Mining the network value of customers

article2001KDD3,106 citations

Proposes a Markov random field framework to model viral marketing by quantifying a customer's network influence from collaborative filtering data, demonstrating that targeting highly connected individuals substantially outperforms traditional direct marketing.

Listen

The paper introduces a new approach to valuing customers that accounts for the influence they exert on others through social networks, rather than treating each customer’s purchases in isolation. Traditional marketing models calculate a customer’s value solely from direct revenue or profit, but this ignores word-of-mouth effects that can multiply or reduce returns on marketing spend. The authors argue that in networked markets—such as online communities, recommendation systems, or consumer product categories—identifying customers with high network value can materially change targeting decisions and overall campaign profitability.

The work sets out to estimate each customer’s total expected contribution to firm profit when marketing actions can propagate through the customer network, and to demonstrate that incorporating this network value improves marketing outcomes compared with conventional direct-marketing approaches. The authors model the market as a social network in which each customer’s purchase probability depends on both intrinsic preferences and the observed or expected behavior of network neighbors. They represent these dependencies with a Markov random field and develop an iterative inference procedure that computes the network-adjusted value of every customer. The method is applied to a large collaborative-filtering data set of movie ratings and purchases, using historical transaction records to estimate influence parameters and then running controlled marketing simulations that compare different targeting policies.

The analysis shows that network value varies widely across customers and often exceeds direct value by a substantial margin; a small fraction of customers generate the majority of propagated profit. When marketing budgets are allocated to the highest network-value customers, simulated profit rises sharply—by factors of two to ten relative to targeting on direct value alone—while random or purely demographic targeting produces far smaller gains. The same experiments indicate that simple one-step or greedy selection rules capture most of the available uplift, but more exhaustive search yields additional improvement at higher computational cost. Results are robust across different assumed strengths of social influence and different marketing cost structures.

These findings imply that customer-acquisition and retention programs should shift from ranking customers by historical spend or margin toward ranking them by expected network-adjusted contribution. In practice this means investing in data that reveal social ties, building models that propagate influence, and reallocating marketing dollars toward a smaller set of high-leverage customers. The approach also suggests that viral or referral incentives may be more cost-effective than broad direct campaigns when network effects are strong.

The authors recommend that firms begin by mapping observable social or transaction links among customers, then pilot network-value scoring on a modest segment before scaling. They note that further gains will require richer longitudinal data on actual influence, faster inference algorithms for very large graphs, and integration with existing CRM platforms. The current results rest on a single domain and several modeling assumptions about how influence travels; validation on additional data sets and sensitivity checks on those assumptions are needed before large-scale deployment.

University of Washington collaborative filtering data mining database systems Mining Network Value Model Market as Social Network Mining Social Network from Collaborative Filtering Database

Domingos et al (2001).pdf

Paper: Maximizing the spread of influence through a social network, David Kempe et al. (2003). This foundational paper provides the algorithmic framework and approximation guarantees for maximizing influence propagation in social networks, directly building upon the network-value concepts introduced in the source.

Paper: Design and analysis of experiments in networks: Reducing bias from interference, Dean Eckles et al. (2017). This paper extends the study of networked customer values by addressing how social network interference biases randomized marketing experiments, providing essential methods for evaluating such interventions.

Cover for Mining the network value of customers

Abstract

One of the major applications of data mining is in helping companies determine which potential customers to market to. If the expected profit from a customer is greater than the cost of marketing to her, the marketing action for that customer is executed. So far, work in this area has considered only the intrinsic value of the customer (i.e, the expected profit from sales to her). We propose to model also the customer's network value: the expected profit from sales to other customers she may influence to buy, the customers those may influence, and so on recursively. Instead of viewing a market as a set of independent entities, we view it as a social network and model it as a Markov random field. We show the advantages of this approach using a social network mined from a collaborative filtering database. Marketing that exploits the network value of customers—also known as viral marketing—can be extremely effective, but is still a black art. Our work can be viewed as a step towards providing a more solid foundation for it, taking advantage of the availability of large relevant databases.

