WO2009077655A1 - A method and an arrangement for segmentation of customers in a customer management system - Google Patents
A method and an arrangement for segmentation of customers in a customer management system Download PDFInfo
- Publication number
- WO2009077655A1 WO2009077655A1 PCT/FI2008/050740 FI2008050740W WO2009077655A1 WO 2009077655 A1 WO2009077655 A1 WO 2009077655A1 FI 2008050740 W FI2008050740 W FI 2008050740W WO 2009077655 A1 WO2009077655 A1 WO 2009077655A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- customer
- customers
- segmentation
- interests
- retrieved
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
Definitions
- the present invention relates in general to customer relationship managements systems (CRM) and more specifically to a method and an arrangement for segmentation of custom- ers in a customer management system that can be further utilized in prediction of customer behavior and interest.
- CRM customer relationship managements systems
- the background of the invention is discussed briefly in the following.
- the invention relates to a problem on how can the marketers of a certain product or service target networked individuals with relevant information on products, without bothering them with irrelevant advertising. Solving this problem requires finding the customers or customer groups who are most interested in the product.
- the target group to which a marketing message is sent is defined usually by the user' s demographics and/or previous purchase patterns.
- One of the typical ways to define the target group of users is to se- lect the most potential age and education level for a product. This way of selecting the target group of users to which marketing messages are sent is however inefficient, in a way that large group of messages is sent to different users without any response from the potential buyers. Therefore large group of unnecessary messages are sent through a network (e.g. Internet) .
- the marketing message covers traditional mail, commercials (on TV or radio), e-mails, mobile messages, etc.
- Another prior art solution that is used is to send e-mail messages to all possible e-mail addresses.
- the recent studies have revealed that about half of the e-mails sent in communications networks are already e-marketing messages. This method causes a lot of unnecessary traffic in the communications networks.
- a social network management system is any system which deals with analyzing, managing or visualizing social networks.
- social networks are considered in a wide sense.
- the network represents direct interactions between individuals (e.g., telephone calls, e-mails and online messaging) or indirect relationships between individuals with similar behaviour, interests or demographic features.
- Nodes represent the individuals and connections the similarity or amount of communication between the individuals. The connections may be either directed or indirected.
- the network may be either weighted or unweighted.
- the social network may represent the relationships between a set of items (such as books, videos or stores) which are connected through individuals who buy, use or share interests in the items.
- the nodes of the social network correspond to the items and the connections correspond to the strength of the similarity between the items.
- US 6,266,649 there is disclosed an service for recom- mending items to individual users based on a set of items that are known to be of interest to the user.
- the present invention realizes a method and an arrangement for segmentation of customers in a customer management system that can be further utilized in prediction of customer behavior and interest, and that better solves the presented problems in comparison to solutions according to prior art.
- a challenge with using network information in segmentation is that network data is noisy and incomplete.
- the present invention solves these problems by using Bayesian methods, which can flexibly deal with uncertainty and randomness.
- a method for segmentation of customers in a customer management system comprises the step of - segmenting customers to retrieve one or more customer segments by using social network topology and/or the combination of social network topology and other properties of customers, such as demography.
- the method further comprises the step of using said retrieved one or more customer segments in prediction of customer interests and preferences.
- the method further comprises the steps of - constructing a predictive model for predicting customer behavior, demography or interests by using said retrieved one or more customer segments, and using said retrieved one or more customer segments together with said retrieved predictive model in predic- tion of customer interests and preferences.
- the method further comprises the steps of predicting customer behavior, demography or interests to retrieve a predictive model by using said retrieved one or more customer segments and social network topology, and using said retrieved one or more customer segments together with said retrieved predictive model in prediction of customer interests and preferences .
- the method further comprises the steps of predicting customer behavior, demography or interests to retrieve a predictive model by using said retrieved one or more customer segments and the combination of social network topology and other properties of customers, such as demography, and using said retrieved one or more customer segments together with said retrieved predictive model in predic- tion of customer interests and preferences .
- the segmentation of customers is performed by using hierarchical clustering.
- the segmentation of customers is performed by using Bayesian clus- tering.
- the segmentation is optimized with collapsed Gibbs sampling.
- the segmentation is optimized with Expectation Maximization algorithm.
- the segmentation is optimized with Markov
- the predicting customer interests and preferences is carried out by averaging over segments.
- the predicting customer interests and preferences is carried out by using network segments as inputs to a predictive model, such as a regression model.
- an arrangement for segmentation of customers in a customer management system which arrangement has : - means for segmenting customers to retrieve one or more customer segments by using social network topology and/or the combination of social network topology and other properties of customers, such as demography.
- the arrangement further has: means for predicting customer interests and preferences by using said retrieved one or more customer segments .
- the arrangement further has: means for constructing a predictive model for predicting customer behavior, demography or interests by using said retrieved one or more customer segments, and means for predicting customer interests and prefer- ences by using said retrieved one or more customer segments together with said retrieved predictive model.
- the arrangement further has: means for retrieving a predictive model by using said retrieved one or more customer segments and social network topology in predicting customer behavior, demography or interests, and means for predicting customer interests and preferences by using said retrieved one or more customer seg- ments together with said retrieved predictive model.
- the arrangement further has: means for retrieving a predictive model by using said retrieved one or more customer segments and the combina- tion of social network topology and other properties of customers, such as demography in predicting customer behavior, demography or interests, and means for predicting customer interests and preferences by using said retrieved one or more customer seg- ments together with said retrieved predictive model.
- the means for segmenting are suited for performing the segmentation of customers by using hierarchical clustering.
- the means for segmenting are suited for performing the segmentation of customers by using Bayesian clustering.
- the means for segmenting are suited for optimizing the segmentation of customers with collapsed Gibbs sampling.
- the means for segmenting are suited for optimizing the segmentation of customers with Expectation Maximization algorithm.
- the means for segmenting are suited for optimizing the segmentation of customers with Markov Chain Monte Carlo method.
- the means for segmenting are suited for optimizing the segmentation of customers with approximate Bayesian inference method.
- the means for predicting customer interests and preferences are suited for carrying out the predicting by averaging over segments.
