#### DMCA

## Mining the Network Value of Customers (2002)

### Cached

### Download Links

- [www.cs.washington.edu]
- [www.cs.washington.edu]
- [research.microsoft.com]
- DBLP

### Other Repositories/Bibliography

Venue: | In Proceedings of the Seventh International Conference on Knowledge Discovery and Data Mining |

Citations: | 568 - 11 self |

### Citations

5126 | Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images
- Geman, Geman
- 1984
(Show Context)
Citation Context ...er of unknown neighbors of X i . If this number is small (e.g., 5), this should not be a problem; otherwise, an approximate solution is necessary. A standard method for this purpose is Gibbs sampling =-=[16]-=-. An alternative based on an ecient k-shortest-path algorithm is proposed in Chakrabarti et al. [6]. Given N i and Y, X i should be independent of the marketing actions for other customers. Assuming a... |

4673 | The Anatomy of a Large-Scale Hypertextual Web Search Engine
- Brin, Page
- 1998
(Show Context)
Citation Context ... sociology and statistics has suered from a lack of data and focused almost exclusively on very small networks, typically in the low tens of individuals [41]. Interestingly, the Google search engine [=-=4]-=- and Kleinberg's (1998) HITS algorithm forsnding hubs and authorities on the Web are based on social network ideas. The success of these approaches, and the discovery of widespread network topologies ... |

3632 | Authoritative sources in a hyperlinked environment - Kleinberg - 1998 |

3613 |
Social Network Analysis: Methods and Applications
- Wasserman, Faust
- 1994
(Show Context)
Citation Context ... study for some time, but previous work within sociology and statistics has suered from a lack of data and focused almost exclusively on very small networks, typically in the low tens of individuals [=-=41]-=-. Interestingly, the Google search engine [4] and Kleinberg's (1998) HITS algorithm forsnding hubs and authorities on the Web are based on social network ideas. The success of these approaches, and th... |

1634 |
Spatial interaction and statistical analysis of lattice systems
- Besag
- 1974
(Show Context)
Citation Context ... ) P (X i jN i ; Y;M) Y X j 2N u i P (X j jX k ; Y;M) (2) The set of variables X u , with joint probability conditioned on X k , Y and M described by Equation 2, is an instance of a Markov randomseld =-=[2, 25, 7]-=-. Because Equation 2 expresses the probabilities P (X i jX k ; Y;M) as a function of themselves, it can be applied iteratively tosnd them, starting from a suitable initial assignment. This procedure i... |

1546 | Grouplens: An open architecture for collaborative filtering of netnews
- Resnick, Iacovou, et al.
- 1994
(Show Context)
Citation Context ...ple, Breese et al. [3]). The most widely used method, and the one that we will assume here, is the one proposed in GroupLens, the project that originally introduced quantitative collaborativesltering =-=[35]-=-. The basic idea in this method is to predict a user's rating of an item as a weighted average of the ratings given by similar users, and then recommend items with high predicted ratings. The similari... |

1495 | Empirical analysis of predictive algorithms for collaborative filtering
- BREESE, HECKERMAN, et al.
- 1998
(Show Context)
Citation Context ...tars to the book, depending on how much she liked it). Many algorithms have been proposed for choosing which items to recommend given the incomplete matrix of ratings (see, for example, Breese et al. =-=[3]-=-). The most widely used method, and the one that we will assume here, is the one proposed in GroupLens, the project that originally introduced quantitative collaborativesltering [35]. The basic idea i... |

1288 |
The Small World Problem
- Milgram
- 1967
(Show Context)
Citation Context ...et al. [28], Barabasi et al. [1]). Some of this work might be applicable in our context. In retrospect, the earliest sign of the potential of viral marketing was perhaps the classic paper by Milgram [=-=31] esti-=-mating that every person in the world is only six edges away from every other, if an edge between i and j means \i knows j." Schwartz and Wood [37] mined social relationships from email logs. The... |