Categories and Subject Descriptors
General Terms
1. INTRODUCTION
2. MODELING MARKETS AS SOCIAL NETWORKS
3. MINING SOCIAL NETWORKS FROM COLLABORATIVE FILTERING DATABASES
4. EMPIRICAL STUDY
4.1 The Model
4.2 The Data
4.3 Inference and Search
4.4 Model Accuracy
4.5 Network Values
4.6 Marketing Experiments
5. RELATED WORK
6. FUTURE WORK
7. CONCLUSION
8. REFERENCES

Knowls

Knowl 1 — Network Value, Intrinsic Value, and Total Value of a Customer
definition
In a market modeled as a social network of $n$ potential customers, let $X = \{X_1, \ldots, X_n\}$ be binary random variables where $X_i = 1$ indicates that customer $i$ purchases the product and $X_i = 0$ indicates non-purchase. Let $X^k$ be the subset of customers whose purchase decisions are known, $Y$ be the set of product attributes, and $M = \{M_1, \ldots, M_n\}$ be the vector of marketing actions, where $M_0 = (0, \ldots, 0)$ denotes marketing to no one.

Let $f_i^1(M)$ represent setting $M_i = 1$ while leaving all other elements of $M$ unchanged, and $f_i^0(M)$ represent setting $M_i = 0$ . Let $ELP(X^k, Y, M)$ denote the global expected lift in profit across all customers resulting from marketing plan $M$ .
1. Intrinsic Value: The expected profit lift derived solely from customer $i$ 's own purchase decision when evaluated in isolation, ignoring influence on others: $ELP_i(X^k, Y, M) = r_1 P(X_i = 1 \mid X^k, Y, f_i^1(M)) - r_0 P(X_i = 1 \mid X^k, Y, f_i^0(M)) - c$ where $r_1$ is revenue from customer $i$ if marketed to, $r_0$ is revenue if not marketed to, and $c$ is the cost of marketing to customer $i$ .
2. Total Value: The global expected profit lift across the entire network when customer $i$ is marketed to compared to when customer $i$ is not marketed to: $V_{\text{total}}(i) = ELP(X^k, Y, f_i^1(M)) - ELP(X^k, Y, f_i^0(M))$
3. Network Value: The value of customer $i$ derived strictly from their direct and recursive influence on other network members: $V_{\text{network}}(i) = V_{\text{total}}(i) - ELP_i(X^k, Y, M)$
Knowl 2 — Markov Random Field Formulation of Market Social Networks
model/method
A market is modeled as a Markov random field (or dependency network) over customer purchase variables $X = \{X_1, \ldots, X_n\}$ , where $X_i \in \{0, 1\}$ . Each customer $X_i$ is directly influenced by a set of neighbors $N_i \subseteq X \setminus \{X_i\}$ , such that $X_i$ is conditionally independent of $X \setminus (N_i \cup \{X_i\})$ given $N_i$ .

Given known customer states $X^k$ , product attributes $Y$ , and marketing actions $M$ , the marginal probability for unknown customer $X_i \in X^u = X \setminus X^k$ is approximated using a mean-field expansion over unknown neighbors $N_i^u = N_i \cap X^u$ : $P(X_i \mid X^k, Y, M) = \sum_{C(N_i^u)} P(X_i \mid N_i, Y, M) \prod_{X_j \in N_i^u} P(X_j \mid X^k, Y, M)$ where $C(N_i^u)$ is the set of all $2^{|N_i^u|}$ possible binary assignments of $0$ and $1$ to the unknown neighbors.