- the means for predicting customer interests and preferences are suited for carrying out the predicting by using network segments as inputs to a predictive model, such as a regression model .
- Figure 1 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to the present invention.
- Figure 2 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to the present invention.
- Figure 3 illustrates a simplified network topology of a social network according to the present invention. Detailed description of certain embodiments
- the solution according to the present invention presents a new method and a new arrangement for segmentation of cus- tomers in a customer management system that can be further utilized in prediction of customer behavior and interest.
- Figure 1 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to the present invention.
- a social network is created 1 based on the relations between customers or potential customers. These relations may be formed based on interactions, communication or similarity of behaviour of individuals. Alternatively, one may also combine interaction, relation and behaviour data in the construction of the relations between individuals.
- the network is segmented either based on the social network topology alone 2a or on the combination of social network topology and demography 2b.
- the segmenting of customers for marketing/CRM by using social network topology is carried out by: combining topology and demographics, or using topology alone.
- This segmentation can be performed by for example: using a Bayesian method, or using hierarchical clustering.
- Figure 2 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to an alternative embodiment of the pre- sent invention.
- a social network is created 1 based on the relations between customers or potential customers. These relations may be formed based on interactions, communication or similarity of behaviour of individuals. Alternatively, one may also combine interaction, relation and behaviour data in the construction of the relations between individuals.
- the network is segmented either based on the social network topology alone 2a or on the combination of social network topology and demography 2b.
- said retrieved one or more customer segments can be used directly 3a in prediction of customer interests and preferences.
- said retrieved one or more customer segments can be used as an input in predictive modeling 3b of customer behavior, demography or in- terests to retrieve a predictive model.
- This said predictive model can then be used together with said retrieved one or more customer segments in prediction of customer interests and preferences.
- social network topology 3c can be used as input in the predictive model in addition to said retrieved one or more customer segments.
- combination of social network topology and demography 3d can be used as input in the predictive model in addition to said retrieved one or more customer segments.
- the segmentation and predictive model is used 4 for example in CRM, Customer Insight, Marketing and Targeting.
- the solution according to an alternative embodiment of the present invention involves: segmenting customers for marketing/CRM by using social network topology, predicting customer behavior, demography or interests on the basis of social network topology combined with be- havior, demography and/or interests of other customers, and using topology based segments in prediction of customer interests and preferences.
- the segmenting of customers for marketing/CRM by using social network topology is carried out by: combining topology and demographics, or - using topology alone.
- This segmentation can be performed by for example: using a Bayesian method, or using hierarchical clustering.
- topology based segments in prediction of customer interests and preferences is carried out by: - by trivial methods, such as averaging over segments, or by using network segments as inputs to a predictive model, such as a regression model (or other statistical predictive tools) .
- Figure 3 illustrates a simplified network topology of a social network according to the present invention.
- the social network segmentation is either based on the topology alone or by combining the topology with other information.
- the segmentation in itself is a form of network clustering problem.
- Network clustering according to the present invention can be performed using many methods, such as hierarchical clustering or Bayesian approaches.
- Bayesian methods make it possible to use both the social network topology and demographics in the segmentation. Alternatively, only the social network topology can also be used.
- Social networks according to the present invention may contain tens of millions of nodes and hundreds of millions edges. Therefore, the clustering method according to the present invention is selected to scale to very large net- works .
- the segmentation based on topography of social networks is important especially for analyzing online customers, for example visitors of web sites or web forums, or mobile pre-paid customers whose demographical details are not known or incomplete .
- the segmentation based on topography of social networks can be used in gaining customer insight in order to find out what kinds of customers are interconnected or what are the "natural" social groups of our customers.
- the segmentation based on topography of social networks is useful in initial targeting of products based on global network segmentation (grouping) from topology; a suitable target segment can be selected for a new product.
- the segmentation according to the present invention can be carried out based on Bayesian clustering approach.
- the segmentation based on Bayesian clustering approach uses a generative component model for constructing the edges of a network and optionally the network node attrib- utes.
- Bayesian clustering approach uses a prior art model for segmenting a network based on topology alone.
- this prior art model is extended to use node attributes in addition to social network topology.
- Bayesian methods have been slow and unusable on large data sets.
- the segmentation based on Bayesian clustering approach of the present invention is usable on net- works with tens of millions of nodes and hundreds of millions of edges.
- the segmentation model of the present invention can learn the number of clusters (i.e. groups or segments) in the data automatically.
- the Bayesian segmentation of the present invention can be used in segmenting a network by topology alone or by com- bining the topology and node (person) attributes, such as gender, geographical location and interests.
- the prior art model is a probabilistic component model which models the way edges of a network have been constructed.
- components have a Dirichlet process or a Dirichlet distribution as the prior art.
- the model is assumed to generate edges by drawing edge end points (nodes) from distributions associated with each of the components.
- this prior art model is extended for modeling the node (customer) properties (such as per- son age, gender or location) in addition to the network structure.
- node customer
- properties such as per- son age, gender or location
- the generative process out of which the network arises is the following (where i and j are the edge endpoints and h is an node attribute) :
- the model of the present invention is able to combine data from both network and demographical domains in a probabilistic way in which the uncertainty and randomness in any data items is dealt with in a flexible manner. Also, for data sets with no node attributes, the model behaves in a similar way to the prior art model. In this way, it can be used as an effective segmentation tool for a wide range of purposes without making any modifications to the model.
- each node may belong to multiple clusters. This provides a "smooth" or soft segmentation of the nodes which is richer in structure than that obtained with segmentation methods based on hard clustering.
- the segments can be estimated using Bayesian inference.
- a property of generative models with conjugate priors of this type is that when the data generated with the model (i.e., social network and node attributes) is known, we may infer or estimate, what type of clustering (segmentation) is the most likely to have created the network structure. This provides a probabilistic estimation of most likely segments for each of the network edges and nodes .
- model parameters can be simply and effectively estimated using collapsed Gibbs sampling, which is a form of MCMC (Markov Chain Monte Carlo) method.
- collapsed Gibbs sampling is a form of MCMC (Markov Chain Monte Carlo) method.