776 |
An algorithmic framework for performing collaborative filtering
- Herlocker, Konstan, et al.
- 1999
(Show Context)
Citation Context ...is commonly used in collaborative ltering systems to avoid concluding that two users are very highly correlated simply because they rated very few movies in common, and by chance rated them similarly =-=[18-=-]. The neighbors of X i were the X j 's for which W ji was highest. With n i =5, a number we believe provides a reasonable tradeo between model accuracy and speed, the averagesW ji of neighbors was 0.... |

613 | Learning Probabilistic Relational Models
- Friedman, Getoor, et al.
- 1999
(Show Context)
Citation Context ...ection is towards more detailed node models and multiple types of relations between nodes. A theoretical framework for this could be provided by the probabilistic relational models of Friedman et al. =-=[14]-=-. We would also like to extend our approach to consider multiple types of marketing actions and product-design decisions, and to multi-player markets (i.e., markets where the actions of competitors mu... |

453 | Enhanced hypertext categorization using hyperlinks
- Chakrabarti, Dom, et al.
- 1998
(Show Context)
Citation Context ...herwise, an approximate solution is necessary. A standard method for this purpose is Gibbs sampling [16]. An alternative based on an ecient k-shortest-path algorithm is proposed in Chakrabarti et al. =-=[6]-=-. Given N i and Y, X i should be independent of the marketing actions for other customers. Assuming a naive Bayes model for X i as a function of N i , Y1 ; : : : ; Ym and M i [11], P (X i jN i ; Y;M) ... |

448 |
Information Rules: A Strategic Guide to the Network Economy:
- Shapiro, Varian
- 1998
(Show Context)
Citation Context ...onomics literature as network externalities) are of critical importance in many industries, including notably those associated with information goods (e.g., software, media, telecommunications, etc.) =-=[38-=-]. A technically inferior product can often prevail in the marketplace if it better leverages the network of users (for example, VHS prevailed over Beta in the VCR market). Ignoring network eects when... |

350 | Scale-free characteristics of random networks: the topology of the world-wide web,”
- Barabasi, Albert, et al.
- 2000
(Show Context)
Citation Context ...and the discovery of widespread network topologies with nontrivial properties [42], has led to asurry of research on modeling the Web as a semi-random graph (e.g., Kumar et al. [28], Barabasi et al. [=-=1]-=-). Some of this work might be applicable in our context. In retrospect, the earliest sign of the potential of viral marketing was perhaps the classic paper by Milgram [31] estimating that every person... |

245 | Phoaks: a system for sharing recommendations.
- Terveen, Hill, et al.
- 1997
(Show Context)
Citation Context ...levant probabilities are the same for all customers, and is only applied to a made-up network with seven nodes. Collaborativesltering systems proposed in the literature include GroupLens [35], PHOAKS =-=[40]-=-, Siteseer [36], and others. A list of collaborativesltering systems, projects and related resources can be found at www.sims.berkeley.- edu/resources/collab/. 6. FUTURE WORK The type of data mining p... |

197 |
Markov random fields and their applications,
- Kindermann, Snell
- 1980
(Show Context)
Citation Context ... ) P (X i jN i ; Y;M) Y X j 2N u i P (X j jX k ; Y;M) (2) The set of variables X u , with joint probability conditioned on X k , Y and M described by Equation 2, is an instance of a Markov randomseld =-=[2, 25, 7]-=-. Because Equation 2 expresses the probabilities P (X i jX k ; Y;M) as a function of themselves, it can be applied iteratively tosnd them, starting from a suitable initial assignment. This procedure i... |

194 |
Estimating probabilities: A crucial task in machine learning
- CESTNIK
- 1990
(Show Context)
Citation Context ...k jX i ), and P (R i jY). P (X i ) is simply the fraction of movies X i rated. We used a naive Bayes model for P (R j jY). P (Yk jX i ), P (R j jY), and P (X i ) were all smoothed using an m-estimate =-=[5-=-] with m=1 and the population average as the prior. We did not know the true values of P (M i jX i ). We expected marketing to have a larger eect on a customer who was already inclined to see the movi... |