Inference is performed via relaxation labeling: all unknown marginal probabilities $P(X_i \mid X^k, Y, M)$ are initialized to network-less prior probabilities $P(X_i \mid Y, M)$ and iteratively updated using the equation above until the maximum change in probability across all nodes in an iteration is below a convergence threshold $\gamma$ (set to $\gamma = 10^{-5}$ ). A queue of modified nodes is maintained so that only neighbors of modified nodes are recomputed.
Knowl 3 — Naive Bayes Decomposition of Local Purchase Probability
model/method
To evaluate the local conditional probability $P(X_i \mid N_i, Y, M)$ in a Markov random field market model, $X_i$ is assumed conditionally independent of marketing actions for other customers $M \setminus \{M_i\}$ given $N_i, Y, M_i$ . Applying Bayes' theorem with a naive Bayes assumption that product attributes $Y = \{Y_1, \ldots, Y_m\}$ and marketing action $M_i$ are conditionally independent given $X_i$ yields: $P(X_i \mid N_i, Y, M) = \frac{P(X_i \mid N_i) P(M_i \mid X_i) \prod_{k=1}^m P(Y_k \mid X_i)}{P(Y, M_i \mid N_i)}$ where the denominator normalizer is: $P(Y, M_i \mid N_i) = \sum_{x \in \{0, 1\}} P(X_i = x \mid N_i) P(M_i \mid X_i = x) \prod_{k=1}^m P(Y_k \mid X_i = x)$

The corresponding network-less prior probability is: $P(X_i \mid Y, M) = \frac{P(X_i) P(M_i \mid X_i) \prod_{k=1}^m P(Y_k \mid X_i)}{P(Y, M_i)}$
Knowl 4 — Global Expected Lift in Profit Objective Function
equation
Let $X = \{X_1, \ldots, X_n\}$ be binary purchase decisions ( $X_i \in \{0, 1\}$ ) of $n$ customers, $X^k$ be known purchases, $Y$ be product attributes, and $M = \{M_1, \ldots, M_n\}$ be a binary marketing decision vector where $M_i = 1$ indicates that customer $i$ is marketed to and $M_i = 0$ indicates otherwise. Let $M_0 = (0, \ldots, 0)$ be the null marketing plan.