- the segmentation of customers can also be performed by using some other network clustering method, such as non- probabilistic agglomerative clustering or non- agglomerative clustering.
- network clustering method such as non- probabilistic agglomerative clustering or non- agglomerative clustering.
- the predicting customer behavior, demography or interests according to the present invention is carried out by using predictive modeling on the basis of social network topology combined with behavior, demography and/or interests of other customers.
- the social network features are used in predicting the properties of customers whose information is missing.
- a standard approach to prediction of missing data is to use linear predictive model, such as a regression (or other statistical prediction tools) to build a model of the parameters.
- This model can be then used to estimate missing data.
- the model may incorporate features, such as node demographics, customer behavior, social network properties and segmentation of the customers .
- Information on customers who belong to the same segment can be used to estimate the missing data. Based on attributes known for even only a few nodes in a segment, the attributes of all the other nodes can be predicted, and
- Node attributes can be predicted directly based on the attributes of its neighbors. In this case an implicit segmentation of the network is unnecessary.
- the social network of customers can be used in marketing, advertising and customer relationship manage- ment (CRM) purposes, such as: predicting customer preferences and interests, grouping similar customers (segmentation) , marketing products to existing customers, and/or marketing products to new customers.
- CRM customer relationship manage- ment
- the solution according to the present invention can be used in clustering the customers into segments (groups) that are similar to each other and marketing products to them.
- the solution according to the present invention can also be used in using network attributes as explanatory variables in predictive models, and using the models in finding customers that are likely to match certain properties.
- the solution according to the present invention can also be used in combining the clustering and the using of network attributes as parameters. This can be done by, in addition to other parameters, using also the segments of customers in a customer management system as a parameter in a predictive model.
- topology When the topology of a social network is known but not all of the attributes of the nodes and edges in a network, topology can provide information that makes it possible to group similar persons (nodes) or connections (edges) together and to predict properties of persons and connections .
- the social networks can be used to segment more effectively; e.g. in Internet communities and mobile networks, a social network can be built based on which people interact with each other or have listed each other as ' friends' .
- a method and an arrangement for segmentation of customers in a customer management system that can be further utilized in prediction of customer be- havior and interest, and that better solves the presented problems in comparison to solutions according to prior art .
Abstract
The present invention relates in general to customer relationship managements systems (CRM) and more specifically to a method and an arrangement for segmentation of customers in a customer management system that can be further utilized in prediction of customer behavior and interest. With the help of the solution according to the present invention there is provided a method and an arrangement for segmentation of customers in a customer management system that comprises means for segmenting customers to retrieve one or more customer segments by using social network topology and/or the combination of social network topology and other properties of customers, such as demography.
Description
A METHOD AND AN ARRANGEMENT FOR SEGMENTATION OF CUSTOMERS IN A CUSTOMER MANAGEMENT SYSTEM
Technical field of the invention
The present invention relates in general to customer relationship managements systems (CRM) and more specifically to a method and an arrangement for segmentation of custom- ers in a customer management system that can be further utilized in prediction of customer behavior and interest.
Background of the invention
The background of the invention is discussed briefly in the following. The invention relates to a problem on how can the marketers of a certain product or service target networked individuals with relevant information on products, without bothering them with irrelevant advertising. Solving this problem requires finding the customers or customer groups who are most interested in the product.
Nowadays active users of fast developing products, e.g. computer software, want to know of any new versions of the software or updates thereto. They also want to know about the new features and advantages (when compared to older version) of those products before their releases. Also some of the users are also interested of the release dates of new version and other possible information they may re- ceive of the new product. Another interest of certain group of people, where from they want to know new releases, is books and movies. In this case the persons may be interested of certain writer (or certain type of books) or filmmaker (or certain type of movies) . These persons wish to receive information of any new release of that certain writer or filmmaker.
However, since interests of people varies a lot, nowadays there is no real solution in which marketing could be directed to people that are interested of a new product.
In one marketing solution, the target group to which a marketing message is sent is defined usually by the user' s demographics and/or previous purchase patterns. One of the typical ways to define the target group of users is to se- lect the most potential age and education level for a product. This way of selecting the target group of users to which marketing messages are sent is however inefficient, in a way that large group of messages is sent to different users without any response from the potential buyers. Therefore large group of unnecessary messages are sent through a network (e.g. Internet) . In this connection the marketing message covers traditional mail, commercials (on TV or radio), e-mails, mobile messages, etc.
Another prior art solution that is used is to send e-mail messages to all possible e-mail addresses. The recent studies have revealed that about half of the e-mails sent in communications networks are already e-marketing messages. This method causes a lot of unnecessary traffic in the communications networks.
In addition to the above drawbacks of the traditional marketing efforts, also the sales and marketing costs are unnecessarily high since there is a plurality of messages sent to various persons who are not interested of the new product. Also one drawback of so called mass marketing is that persons who would be interested of a new product do not necessarily realize the interesting marketing messages from all the messages received.
The technological development has improved the gain of information of various products and services. People get more and more information from various sources whether by finding that by himself/herself or through advertising campaigns. Especially electronic mail has provided an easy way to advertise products. On the other hand the effectiveness of the electronic mail relies on mass marketing, which also causes a lot of unnecessary sending of mail to people who does not even want to know about some of these products.
Even though marketing their products to possible customers is very important for companies, there is not easy enough way to target the marketing effort in order to effectively find the people who would be interested of their products or services.
Further, when companies (e.g. chain of stores) wish to market their products to their customers, they may only use their own databases for sending information of offers. The marketing messages (whether through conventional mail, broadcasting, or other marketing) may only be directed to the customers whose profiles are stored in their databases (not e.g. for occasional shopper in a store) . For target marketing of special offers to certain customers, the store (or chain of stores) needs to rely on their own databases of their customers' interests (what customers have informed the store e.g. through registration to valued customer program and the possible use of loyalty card when shopping) when targeting offers. Therefore, the store does not even know any of other purchases in other stores or other interests of their customers.
