model based on big data and social network analysis;
  2. To develop a prototype system based on the procurement fraud detection model;
  3. To publish a paper and apply for a software copyright based on the model and prototype;
  4. To train several master students concentrating on big data development and application technologies.

    Project Definition

        The procurement of governments or large companies always involves with many people and purchasing links. If not effectively monitored, it is prone to fraud and corruption. The traditional way to strengthen supervision by human leads to huge administrative cost. With the rapid development of information technology, an effective way to discover fraud by analyzing the data produced during the procurement is considered. However, analyzing only the internal information related to procurement staff is not enough, lots of potential irregularity information is hidden in the external information (e.g. social network, forum etc.). But, the amount of external information is very big, traditional data storage and analysis techniques are unable to deal with it. This project plans to solve this problem with the big data technology developed by College of Software Technology, Zhejiang University. The internal and external information of the procurement staff, the information of the suppliers and commodities on the internet will all be collected, stored and analyzed to effectively detect and eliminate the potential fraudulent behavior. The prototype system will provide the basic interaction with users and the visualization of the detection results. The framework of the project is shown in the following figure.

 

 

The project will focus on the following aspects of the research and development:

  • The acquisition and storage of procurement big data: all the individual data objects (including purchaser, supplier and commodity) involved in the procurement process will be considered. We will first analyze the data availability of the data objects and then collect the information from all the aspects. The stored information will be collected from social network, forum, post bar, e-commerce website, internal e-mail system and other related structured and unstructured data.
  • Procurement fraud analysis and mining: a procurement fraud detection model will be established based on the above collected data. A set of data mining methods, such as machine learning, statistical analysis, text matching, social relations mining etc., will be employed to develop a procurement fraud detection prototype system, which can significantly improve the efficiency and accuracy of procurement fraud detection.

 

    Innovation

  • Propose a method for person profile construction by matching and integrating multi-source heterogeneous data. In the project, we will collect data from a wide range of sources, including popular social networks (such as weibo, LinkedIn, twitter, etc), as well as internal data (such as email, database, etc). The method developed in this project will take basic personal information, results of social network analysis and text similarity analysis into consideration comprehensively to determine whether the descriptions from different data sources are about the same person.
  • Propose a procurement fraud detection model based on social network. The model integrates information collected from both public social networks and private internal data of the company to build personal relation network. By applying methods of advanced machine learning to simple arithmetic queries, statistical tests, outlier detection methods, text analytics built using SPSS, Java and SQL, the potential improper behavior could be detected.

 

    Technical Value

Procurement fraud has a detrimental effect on both governmental agencies and companies. However, it is difficult to detect such kind of fraud due to reasons such as complexity of procurement and invisibility of fraud.

        In this project, we propose a procurement fraud detection model based on social networks. By collecting data from popular social networks and internal sources (such as email content, databases), the personal relationship model can be constructed. Then, we also develop methods including advanced machine learning methods to simple queries, which can help discover whether the relationship exists between specific targets and what kind the relationship is. The result could be useful to detect the potential procurement fraud behaviors.

        With the growing popularity of social networks, more and more people publish information. The increasing data volume will improve the accuracy of our model, which could help to make the procurement practices more transparent.

        Besides, the model and method proposed in this project can also be used for general person investigation and personal relationship analysis, and their future application in areas such as social, intelligence, business and finance is also promising.





0