The global Expected Lift in Profit ( $ELP$ ) across the entire network is given by: $ELP(X^k, Y, M) = \sum_{i=1}^n r_i P(X_i = 1 \mid X^k, Y, M) - r_0 \sum_{i=1}^n P(X_i = 1 \mid X^k, Y, M_0) - |M|c$ where:
- $r_1$ is the revenue obtained from selling to customer $i$ when marketing action $M_i = 1$ is taken.
- $r_0$ is the revenue obtained from selling to customer $i$ when no marketing action ( $M_i = 0$ ) is taken.
- $r_i = r_1$ if $M_i = 1$ , and $r_i = r_0$ if $M_i = 0$ .
- $c$ is the unit cost of marketing to one customer.
- $|M| = \sum_{i=1}^n M_i$ is the total number of customers marketed to.
Knowl 5 — Target Selection Algorithms for Viral Marketing
algorithm
Three heuristic search algorithms find a marketing assignment $M \in \{0, 1\}^n$ that approximately maximizes global expected lift in profit $ELP(X^k, Y, M)$ :
Input: Customer network X, known purchases X^k, product attributes Y, cost c, revenues r_0, r_1
Output: Target marketing vector M in {0, 1}^n
Method 1: Single Pass
  Initialize M = (0, ..., 0)
  for each customer i in {1, ..., n} do
    Compute intrinsic lift Delta_i = ELP_i(X^k, Y, f_i^1(M_0))
    if Delta_i > 0 then
      M_i <- 1
    else
      M_i <- 0
  return M
Method 2: Greedy Search
  Initialize M = (0, ..., 0)
  repeat
    for each customer i in {1, ..., n} do
      if ELP(X^k, Y, f_i^1(M)) > ELP(X^k, Y, M) then
        M_i <- 1
      else
        M_i <- 0
  until M does not change during a complete pass over all n customers
  return M
Method 3: Hill-Climbing Search
  Initialize M = (0, ..., 0)
  repeat
    Find i* = argmax_i ELP(X^k, Y, f_i^1(M))
    if ELP(X^k, Y, f_{i*}^1(M)) > ELP(X^k, Y, M) then
      M_{i*} <- 1
  until no single customer change M_i = 1 increases ELP
  return M
Knowl 6 — Collaborative Filtering Estimation of Social Network Topology and Influence
model/method
A social network and its influence probabilities are constructed from a collaborative filtering database of explicit ratings $R_{ik} \in \{1, \ldots, 5\}$ across users $i$ and items $k$ .
1. User Similarity: The similarity $w_{ij}$ between users $i$ and $j$ is calculated via the Pearson correlation coefficient: $w_{ij} = \frac{\sum_k (R_{ik} - \bar{R}_i)(R_{jk} - \bar{R}_j)}{\sqrt{\sum_k (R_{ik} - \bar{R}_i)^2 \sum_k (R_{jk} - \bar{R}_j)^2}}$ where $\bar{R}_i$ is the average rating of user $i$ . A significance weighting penalty subtracts $0.05$ from $w_{ij}$ for each rated item in common fewer than $10$ .
2. Neighborhood Selection: The neighborhood $N_i$ consists of the $n_i$ users with the highest $w_{ji}$ (typically $n_i = 5$ , giving an average $w_{ji} = 0.91$ ).
3. Predicted Rating: The predicted rating deviation $\hat{R}_i$ for an item is: $\hat{R}_i = \bar{R}_i + \rho \sum_{X_j \in N_i} w_{ji} (R_j - \bar{R}_j)$ where $\rho = 1 / \sum_{X_j \in N_i} |w_{ji}|$ . If neighbor $j$ 's rating is unknown, $R_j - \bar{R}_j$ is set to $0$ if $X_j = 0$ , or to $E[R_j \mid Y] - \bar{R}_j$ if $X_j = 1$ .
4. Influence Distribution: Because $X_i$ depends on $N_i$ solely through $\hat{R}_i$ , the conditional purchase probability is $P(X_i \mid N_i) = P(X_i \mid \hat{R}_i(N_i))$ , which is estimated using a piecewise linear regression of $X_i$ on $\hat{R}_i$ smoothed with an $m$ -estimate ( $m=1$ ).
Knowl 7 — Five Characteristics of High Network Value Customers
empirical result
Customer network values in collaborative filtering networks are highly right-skewed: the vast majority of users have near-zero network value, while a very small subset exhibits high network value. A high network value customer is characterized by five properties:
1. High Purchase Probability and Rating: The customer has a high probability of buying the product and giving it a high rating ( $X_i = 1$ and high $R_i$ ).
