Research on enterprise crisis prevention under the network era

■2012 JSPS Asian CORE Program, Nagoya University and VNU University of Economics and Business Research on enterprise crisis prevention under the network era Beijing Institute of Technology Chen Yan* ABSTRACT: Enterprises will be the impact of various factors in the development process Sometimes the unconventional factors lead to crisis This paper examines the impact of the network era to the enterprises crisis The research focus on using internet information filtering model to detect the crisis signals and then taking the necessary measures to prevent the outbreak of corporate crisis KEYWORDS : crisis management; network era; internet information filtering enterprises crisis cases causes-statistics Through a Introduction comparative study of the existing literature, contribute Enterprises in the development process are subject to to the comprehensiveness of the index system Then various factors, not always in steady state Sometimes the causes of crisis contain a lot of crisis signs which they meet crisis Great challenges are brought to have the internal origin of the crisis Using the crisis enterprise crisis management under the network ear, case causes analysis filter indicators can make sure that such as the information spread fast, the indicators can sensitively reflect the potential risks has many communication channels, covers widely and the source Crisis management theory research began in the 2003 is difficulty to predict Under network era, the crisis with the occurrence of SARS in China Searching the information disseminated on the internet lead to the cnki (China National Knowledge Infrastructure) enterprise crisis and furthermore to stride forward crisis database with "enterprise crisis" as a keyword, we find According the situation, the paper focus on a network 2,578 literatures from 2003 to 2011 Among them, information filter model to monitor the internet and try there are 427 literatures involving crisis warning to get crises omen in order to prevent from the crisis indicators In this study, by the literature statistics, the indicators are composed of two parts of the enterprise Building index system about crisis signals external indicators (contain 14 second index) and internal indicators (contain 15 second index) Building crisis signals index system is the basis of the The enterprise crisis causes are varied Sometimes the crisis information gathering work Indicators which contingent and special events or changes in the effectively reflect the severity and the characteristic of environment are likely to lead to a crisis The cause‟s the crisis contribute to effective, scientific complete the analysis, in theory, should contain all the reasons, but follow-up work in practical work, it is neither possible nor necessary to The index system comes from literature research and include all corporate crisis causes according to the * Beijing Institute of Technology, School of Management and Economics manpower, resources and funding constraints In order system is showed as a template In accordance with to filter the indicators, we collect 166 cases from 2001 user-defined templates, we check the inflow of to 2011 Due to the cause‟s frequency analysis, we information flow to judge if the information meets the delete some indicators and then build up the crisis needs of the users The main technique of information signals index system (see table 1) filtering is in three aspects: First, the user's information needs model (user profiles) and the express technique Table1 Crisis signals index system First Index of web pages, the second is matching technology, and the third is use techniques of relative feedback Second Index Information filtering flow chart is shown in Figure Changes of policy Initialization Industry factors User Profile Incoming … Media influence Force Majeure … User User Public opinion External Index Web Page Rumor Profile Accept Information Cultural conflict Changes of the related Learning Feedback enterprises Figure Information filtering Flow Counterfeit products hazards Spokesperson problem Internal Index Organization management 3.1 User profile Financial position The construction of user profile refers to the collection Product / service quality and description of a user or a group of users‟ the Personnel Flow long-term information needs User needs template is management layer the core of information filtering system, because the Enterprise Strategy main purpose of filtering is judging correlation Brand reputation between the data and the demand of users based on the Public relations problems user needs template We build an initial user profile with keywords and their weight The initial keywords set up based on the Network information filtrating system model indicators plus company name plus some degree words, An information filtering system is a system that for example “Coca-Cola product bad” The initial removes redundant or unwanted information from an weights are assigned using entropy method through the information 166 enterprise crisis cases stream using (semi)automated or computerized methods prior to presentation to a human user Information filtering is a process, from which 3.2 Representation of the web page people query the information meeting users‟ specific The properties of the web page that are used to form needs from the dynamic information data stream, the their vector representations are the key terms taken specific needs of the users in information filtering from their respective text The 1st step in the feature extraction process usually consists of a text transformation of following kind: two empirical observations regarding text First, the  Remove HTML mark-up tags like , more times a term appears in a document, the more and so on relevant it is to the topic of the documents Second, the  Recognize individual words more times the term occurs throughout all documents  Use a stop-list to eliminate unwanted words in the collection to be filtered, the more poorly it (Stopwords are frequent words that carry no discriminates between documents The weight is given information These are words like pronouns, as prepositions, conjunctions etc Stopwords are wik usually removed from the term set of the document)  Where N is the total number of documents in the Perform suffix removal to generate word stems Then page is represented as a vector space model For a web page Di , with to terms t j , its vector representation is Di (1) f ik * log( N / nk ) (ti1 ,, tin ) Both the profiles collection and the factor log( N / nk ) is an inverse collection frequency factor which is small for terms that are widely used in the total collection 3.3 Matching Technology and the input pages are represented