In the scope of the present invention, a social network management system is any system which deals with analyzing, managing or visualizing social networks. Furthermore,
in the scope of the present invention, social networks are considered in a wide sense. The network represents direct interactions between individuals (e.g., telephone calls, e-mails and online messaging) or indirect relationships between individuals with similar behaviour, interests or demographic features. Nodes represent the individuals and connections the similarity or amount of communication between the individuals. The connections may be either directed or indirected. The network may be either weighted or unweighted.
In addition, the social network may represent the relationships between a set of items (such as books, videos or stores) which are connected through individuals who buy, use or share interests in the items. In this case, the nodes of the social network correspond to the items and the connections correspond to the strength of the similarity between the items. For example in US patent specification US 6,266,649 there is disclosed an service for recom- mending items to individual users based on a set of items that are known to be of interest to the user.
There are some prior art methods and arrangements for segmentation of customers in a customer management system, i.e. clustering of individuals in social networks. Some of these prior art solutions involve a set of individuals which has been divided into segments i.e. communities, which differ in demographic features, such as age and gender. Other prior art solutions involve using predictive models based on demographics alone.
One prior art solution related to a system and method for discovering knowledge communities has been disclosed in US patent application publication US 2006/0112105.
One prior art solution related to a system and method for discovering communities in networks has been disclosed in US patent application publication US 2006/0080422.
One prior art solution related to a solution for discovering communities-of-practice has been disclosed in US patent application publication US 2005/0138070.
One prior art solution related to a method and apparatus for learning probabilistic relational models has been disclosed in US patent application publication US 2002/0103793.
However, the presented prior art solution do not take the networked nature of the individuals into account and do not adequately solve the problem on how can the marketers of a certain product or service target networked individuals with relevant information on products, without bothering them with irrelevant advertising.
Summary of the present invention
It is an object of the present invention to overcome or at least mitigate the disadvantages of the prior art. The present invention realizes a method and an arrangement for segmentation of customers in a customer management system that can be further utilized in prediction of customer behavior and interest, and that better solves the presented problems in comparison to solutions according to prior art.
A challenge with using network information in segmentation is that network data is noisy and incomplete. The present invention solves these problems by using Bayesian methods, which can flexibly deal with uncertainty and randomness.
According to a first aspect of the present invention there is presented a method for segmentation of customers in a customer management system, which method comprises the step of - segmenting customers to retrieve one or more customer segments by using social network topology and/or the combination of social network topology and other properties of customers, such as demography.
Preferably, the method further comprises the step of using said retrieved one or more customer segments in prediction of customer interests and preferences.
Alternatively, the method further comprises the steps of - constructing a predictive model for predicting customer behavior, demography or interests by using said retrieved one or more customer segments, and using said retrieved one or more customer segments together with said retrieved predictive model in predic- tion of customer interests and preferences.
Further alternatively, the method further comprises the steps of predicting customer behavior, demography or interests to retrieve a predictive model by using said retrieved one or more customer segments and social network topology, and using said retrieved one or more customer segments together with said retrieved predictive model in prediction of customer interests and preferences .
Further alternatively, the method further comprises the steps of predicting customer behavior, demography or interests to retrieve a predictive model by using said retrieved one or more customer segments and the combination of social
network topology and other properties of customers, such as demography, and using said retrieved one or more customer segments together with said retrieved predictive model in predic- tion of customer interests and preferences .
Preferably, the segmentation of customers is performed by using hierarchical clustering. Alternatively, the segmentation of customers is performed by using Bayesian clus- tering.
Preferably, the segmentation is optimized with collapsed Gibbs sampling. Alternatively, the segmentation is optimized with Expectation Maximization algorithm. Further al- ternatively, the segmentation is optimized with Markov
Chain Monte Carlo method. Further alternatively, the segmentation is optimized with approximate Bayesian inference method.
Preferably, the predicting customer interests and preferences is carried out by averaging over segments. Alternatively, the predicting customer interests and preferences is carried out by using network segments as inputs to a predictive model, such as a regression model.
According to a second aspect of the present invention there is presented an arrangement for segmentation of customers in a customer management system, which arrangement has : - means for segmenting customers to retrieve one or more customer segments by using social network topology and/or the combination of social network topology and other properties of customers, such as demography.
Preferably, the arrangement further has:
means for predicting customer interests and preferences by using said retrieved one or more customer segments .
Alternatively, the arrangement further has: means for constructing a predictive model for predicting customer behavior, demography or interests by using said retrieved one or more customer segments, and means for predicting customer interests and prefer- ences by using said retrieved one or more customer segments together with said retrieved predictive model.
Further alternatively, the arrangement further has: means for retrieving a predictive model by using said retrieved one or more customer segments and social network topology in predicting customer behavior, demography or interests, and means for predicting customer interests and preferences by using said retrieved one or more customer seg- ments together with said retrieved predictive model.
Further alternatively, the arrangement further has: means for retrieving a predictive model by using said retrieved one or more customer segments and the combina- tion of social network topology and other properties of customers, such as demography in predicting customer behavior, demography or interests, and means for predicting customer interests and preferences by using said retrieved one or more customer seg- ments together with said retrieved predictive model.
Preferably, the means for segmenting are suited for performing the segmentation of customers by using hierarchical clustering. Alternatively, the means for segmenting are suited for performing the segmentation of customers by using Bayesian clustering.
Preferably, the means for segmenting are suited for optimizing the segmentation of customers with collapsed Gibbs sampling. Alternatively, the means for segmenting are suited for optimizing the segmentation of customers with Expectation Maximization algorithm. Further alternatively, the means for segmenting are suited for optimizing the segmentation of customers with Markov Chain Monte Carlo method. Further alternatively, the means for segmenting are suited for optimizing the segmentation of customers with approximate Bayesian inference method.
Preferably, the means for predicting customer interests and preferences are suited for carrying out the predicting by averaging over segments. Alternatively, the means for predicting customer interests and preferences are suited for carrying out the predicting by using network segments as inputs to a predictive model, such as a regression model .
Brief description of the drawings
For a better understanding of the present invention and in order to show how the same may be carried into effect ref- erence will now be made to the accompanying drawings, in which :
Figure 1 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to the present invention. Figure 2 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to the present invention.
Figure 3 illustrates a simplified network topology of a social network according to the present invention.