186 | Graph-based data mining.
- Cook, Holder
- 2000
(Show Context)
Citation Context ...ge corresponding to properties of the product. Neville and Jensen [32] proposed a simple iterative algorithm for labeling nodes in social networks, based on the naive Bayes classier. Cook and Holder [=-=9-=-] developed a system for mining graph-based data. Flake et al. [13] used graph algorithms to mine communities from the Web (dened as sets of sites that have more links to each other than to non-member... |

156 | Data Mining for Direct Marketing: Problems and Solutions',
- Ling, Li
- 1998
(Show Context)
Citation Context ... [19]. Data mining plays a key role in this process, by allowing the construction of models that predict a customer's response given her past buying behavior and any available demographic information =-=[2-=-9]. When successful, this approach can signicantly increase prots [34]. One basic limitation of it is that it treats each customer as making a buying decision independently of all other customers. In ... |

131 | ReferralWeb: Combining social networks and collaborative filtering.
- Kautz, Selman, et al.
- 1997
(Show Context)
Citation Context ...and j means \i knows j." Schwartz and Wood [37] mined social relationships from email logs. The ReferralWeb project mined a social network from a wide variety of publicly-available online informa=-=tion [24]-=-, and used it to help individualssnd experts who could answer their questions. The COBOT project Free Movie 0 2 4 6 8 10 12 1 1.5 2 2.5 3 Alpha hill greedy single-pass direct Advertising -30 0 30 60 9... |

124 |
SiteSeer: personalized navigation for the Web,
- Rucker, Polanco
- 1997
(Show Context)
Citation Context ...ities are the same for all customers, and is only applied to a made-up network with seven nodes. Collaborativesltering systems proposed in the literature include GroupLens [35], PHOAKS [40], Siteseer =-=[36]-=-, and others. A list of collaborativesltering systems, projects and related resources can be found at www.sims.berkeley.- edu/resources/collab/. 6. FUTURE WORK The type of data mining proposed here op... |

121 | Extracting large-scale knowledge bases from the web.
- Kumar, Raghavan, et al.
- 1999
(Show Context)
Citation Context ...s of these approaches, and the discovery of widespread network topologies with nontrivial properties [42], has led to asurry of research on modeling the Web as a semi-random graph (e.g., Kumar et al. =-=[28-=-], Barabasi et al. [1]). Some of this work might be applicable in our context. In retrospect, the earliest sign of the potential of viral marketing was perhaps the classic paper by Milgram [31] estima... |

111 | Scalable techniques for mining causal structures,”
- Silverstein, Brin, et al.
- 2000
(Show Context)
Citation Context ...nce of causal relations between individuals (as opposed to purely correlational ones) is key. While mining causal knowledge from observational databases is dicult, there has been much recent progress =-=[10, 39]-=-. We have also assumed so far that the relevant social network is completely known. In many (or most) applications this will not be the case. For example, a long-distance telephone company may know th... |

86 |
Discovering shared interests using graph analysis, In:
- Schwartz, Wood
- 1993
(Show Context)
Citation Context ...keting was perhaps the classic paper by Milgram [31] estimating that every person in the world is only six edges away from every other, if an edge between i and j means \i knows j." Schwartz and =-=Wood [37]-=- mined social relationships from email logs. The ReferralWeb project mined a social network from a wide variety of publicly-available online information [24], and used it to help individualssnd expert... |

36 |
The buzz on buzz’,
- Dye
- 2000
(Show Context)
Citation Context ... systems (e.g., Breese at al. [3]). Motion picture marketing is an interesting application for the techniques we propose because the success of a movie is known to be strongly driven by word of mouth =-=[12]-=-. EachMovie is composed of three databases: one containing the ratings, one containing demographic information about the users (which we did not use), and one containing information about the movies. ... |