2. High Influence Weight: The customer has large similarity weights $w_{ji}$ with many neighbors, giving them strong weight in determining neighbors' predicted ratings $\hat{R}_j$ .
3. Easily Influenced Neighbors: The customer's neighbors have purchase decisions that are highly sensitive to their received rating predictions.
4. High Neighbor Purchase Propensity: The customer's neighbors have a high baseline probability of purchasing the product upon recommendation.
5. Recursive Multiplier: The customer's neighbors themselves have neighbors who possess characteristics (1) through (4).
Knowl 8 — Profitability and Computational Efficiency of Network-Based Marketing Strategies
empirical result
Evaluation on the EachMovie dataset ( $2.8$ million ratings, $1628$ movies, $72916$ users, with $9585$ active test users) across three marketing models under the assumption $P(X_i=1 \mid M_i=1) = \min\{\alpha P(X_i=1 \mid M_i=0), 1\}$ with $\alpha > 1$ demonstrated:
1. Free Movie Scenario ( $r_0 = 1, r_1 = 0, c = 0.1$ ): Mass marketing loses profit ( $-2057$ to $-2712$ ). Direct marketing achieves zero profit because intrinsic lift is negative when giving a product away. Network marketing achieves positive profit (up to $\approx 11$ at $\alpha = 2.5$ ) by identifying influential customers who induce paying purchases by neighbors.
2. Discounted Movie Scenario ( $r_0 = 1, r_1 = 0.5, c = 0.1$ ): Direct marketing produces net losses, whereas network marketing methods achieve positive profits (up to $\approx 11$ at $\alpha = 2$ ).
3. Advertising Scenario ( $r_0 = 1, r_1 = 1, c = 0.02$ ): Mass marketing produces net losses ( $-143$ to $-381$ ). Direct marketing is profitable only at higher $\alpha$ , while network marketing substantially outperforms direct marketing across all $\alpha$ (profit $> 140$ at $\alpha = 1.4$ ).
4. Search Method Trade-offs: Greedy search produces profits within $9.6\%$ , $4.0\%$ , and $0.0\%$ of Hill-Climbing in the free movie, discounted movie, and advertising scenarios respectively, while taking only $3.8$ to $5.5$ minutes compared to $4.6$ to $42.1$ minutes (and $14$ hours for advertising) for Hill-Climbing.
Knowl 9 — Empirical Relationship Between Collaborative Filtering Predictions and Customer Purchase Probability
empirical result
In the EachMovie dataset, modeling customer viewing probability $P(X_i \mid \hat{R}_i)$ as a function of predicted rating deviation $\hat{R}_i$ using piecewise linear regression over nine bins ( $[-5.0, -2.0, -1.0, -0.5, -0.1, 0.1, 0.5, 1.0, 2.0, 5.0]$ ) revealed a U-shaped non-monotonic curve:
1. $P(X_i \mid \hat{R}_i)$ increases as $\hat{R}_i$ deviates substantially from $0$ in both positive and negative directions.
2. This effect is driven by movie popularity: popular movies attract more ratings, have larger absolute rating deviations $|\hat{R}_i|$ , and have higher overall viewing probability.
3. A positive rating deviation provides higher viewing probability than an equivalent negative deviation: $E[P(X_i \mid \hat{R}_i > 0)] = 0.203 > 0.176 = E[P(X_i \mid \hat{R}_i < 0)]$ .
4. Strictly monotonic increase of $P(X_i \mid \hat{R}_i)$ with $\hat{R}_i$ holds only within the narrow $[-0.1, 0.1]$ interval, which contains $96.4\%$ of rating predictions.
Knowl 10 — Optimization of Continuous Customer Acquisition Costs
model/method
When marketing action $M_i = c_i$ is a continuous variable representing the monetary expenditure allocated to customer $i$ , with a known functional relationship between expenditure $c_i$ and purchase probability $P(X_i \mid M_i)$ , the optimal customer acquisition cost $c_i^*$ for customer $i$ is determined by maximizing total customer value: $c_i^* = \arg\max_{c_i} \left( ELP(X^k, Y, f_i^1(M)) - ELP(X^k, Y, f_i^0(M)) \right)$ where $|M|c$ in the global $ELP$ formulation is replaced by the sum of continuous costs $\sum_{j=1}^n c_j$ .