Detailed description of certain embodiments
The solution according to the present invention presents a new method and a new arrangement for segmentation of cus- tomers in a customer management system that can be further utilized in prediction of customer behavior and interest.
Figure 1 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to the present invention. In the method according to the present invention first a social network is created 1 based on the relations between customers or potential customers. These relations may be formed based on interactions, communication or similarity of behaviour of individuals. Alternatively, one may also combine interaction, relation and behaviour data in the construction of the relations between individuals.
Thereafter, in the method according to the present inven- tion the network is segmented either based on the social network topology alone 2a or on the combination of social network topology and demography 2b.
In the solution according to the present invention the segmenting of customers for marketing/CRM by using social network topology is carried out by: combining topology and demographics, or using topology alone.
This segmentation can be performed by for example: using a Bayesian method, or using hierarchical clustering.
Figure 2 illustrates a flowchart presentation of a method for segmentation of customers in a customer management system according to an alternative embodiment of the pre-
sent invention. In the method according to an alternative embodiment of the present invention first a social network is created 1 based on the relations between customers or potential customers. These relations may be formed based on interactions, communication or similarity of behaviour of individuals. Alternatively, one may also combine interaction, relation and behaviour data in the construction of the relations between individuals.
Thereafter, in the method according to an alternative embodiment of the present invention the network is segmented either based on the social network topology alone 2a or on the combination of social network topology and demography 2b.
Next, in the method according to an alternative embodiment of the present invention said retrieved one or more customer segments can be used directly 3a in prediction of customer interests and preferences.
Alternatively, in the method according to an alternative embodiment of the present invention said retrieved one or more customer segments can be used as an input in predictive modeling 3b of customer behavior, demography or in- terests to retrieve a predictive model. This said predictive model can then be used together with said retrieved one or more customer segments in prediction of customer interests and preferences.
In the method according to an alternative embodiment of the present invention also social network topology 3c can be used as input in the predictive model in addition to said retrieved one or more customer segments. Furthermore, in the method according to an alternative embodiment of the present invention also the combination of social network topology and demography 3d can be used as input in
the predictive model in addition to said retrieved one or more customer segments.
Thereafter, in the method according to an alternative em- bodiment of the present invention the segmentation and predictive model is used 4 for example in CRM, Customer Insight, Marketing and Targeting.
The solution according to an alternative embodiment of the present invention involves: segmenting customers for marketing/CRM by using social network topology, predicting customer behavior, demography or interests on the basis of social network topology combined with be- havior, demography and/or interests of other customers, and using topology based segments in prediction of customer interests and preferences.
In the solution according to an alternative embodiment of the present invention the segmenting of customers for marketing/CRM by using social network topology is carried out by: combining topology and demographics, or - using topology alone.
This segmentation can be performed by for example: using a Bayesian method, or using hierarchical clustering.
In the solution according to an alternative embodiment of the present invention the use of topology based segments in prediction of customer interests and preferences is carried out by: - by trivial methods, such as averaging over segments, or
by using network segments as inputs to a predictive model, such as a regression model (or other statistical predictive tools) .
Figure 3 illustrates a simplified network topology of a social network according to the present invention. In the solution according to the present invention the social network segmentation is either based on the topology alone or by combining the topology with other information. The segmentation in itself is a form of network clustering problem. Network clustering according to the present invention can be performed using many methods, such as hierarchical clustering or Bayesian approaches.
In the solution according to the present invention the
Bayesian methods make it possible to use both the social network topology and demographics in the segmentation. Alternatively, only the social network topology can also be used.
Social networks according to the present invention may contain tens of millions of nodes and hundreds of millions edges. Therefore, the clustering method according to the present invention is selected to scale to very large net- works .
With the help of the solution according to the present invention and based on topography of social networks, it is possible to perform segmentation even if very limited or no information on the demographics of individual customers is available.
The segmentation based on topography of social networks is important especially for analyzing online customers, for example visitors of web sites or web forums, or mobile
pre-paid customers whose demographical details are not known or incomplete .
The segmentation based on topography of social networks can be used in gaining customer insight in order to find out what kinds of customers are interconnected or what are the "natural" social groups of our customers.
The segmentation based on topography of social networks is useful in initial targeting of products based on global network segmentation (grouping) from topology; a suitable target segment can be selected for a new product.
The segmentation according to the present invention can be carried out based on Bayesian clustering approach.
The segmentation based on Bayesian clustering approach uses a generative component model for constructing the edges of a network and optionally the network node attrib- utes.
Bayesian clustering approach uses a prior art model for segmenting a network based on topology alone. In the present invention, this prior art model is extended to use node attributes in addition to social network topology.
Typically Bayesian methods have been slow and unusable on large data sets. The segmentation based on Bayesian clustering approach of the present invention is usable on net- works with tens of millions of nodes and hundreds of millions of edges. In addition, the segmentation model of the present invention can learn the number of clusters (i.e. groups or segments) in the data automatically.
The Bayesian segmentation of the present invention can be used in segmenting a network by topology alone or by com-
bining the topology and node (person) attributes, such as gender, geographical location and interests.
To have a better understanding on the model used in the present invention, a short introduction on the prior art model follows. For further information on the prior art model, see Janne Sinkkonen, Janne Aukia, and Samuel Kaski: Inferring vertex properties from topology in large networks, In Working Notes of the 5th International Workshop on Mining and Learning with Graphs (MLG' 07), Florence, Italy, 2007, Universita degli Studi di Firenze, Extended Abstract.
The prior art model is a probabilistic component model which models the way edges of a network have been constructed. In the model components have a Dirichlet process or a Dirichlet distribution as the prior art. The model is assumed to generate edges by drawing edge end points (nodes) from distributions associated with each of the components.
For the prior art model, the generative process out of which the network arises is the following:
In the present invention, this prior art model is extended for modeling the node (customer) properties (such as per-
son age, gender or location) in addition to the network structure. In the model, instead of just generating edge end points from each component, it is assumed that even node attributes are drawn from multinomial distributions associated with each of the components.