as vectors in an Match the web pages and user profile according to n-dimensional space The total number d terms in the calculate the similarity of the page vector and the user document collection is n A profile or page can also be profile vector Formula as follows: M represented as a vector of its terms with weighs assigned based on the „importance‟‟ of the individual term within the specific profile or document and within a collection of similar documents If w j represents   Sim( D, Pi ) u ik Cos wk (2) k M M wk2 u ik2 k k   where D is a vector of the new page: Pi is the ith the weight of the jth term, then each document Di can be vector Di represented by the improved (wi1 ,, win ) category profile vector; θ is the angle of the page vector and the profile vector; M is the dimension of the eigenvector; wk is the weight of the kth dimension in A well-known approach for computing term weights is the TFDIF scheme The TFDIF scheme assigns weight wik to term t k in document Di in proportion to f ik , the number of occurrences of term t k in document Di , and in inverse proportion to nk , the the page vector; uik is the weight of the kth dimension in the ith profile vector Then compare with the similarity of the page vector and each category pofile vector The page is assigned to the category of the greatest similarity The page will be filtered by the system, if the class is one which we   want to filter and Sim( D, Pi ) number of documents in which t k occurs at least once |The assumptions behind TFDIF are based on threshold) 3.4 Feedback algorithms ( is filtering , For many systems, when the user accesses the system and are set to 8, 16, and 4, respectively for the first time, he is often asked to provide feedback The algorithm revises the weight of the template on a number of documents An alternative procedure feature according to the feedback information The which does not require interaction from the user is to formula is described as follows: let the user start with a very comprehensive profile In 3.5 Simulation experiment on information filtering the beginning he then will get many irrelevant pages, model but after a while the system will learn from his The research uses Utility which is suggested by TREC feedback and update the profile relative to his response Feedback from the user may be obtained in two ways: （Text Retrieval Conference）program to evaluate The the user can either explicitly evaluate pages or the formula as follow: system can observe some implicit evidence about the Utility user‟s interest in each page When it comes to explicit where R :the number of actual pages which are feedback, the user usually provides a score for the searched out/ total number of actual pages; N : the pages retrieved by the system number of not actual pages which are searched out/ There are many traditional feedback algorithms Standard rocchio algorithm is the most typical total number of not actual pages; R :the number of actual pages which are not searched out/ total number feedback algorithm in vector space model Learning is of actual pages; N : the number of not actual pages achieved by combining document vectors into a which are not searched out/ total number of not actual profile P ( p1 , p2 , pm ) An initial profile can be computed as the weighted difference between the average of the relevant training samples c , and the average of the irrelevant training samples P where the parameters c , and control the relative impact of positive and negative examples, respectively The lengths of the page vectors ( A * R ) ( B * N ) (C * R ) ( D * N ) (4) pages The user want R and N result N means the system making wrong judgment and R means the system missing some actual information A, B, C, D represent weights and A, D (A, B, C and D are set to 2, -1, -2 and 1, respectively.) By using C# platform, in windows2000 server environment, we download 600 web pages as training set The experiment is repeated times, each time use 300 pages (see table 2) After the first filtering, user gives feedback and corrects the user profile The filtering result and Utility are shown in table and are usually normalized before the computations Table2 Test Pages Typically, the profile resulting from the Rocchio algorithm is restricted to have non-negative weights, , B,C Number of Number of Number of not test pages actual pages actual pages No meaning that if the computed pk ‟s are negative, they are set to The initial profile may be used to fetch new 300 225 75 pages to the user, and the feedback from the user 300 225 75 (relevant or irrelevant) can be utilized to update the initial profile: P new P old c c (3) Norwegian Computing Center HE Jing,LIU Hai-yan,GONG Yun-zhan(2004) “Techniques of improving filtering profiles in content filtering”, Journal of China Table3 Experimental Results No System searched out institute of communications, Vol.25, No.3, pp.112-118 System not searched Luo Chang-ri (2009) “Research on Web Information out Filtering Based on DCM Algorithm” 2009 First International Workshop on Database Technology and Applications, pp Actual Not Actual Actual Not Actual pages pages pages pages 191 22 34 53 205 16 20 59 Table4 System Utility No R N R N 0.85 0.10 0.45 0.71 1.41 0.91 0.07 0.27 0.79 Utility According the result, the feedback improves the system performance and the filtering effect can be accepted Conclusion The crisis comes unexpectedly and leads to huge losses Under the network era, crisis information disseminates faster and wider, brings much difficulty to crisis management Using the internet information filtering model to monitor the crisis signals, the enterprise detects the crisis as soon as possible Taking precautionary measures can prevent the crisis ACKNOWLEDGMENTS This research was supported by the Grant-in-Aid for Asian CORE Program "Manufacturing and Environmental Management in East Asia" of Japan Society for the Promotion of Science (JSPS) REFERENCE Malone TW (1987) “Intelligent information sharing systems.”, Commun ACM, Vol.30, No.5, pp.390–402 Kjersti Aas(1997) “A survey on personalisted information filtering systems for the world wide web”, Report No 922, 206-209  ... result, the feedback improves the system performance and the filtering effect can be accepted Conclusion The crisis comes unexpectedly and leads to huge losses Under the network era, crisis information... Financial position The construction of user profile refers to the collection Product / service quality and description of a user or a group of users‟ the Personnel Flow long-term information needs User... information disseminates faster and wider, brings much difficulty to crisis management Using the internet information filtering model to monitor the crisis signals, the enterprise detects the crisis