Coverage note — No substantial contributed material was omitted from the knowls.

References

1.A. L. Barabási, R. Albert, and H. Jong. Scale-free characteristics of random networks: The topology of the World Wide Web. Physica A, 281:69–77, 2000.
2.J. Besag. Spatial interaction and the statistical analysis of lattice systems. Journal of the Royal Statistical Society, Series B, 36:192–236, 1974.
3.J. S. Breese, D. Heckerman, and C. Kadie. Empirical analysis of predictive algorithms for collaborative filtering. In Proceedings of the Fourteenth Conference on Uncertainty in Artificial Intelligence, Madison, WI, 1998. Morgan Kaufmann.
4.S. Brin and L. Page. The anatomy of a large-scale hypertextual Web search engine. In Proceedings of the Seventh International World Wide Web Conference, Brisbane, Australia, 1998. Elsevier.
5.B. Cestnik. Estimating probabilities: A crucial task in machine learning. In Proceedings of the Ninth European Conference on Artificial Intelligence, pages 147–149, Stockholm, Sweden, 1990. Pitman.
6.S. Chakrabarti, B. Dom, and P. Indyk. Enhanced hypertext categorization using hyperlinks. In Proceedings of the 1998 ACM SIGMOD International Conference on Management of Data, pages 307–318, Seattle, WA, 1998. ACM Press.
7.R. Chellappa and A. K. Jain, editors. Markov Random Fields: Theory and Application. Academic Press, Boston, MA, 1993.
8.D. M. Chickering and D. Heckerman. A decision theoretic approach to targeted advertising. In Proceedings of the Sixteenth Annual Conference on Uncertainty in Artificial Intelligence, Stanford, CA, 2000. Morgan Kaufmann.
9.D. J. Cook and L. B. Holder. Graph-based data mining. IEEE Intelligent Systems, 15:32–41, 2000.
10.G. F. Cooper. A simple constraint-based algorithm for efficiently mining observational databases for causal relationships. Data Mining and Knowledge Discovery, 1:203–224, 1997.
11.P. Domingos and M. Pazzani. On the optimality of the simple Bayesian classifier under zero-one loss. Machine Learning, 29:103–130, 1997.
12.R. Dye. The buzz on buzz. Harvard Business Review, 78(6):139–146, 2000.
13.G. W. Flake, S. Lawrence, and C. L. Giles. Efficient identification of Web communities. In Proceedings of the Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 150–160, Boston, MA, 2000. ACM Press.
14.N. Friedman, L. Getoor, D. Koller, and A. Pfeffer. Learning probabilistic relational models. In Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence, pages 1300–1307, Stockholm, Sweden, 1999. Morgan Kaufmann.
15.K. Gelbrich and R. Nakhaeizadeh. Value Miner: A data mining environment for the calculation of the customer lifetime value with application to the automotive industry. In Proceedings of the Eleventh European Conference on Machine Learning, pages 154–161, Barcelona, Spain, 2000. Springer.
16.S. Geman and D. Geman. Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 6:721–741, 1984.
17.D. Heckerman, D. M. Chickering, C. Meek, R. Rounthwaite, and C. Kadie. Dependency networks for inference, collaborative filtering, and data visualization. Journal of Machine Learning Research, 1:49–75, 2000.
18.J. Herlocker, J. Konstan, A. Borchers, and J. Riedl. An algorithmic framework for performing collaborative filtering. In Proceedings of the 1999 Conference on Research and Development in Information Retrieval, Berkeley, CA, 1999.
19.A. M. Hughes. The Complete Database Marketer: Second-Generation Strategies and Techniques for Tapping the Power of your Customer Database. Irwin, Chicago, IL, 1996.
20.D. Iacobucci, editor. Networks in Marketing. Sage, Thousand Oaks, CA, 1996.
21.C. L. Isbell, Jr., M. Kearns, D. Korman, S. Singh, and P. Stone. Cobot in LambdaMOO: A social statistics agent. In Proceedings of the Seventeenth National Conference on Artificial Intelligence, pages 36–41, Austin, TX, 2000. AAAI Press.
22.D. R. Jackson. Strategic application of customer lifetime value in direct marketing. Journal of Targeting, Measurement and Analysis for Marketing, 1:9–17, 1994.
23.S. Jurvetson. What exactly is viral marketing? Red Herring, 78:110–112, 2000.
24.H. Kautz, B. Selman, and M. Shah. ReferralWeb: Combining social networks and collaborative filtering. Communications of the ACM, 40(3):63–66, 1997.
25.R. Kindermann and J. L. Snell. Markov Random Fields and Their Applications. American Mathematical Society, Providence, RI, 1980.
26.J. M. Kleinberg. Authoritative sources in a hyperlinked environment. In Proceedings of the Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 668–677, Baltimore, MD, 1998. ACM Press.
27.D. Krackhardt. Structural leverage in marketing. In D. Iacobucci, editor, Networks in Marketing, pages 50–59. Sage, Thousand Oaks, CA, 1996.
28.R. Kumar, P. Raghavan, S. Rajagopalan, and A. Tomkins. Extracting large-scale knowledge bases from the Web. In Proceedings of the Twenty-Fifth International Conference on Very Large Databases, pages 639–650, Edinburgh, Scotland, 1999. Morgan Kaufmann.
29.C. X. Ling and C. Li. Data mining for direct marketing: Problems and solutions. In Proceedings of the Fourth International Conference on Knowledge Discovery and Data Mining, pages 73–79, New York, NY, 1998. AAAI Press.
30.D. R. Mani, J. Drew, A. Betz, and P. Datta. Statistics and data mining techniques for lifetime value modeling. In Proceedings of the Fifth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 94–103, New York, NY, 1999. ACM Press.
31.S. Milgram. The small world problem. Psychology Today, 2:60–67, 1967.
32.J. Neville and D. Jensen. Iterative classification in relational data. In Proceedings of the AAAI-2000 Workshop on Learning Statistical Models from Relational Data, pages 42–49, Austin, TX, 2000. AAAI Press.
33.L. Pelkowitz. A continuous relaxation labeling algorithm for Markov random fields. IEEE Transactions on Systems, Man and Cybernetics, 20:709–715, 1990.
34.G. Piatetsky-Shapiro and B. Masand. Estimating campaign benefits and modeling lift. In Proceedings of the Fifth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 185–193, San Diego, CA, 1999. ACM Press.
35.P. Resnick, N. Iacovou, M. Suchak, P. Bergstrom, and J. Riedl. GroupLens: An open architecture for collaborative filtering of netnews. In Proceedings of the ACM 1994 Conference on Computer Supported Cooperative Work, pages 175–186, New York, NY, 1994. ACM Press.
36.J. Rucker and M. J. Polanco. Siteseer: Personalized navigation for the web. Communications of the ACM, 40(3):73–76, 1997.
37.M. F. Schwartz and D. C. M. Wood. Discovering shared interests using graph analysis. Communications of the ACM, 36(8):78–89, 1993.
38.C. Shapiro and H. R. Varian. Information Rules: A Strategic Guide to the Network Economy. Harvard Business School Press, Boston, MA, 1999.
39.C. Silverstein, S. Brin, R. Motwani, and J. Ullman. Scalable techniques for mining causal structures. Data Mining and Knowledge Discovery, 4:163–192, 2000.
40.L. Terveen, W. Hill, B. Amento, D. McDonald, and J. Creter. PHOAKS: A system for sharing recommendations. Communications of the ACM, 40(3):59–62, 1997.
41.S. Wasserman and K. Faust. Social Network Analysis: Methods and Applications. Cambridge University Press, Cambridge, UK, 1994.
42.D. J. Watts and S. H. Strogatz. Collective dynamics of "small-world" networks. Nature, 393:440–442, 1998.