In the case of the model of the present invention, the generative process out of which the network arises is the following (where i and j are the edge endpoints and h is an node attribute) :
The model of the present invention is able to combine data from both network and demographical domains in a probabilistic way in which the uncertainty and randomness in any data items is dealt with in a flexible manner. Also, for data sets with no node attributes, the model behaves in a similar way to the prior art model. In this way, it can be used as an effective segmentation tool for a wide range of purposes without making any modifications to the model. In the model, each node (person) may belong to multiple clusters. This provides a "smooth" or soft segmentation of the nodes which is richer in structure than that obtained with segmentation methods based on hard clustering.
The segments (components) can be estimated using Bayesian inference. A property of generative models with conjugate priors of this type is that when the data generated with the model (i.e., social network and node attributes) is known, we may infer or estimate, what type of clustering
(segmentation) is the most likely to have created the network structure. This provides a probabilistic estimation of most likely segments for each of the network edges and nodes .
By using a Polya or Blackwell-MacQueen urn representation (For further information see, Johnson, Urn models and their applications, John Wiley and Sons, 1977; Tavare and Ewens, The Ewens sampling formula, In Multivariate dis- crete distributions, John Wiley & Sons, New York, USA,
1997; Blackwell and MacQueen, Ferguson distributions via Polya urn schemes, Annals of Statistics, 1:353-355, 1973), the model parameters can be simply and effectively estimated using collapsed Gibbs sampling, which is a form of MCMC (Markov Chain Monte Carlo) method. For further information on collapsed Gibbs sampling, see Radford M. Neal . Markov chain sampling methods for Dirichlet process mixture models, Journal of Computational and Graphical Statistics, 9 (2) :249-265, 2000.
Instead of Gibbs sampling, other methods, such as EM- algorithm (Expectation Maximization) , other MCMC approaches, or approximate Bayesian inference (e.g. variational methods) , may be used for estimating the model pa- rameters .
The segmentation of customers can also be performed by using some other network clustering method, such as non- probabilistic agglomerative clustering or non- agglomerative clustering. This includes, for example, hierarchical clustering methods (see for example Clauset, Newman and Moore, 2004; Finding community structure in very large networks, Physical Review E, 70 ( 6) : 066111, Newman and Girvan, 2004; Finding and evaluating community structure in networks, Physical Review E, 69 (2) : 026113) .
The predicting customer behavior, demography or interests according to the present invention is carried out by using predictive modeling on the basis of social network topology combined with behavior, demography and/or interests of other customers.
In the predictive modeling according to the present invention the social network features are used in predicting the properties of customers whose information is missing.
In the predictive modeling according to the present invention a standard approach to prediction of missing data is to use linear predictive model, such as a regression (or other statistical prediction tools) to build a model of the parameters. This model can be then used to estimate missing data. Furthermore, the model may incorporate features, such as node demographics, customer behavior, social network properties and segmentation of the customers .
In the predictive modeling according to the present invention alternative approaches to using social networks in prediction of customer features include:
Information on customers who belong to the same segment can be used to estimate the missing data. Based on attributes known for even only a few nodes in a segment, the attributes of all the other nodes can be predicted, and
Node attributes can be predicted directly based on the attributes of its neighbors. In this case an implicit segmentation of the network is unnecessary.
With the help of the solution according to the present invention, the social network of customers can be used in marketing, advertising and customer relationship manage- ment (CRM) purposes, such as: predicting customer preferences and interests,
grouping similar customers (segmentation) , marketing products to existing customers, and/or marketing products to new customers.
The solution according to the present invention can be used in clustering the customers into segments (groups) that are similar to each other and marketing products to them.
The solution according to the present invention can also be used in using network attributes as explanatory variables in predictive models, and using the models in finding customers that are likely to match certain properties.
The solution according to the present invention can also be used in combining the clustering and the using of network attributes as parameters. This can be done by, in addition to other parameters, using also the segments of customers in a customer management system as a parameter in a predictive model.
When the topology of a social network is known but not all of the attributes of the nodes and edges in a network, topology can provide information that makes it possible to group similar persons (nodes) or connections (edges) together and to predict properties of persons and connections .
With the help of the solution according to the present in- vention the social networks can be used to segment more effectively; e.g. in Internet communities and mobile networks, a social network can be built based on which people interact with each other or have listed each other as ' friends' .
With the help of the solution according to the present invention there is provided a method and an arrangement for segmentation of customers in a customer management system that can be further utilized in prediction of customer be- havior and interest, and that better solves the presented problems in comparison to solutions according to prior art .
Claims
1. A method for segmentation of customers in a customer management system, characterized in that the method com- prises the step of segmenting customers to retrieve one or more customer segments by using social network topology and/or the combination of social network topology and other properties of customers, such as demography.
2. A method according to Claim 1, characterized in that the method further comprises the step of using said retrieved one or more customer segments in prediction of customer interests and preferences.
3. A method according to Claim 1, characterized in that the method further comprises the steps of constructing a predictive model for predicting customer behavior, demography or interests by using said re- trieved one or more customer segments, and using said retrieved one or more customer segments together with said constructed predictive model in prediction of customer interests and preferences .
4. A method according to Claim 1, characterized in that the method further comprises the steps of predicting customer behavior, demography or interests to retrieve a predictive model by using said retrieved one or more customer segments and social network topology, and - using said retrieved one or more customer segments together with said retrieved predictive model in prediction of customer interests and preferences .
5. A method according to Claim 1, characterized in that the method further comprises the steps of predicting customer behavior, demography or interests to retrieve a predictive model by using said retrieved one or more customer segments and the combination of social network topology and other properties of customers, such as demography, and using said retrieved one or more customer segments together with said retrieved predictive model in prediction of customer interests and preferences.
6. A method according to any one of the claims 1-5, characterized in that the segmentation of customers is performed by using hierarchical clustering.
7. A method according to any one of the claims 1-5, characterized in that the segmentation of customers is performed by using Bayesian clustering.
8. A method according to Claim 7, characterized in that the segmentation is optimized with collapsed Gibbs sam- pling.
9. A method according to Claim 7, characterized in that the segmentation is optimized with Expectation Maximization algorithm.
10. A method according to Claim 7, characterized in that the segmentation is optimized with Markov Chain Monte Carlo method.