32 |
On the optimality of the simple Bayesian classi under zero-one loss
- Domingos, Pazzani
- 1997
(Show Context)
Citation Context ...in Chakrabarti et al. [6]. Given N i and Y, X i should be independent of the marketing actions for other customers. Assuming a naive Bayes model for X i as a function of N i , Y1 ; : : : ; Ym and M i =-=[11]-=-, P (X i jN i ; Y;M) = P (X i jN i ; Y;M i ) = P (X i )P (N i ; Y;M i jX i ) P (N i ; Y;M i ) = P (X i )P (N i jX i )P (M i jX i ) P (N i ; Y;M) m Y k=1 P (Yk jX i ) = P (X i jN i )P (M i jX i ) P (Y;... |

27 |
Statistics and data mining techniques for lifetime value modeling.
- Mani, Drew, et al.
- 1999
(Show Context)
Citation Context ... researchers have studied the problem of estimating a customer's lifetime value from data [22]. This line of research generally focuses on variables like an individual's expected tenure as a customer =-=[30]-=- and future frequency of purchases [15]. Customer networks have received some attention in the marketing literature [20]. Most of these studies are purely qualitative; where data sets appear, they are... |

21 | Cobot in LambdaMOO: A Social Statistics Agent
- Isbell, Kearns, et al.
- 2000
(Show Context)
Citation Context ...s and runtimes obtained using dierent marketing strategies. gathered social statistics from participant interactions in the LambdaMoo MUD, but did not explicitly construct a social network from them [=-=21-=-]. A Markov randomseld formulation similar to Equation 2 was used by Chakrabarti et al. [6] for classication of Web pages, with pages corresponding to customers, hyperlinks between pages corresponding... |

15 |
Structural leverage in marketing
- Krackhardt
- 1996
(Show Context)
Citation Context ...received some attention in the marketing literature [20]. Most of these studies are purely qualitative; where data sets appear, they are very small, and used only for descriptive purposes. Krackhardt =-=[2-=-7] proposes a very simple model for optimizing which customers to oer a free sample of a product to. The model only considers the impact on the customer's immediate friends, ignores the eect of produc... |

12 |
The Complete Database Marketing Second Generation Strategies and Techniques for Tapping the Power of Your Customer Database. Revised ed
- Hughes
- 1996
(Show Context)
Citation Context ...copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Copyright 2001 ACM X-XXXXX-XX-X/XX/XX ...$5.00. and market only to those =-=[19]-=-. Data mining plays a key role in this process, by allowing the construction of models that predict a customer's response given her past buying behavior and any available demographic information [29].... |

12 |
A continuous relaxation labeling algorithm for markov random fields
- Pelkowitz
- 1990
(Show Context)
Citation Context ...jX k ; Y;M) (1) where C(N u i ) is the set of all possible congurations of the unknown neighbors of X i (i.e., the set of all possible 2 jN u i j assignments of 0 and 1 to them). Following Pelkowitz [=-=33]-=-, we approximate P (N u i jX k ; Y;M) by its maximum entropy estimate given the marginals P (X j jX k ; Y;M), for X j 2 N u i . This yields 1 P (X i jX k ; Y;M) = X C(N u i ) P (X i jN i ; Y;M) Y X j ... |

9 | A decision theoretic approach to targeted advertising
- DM, Heckerman
- 2000
(Show Context)
Citation Context ...g M i to 1 and leaving the rest of M unchanged, and similarly for f 0 i (M). The expected lift in prot from marketing to customer i in isolation (i.e., ignoring her eect on other customers) is then [8=-=-=-] ELP i (X k ; Y;M) = r1P (X i =1jX k ; Y; f 1 i (M)) r0P (X i =1jX k ; Y; f 0 i (M)) c (4) Let M0 be the null vector (all zeros). The global lift in prot that results from a particular choice M of cu... |

9 |
Ecient identi of web communities
- Flake, Lawrence, et al.
- 2000
(Show Context)
Citation Context ... [32] proposed a simple iterative algorithm for labeling nodes in social networks, based on the naive Bayes classier. Cook and Holder [9] developed a system for mining graph-based data. Flake et al. [=-=13-=-] used graph algorithms to mine communities from the Web (dened as sets of sites that have more links to each other than to non-members). Several researchers have studied the problem of estimating a c... |