Citation

MLA

Domingos, P., and M. Richardson. “Mining the Network Value of Customers”. Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2001, pp. 57–66, https://doi.org/10.1145/502512.502525.

APA

Domingos, P., & Richardson, M. (2001). Mining the network value of customers. Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 57–66. https://doi.org/10.1145/502512.502525

Chicago

Domingos, P., and M. Richardson. 2001. “Mining the Network Value of Customers”. Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 57–66. https://doi.org/10.1145/502512.502525.

Harvard

Domingos, P. and Richardson, M. (2001) “Mining the network value of customers”, Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, pp. 57–66. Available at: https://doi.org/10.1145/502512.502525.

Vancouver

1. Domingos P, Richardson M (2001) Mining the network value of customers. In: Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, pp 57–66

BibTeX

@inproceedings{Domingos_2001, series={KDD01}, title={Mining the network value of customers}, url={http://dx.doi.org/10.1145/502512.502525}, DOI={10.1145/502512.502525}, booktitle={Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining}, publisher={ACM}, author={Domingos, Pedro and Richardson, Matt}, year={2001}, month=Aug, pages={57–66}, collection={KDD01} }

Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

Mining the network value of customers

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — Network Value, Intrinsic Value, and Total Value of a Customer

Knowl 2 — Markov Random Field Formulation of Market Social Networks

Knowl 3 — Naive Bayes Decomposition of Local Purchase Probability

Knowl 4 — Global Expected Lift in Profit Objective Function

Knowl 5 — Target Selection Algorithms for Viral Marketing

Knowl 6 — Collaborative Filtering Estimation of Social Network Topology and Influence

Knowl 7 — Five Characteristics of High Network Value Customers

Knowl 8 — Profitability and Computational Efficiency of Network-Based Marketing Strategies

Knowl 9 — Empirical Relationship Between Collaborative Filtering Predictions and Customer Purchase Probability

Knowl 10 — Optimization of Continuous Customer Acquisition Costs

References

Citation

BibTeX

Access the Paper