11. A method according to Claim 7, characterized in that the segmentation is optimized with approximate Bayesian inference method.
12. A method according to any one of the claims 2-11, characterized in that the predicting customer interests and preferences is carried out by averaging over segments.
13. A method according to any one of the claims 2-11, characterized in that the predicting customer interests and preferences is carried out by using network segments as inputs to a predictive model, such as a regression model .
14. An arrangement for segmentation of customers in a customer management system, the arrangement being charac- terized in that the arrangement having means for segmenting customers to retrieve one or more customer segments by using social network topology and/or the combination of social network topology and other properties of customers, such as demography.
15. An arrangement according to Claim 14, characterized in that the arrangement further has : means for predicting customer interests and preferences by using said retrieved one or more customer seg- ments .
16. An arrangement according to Claim 14, characterized in that the arrangement further has : means for constructing a predictive model for pre- dieting customer behavior, demography or interests by using said constructed one or more customer segments, and means for predicting customer interests and preferences by using said retrieved one or more customer segments together with said retrieved predictive model.
17. An arrangement according to Claim 14, characterized in that the arrangement further has : means for retrieving a predictive model by using said retrieved one or more customer segments and social network topology in predicting customer behavior, demography or interests, and means for predicting customer interests and preferences by using said retrieved one or more customer segments together with said retrieved predictive model.
18. An arrangement according to Claim 14, characterized in that the arrangement further has : means for retrieving a predictive model by using said retrieved one or more customer segments and the combination of social network topology and other properties of customers, such as demography in predicting customer behavior, demography or interests, and means for predicting customer interests and preferences by using said retrieved one or more customer segments together with said retrieved predictive model.
19. An arrangement according to any one of the claims 14- 18, characterized in that the means for segmenting are suited for performing the segmentation of customers by using hierarchical clustering.
20. An arrangement according to any one of the claims 14- 18, characterized in that the means for segmenting are suited for performing the segmentation of customers by using Bayesian clustering.
21. An arrangement according to Claim 20, characterized in that the means for segmenting are suited for optimizing the segmentation of customers with collapsed Gibbs sampling .
22. An arrangement according to Claim 20, characterized in that the means for segmenting are suited for optimizing the segmentation of customers with Expectation Maximization algorithm.
23. An arrangement according to Claim 20, characterized in that the means for segmenting are suited for optimizing the segmentation of customers with Markov Chain Monte Carlo method.
24. An arrangement according to Claim 20, characterized in that the means for segmenting are suited for optimizing the segmentation of customers with approximate Bayesian inference method.
25. An arrangement according to any one of the claims 15- 24, characterized in that the means for predicting customer interests and preferences are suited for carrying out the predicting by averaging over segments.
26. An arrangement according to any one of the claims 15- 24, characterized in that the means for predicting customer interests and preferences are suited for carrying out the predicting by using network segments as inputs to a predictive model, such as a regression model.
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
FI20075914A FI20075914A0 (en) | 2007-12-14 | 2007-12-14 | A method and arrangement for segmenting customers |
FI20075914 | 2007-12-14 | ||
FI20085399A FI20085399A0 (en) | 2007-12-14 | 2008-04-30 | A method and arrangement for segmenting clients in a client management system |
FI20085399 | 2008-04-30 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2009077655A1 true WO2009077655A1 (en) | 2009-06-25 |
Family
ID=39386022
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/FI2008/050740 WO2009077655A1 (en) | 2007-12-14 | 2008-12-15 | A method and an arrangement for segmentation of customers in a customer management system |
Country Status (2)
Country | Link |
---|---|
FI (1) | FI20085399A0 (en) |
WO (1) | WO2009077655A1 (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010026297A1 (en) * | 2008-09-08 | 2010-03-11 | Xtract Oy | A method and an arrangement for predicting customer demographics |
WO2011019731A3 (en) * | 2009-08-10 | 2011-04-28 | Mintigo Ltd. | Systems and methods for gererating leads in a network by predicting properties of external nodes |
WO2012079835A1 (en) * | 2010-12-15 | 2012-06-21 | International Business Machines Corporation | Method and system for carrying out predictive analysis relating to nodes of a communication network |
US8208943B2 (en) | 2009-02-02 | 2012-06-26 | Waldeck Technology, Llc | Anonymous crowd tracking |
US8417780B2 (en) | 2007-12-21 | 2013-04-09 | Waldeck Technology, Llc | Contiguous location-based user networks |
US8560608B2 (en) | 2009-11-06 | 2013-10-15 | Waldeck Technology, Llc | Crowd formation based on physical boundaries and other rules |
US8688717B2 (en) * | 2012-02-16 | 2014-04-01 | Accenture Global Service Limited | Method and apparatus for generating and using an interest graph |
US8711737B2 (en) | 2009-12-22 | 2014-04-29 | Waldeck Technology, Llc | Crowd formation based on wireless context information |
US8898288B2 (en) | 2010-03-03 | 2014-11-25 | Waldeck Technology, Llc | Status update propagation based on crowd or POI similarity |
US20150193854A1 (en) * | 2014-01-06 | 2015-07-09 | Palo Alto Research Center Incorporated | Automated compilation of graph input for the hipergraph solver |
US9763048B2 (en) | 2009-07-21 | 2017-09-12 | Waldeck Technology, Llc | Secondary indications of user locations and use thereof by a location-based service |