9 |
Dependency networks for inference, collaborative and data visualization
- Heckerman, Chickering, et al.
- 2000
(Show Context)
Citation Context ... the case. The probabilistic model obtained from it in the way described will then be an instance of a dependency network, a generalization of Markov randomselds recently proposed by Heckerman et al. =-=[17-=-]. Heckerman et al. show that Gibbs sampling applied to such a network denes a joint distribution from which all probabilities of interest can be computed. While in our experimental studies Gibbs samp... |

6 |
Value Miner: A Data Mining Environment for the Calculation of the Customer Lifetime Value with Application to the Automotive Industry, López de Mántaras
- Gelbrich, Nakhaeizadeh
- 2000
(Show Context)
Citation Context ...f estimating a customer's lifetime value from data [22]. This line of research generally focuses on variables like an individual's expected tenure as a customer [30] and future frequency of purchases =-=[15]-=-. Customer networks have received some attention in the marketing literature [20]. Most of these studies are purely qualitative; where data sets appear, they are very small, and used only for descript... |

6 |
Strategic Application of Customer Lifetime Value in the Direct Marketing Environment
- Jackson
- 1994
(Show Context)
Citation Context ...unities from the Web (dened as sets of sites that have more links to each other than to non-members). Several researchers have studied the problem of estimating a customer's lifetime value from data [=-=22]-=-. This line of research generally focuses on variables like an individual's expected tenure as a customer [30] and future frequency of purchases [15]. Customer networks have received some attention in... |

6 |
Collective dynamics of \small-world" networks
- Watts, Strogatz
- 1998
(Show Context)
Citation Context ...S algorithm forsnding hubs and authorities on the Web are based on social network ideas. The success of these approaches, and the discovery of widespread network topologies with nontrivial properties =-=[42-=-], has led to asurry of research on modeling the Web as a semi-random graph (e.g., Kumar et al. [28], Barabasi et al. [1]). Some of this work might be applicable in our context. In retrospect, the ear... |

4 |
A Simple Constraint-based Algorithm for Eciently Mining Observational Databases for Causal Relationships
- Cooper
- 1997
(Show Context)
Citation Context ...nce of causal relations between individuals (as opposed to purely correlational ones) is key. While mining causal knowledge from observational databases is dicult, there has been much recent progress =-=[10, 39]-=-. We have also assumed so far that the relevant social network is completely known. In many (or most) applications this will not be the case. For example, a long-distance telephone company may know th... |

4 |
What exactly is viral marketing? Red Herring
- Jurvetson
- 2000
(Show Context)
Citation Context ... service, which grew from zero to 12 million users in 18 months on a minuscule advertising budget, thanks to the inclusion of a promotional message with the service's URL in every email sent using it =-=[23]-=-. Competitors using conventional marketing fared far less well. This type of marketing, dubbed viral marketing because of its similarity to the spread of an epidemic, is now used by a growing number o... |

2 |
Estimating campaign bene and modeling lift
- Piatetsky-Shapiro, Masand
- 1999
(Show Context)
Citation Context ...onstruction of models that predict a customer's response given her past buying behavior and any available demographic information [29]. When successful, this approach can signicantly increase prots [3=-=4]-=-. One basic limitation of it is that it treats each customer as making a buying decision independently of all other customers. In reality, a person's decision to buy a product is often strongly in uen... |

1 |
Iterative classi in relational data
- Neville, Jensen
- 2000
(Show Context)
Citation Context ...pages corresponding to customers, hyperlinks between pages corresponding to in uence between customers, and the bag of words in the page corresponding to properties of the product. Neville and Jensen =-=[32-=-] proposed a simple iterative algorithm for labeling nodes in social networks, based on the naive Bayes classier. Cook and Holder [9] developed a system for mining graph-based data. Flake et al. [13] ... |