US9886727B2 (en) | 2010-11-11 | 2018-02-06 | Ikorongo Technology, LLC | Automatic check-ins and status updates |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1998033135A1 (en) * | 1997-01-28 | 1998-07-30 | Firefly Network, Inc. | Improved method and apparatus for item recommendation using automated collaborative filtering |
US20050138070A1 (en) * | 2003-12-19 | 2005-06-23 | Huberman Bernardo A. | Discovering communities-of-practice |
US20060143081A1 (en) * | 2004-12-23 | 2006-06-29 | International Business Machines Corporation | Method and system for managing customer network value |
US20070240119A1 (en) * | 2006-04-11 | 2007-10-11 | Palo Alto Research Center | Method, device, and program product to monitor the social health of a persistent virtual environment |
US20080005072A1 (en) * | 2006-06-28 | 2008-01-03 | Microsoft Corporation | Search engine that identifies and uses social networks in communications, retrieval, and electronic commerce |
US20080162260A1 (en) * | 2006-12-29 | 2008-07-03 | Google Inc. | Network node ad targeting |
US20080275849A1 (en) * | 2007-02-01 | 2008-11-06 | Sugato Basu | Method and apparatus for targeting messages to users in a social network |
US20080275861A1 (en) * | 2007-05-01 | 2008-11-06 | Google Inc. | Inferring User Interests |
US20080313251A1 (en) * | 2007-06-15 | 2008-12-18 | Li Ma | System and method for graph coarsening |
-
2008
- 2008-04-30 FI FI20085399A patent/FI20085399A0/en not_active Application Discontinuation
- 2008-12-15 WO PCT/FI2008/050740 patent/WO2009077655A1/en active Application Filing
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1998033135A1 (en) * | 1997-01-28 | 1998-07-30 | Firefly Network, Inc. | Improved method and apparatus for item recommendation using automated collaborative filtering |
US20050138070A1 (en) * | 2003-12-19 | 2005-06-23 | Huberman Bernardo A. | Discovering communities-of-practice |
US20060143081A1 (en) * | 2004-12-23 | 2006-06-29 | International Business Machines Corporation | Method and system for managing customer network value |
US20070240119A1 (en) * | 2006-04-11 | 2007-10-11 | Palo Alto Research Center | Method, device, and program product to monitor the social health of a persistent virtual environment |
US20080005072A1 (en) * | 2006-06-28 | 2008-01-03 | Microsoft Corporation | Search engine that identifies and uses social networks in communications, retrieval, and electronic commerce |
US20080162260A1 (en) * | 2006-12-29 | 2008-07-03 | Google Inc. | Network node ad targeting |
US20080275849A1 (en) * | 2007-02-01 | 2008-11-06 | Sugato Basu | Method and apparatus for targeting messages to users in a social network |
US20080275861A1 (en) * | 2007-05-01 | 2008-11-06 | Google Inc. | Inferring User Interests |
US20080313251A1 (en) * | 2007-06-15 | 2008-12-18 | Li Ma | System and method for graph coarsening |
Non-Patent Citations (7)
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8417780B2 (en) | 2007-12-21 | 2013-04-09 | Waldeck Technology, Llc | Contiguous location-based user networks |
WO2010026297A1 (en) * | 2008-09-08 | 2010-03-11 | Xtract Oy | A method and an arrangement for predicting customer demographics |
US9098723B2 (en) | 2009-02-02 | 2015-08-04 | Waldeck Technology, Llc | Forming crowds and providing access to crowd data in a mobile environment |
US8825074B2 (en) | 2009-02-02 | 2014-09-02 | Waldeck Technology, Llc | Modifying a user'S contribution to an aggregate profile based on time between location updates and external events |
US8321509B2 (en) | 2009-02-02 | 2012-11-27 | Waldeck Technology, Llc | Handling crowd requests for large geographic areas |
US9515885B2 (en) | 2009-02-02 | 2016-12-06 | Waldeck Technology, Llc | Handling crowd requests for large geographic areas |
US8208943B2 (en) | 2009-02-02 | 2012-06-26 | Waldeck Technology, Llc | Anonymous crowd tracking |
US9763048B2 (en) | 2009-07-21 | 2017-09-12 | Waldeck Technology, Llc | Secondary indications of user locations and use thereof by a location-based service |
WO2011019731A3 (en) * | 2009-08-10 | 2011-04-28 | Mintigo Ltd. | Systems and methods for gererating leads in a network by predicting properties of external nodes |
US9300704B2 (en) | 2009-11-06 | 2016-03-29 | Waldeck Technology, Llc | Crowd formation based on physical boundaries and other rules |
US8560608B2 (en) | 2009-11-06 | 2013-10-15 | Waldeck Technology, Llc | Crowd formation based on physical boundaries and other rules |
US9046987B2 (en) | 2009-12-22 | 2015-06-02 | Waldeck Technology, Llc | Crowd formation based on wireless context information |
US8711737B2 (en) | 2009-12-22 | 2014-04-29 | Waldeck Technology, Llc | Crowd formation based on wireless context information |
US8898288B2 (en) | 2010-03-03 | 2014-11-25 | Waldeck Technology, Llc | Status update propagation based on crowd or POI similarity |
US9886727B2 (en) | 2010-11-11 | 2018-02-06 | Ikorongo Technology, LLC | Automatic check-ins and status updates |
TWI505667B (en) * | 2010-12-15 | 2015-10-21 | Ibm | Method and system for carrying out predictive analysis relating to nodes of a communication network |
US8644468B2 (en) | 2010-12-15 | 2014-02-04 | International Business Machines Corporation | Carrying out predictive analysis relating to nodes of a communication network |
CN103250376A (en) * | 2010-12-15 | 2013-08-14 | 国际商业机器公司 | Method and system for carrying out predictive analysis relating to nodes of a communication network |
CN103250376B (en) * | 2010-12-15 | 2017-07-07 | 国际商业机器公司 | Method and system for performing the forecast analysis relevant with the node of communication network |
WO2012079835A1 (en) * | 2010-12-15 | 2012-06-21 | International Business Machines Corporation | Method and system for carrying out predictive analysis relating to nodes of a communication network |
US8688717B2 (en) * | 2012-02-16 | 2014-04-01 | Accenture Global Service Limited | Method and apparatus for generating and using an interest graph |
US20150193854A1 (en) * | 2014-01-06 | 2015-07-09 | Palo Alto Research Center Incorporated | Automated compilation of graph input for the hipergraph solver |
Also Published As
Publication number | Publication date |
---|---|
FI20085399A0 (en) | 2008-04-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2009077655A1 (en) | A method and an arrangement for segmentation of customers in a customer management system | |
Dolega et al. | Going digital? The impact of social media marketing on retail website traffic, orders and sales | |
Hill et al. | Network-based marketing: Identifying likely adopters via consumer networks | |
Jung et al. | Persona generation from aggregated social media data | |
Shen et al. | Learning personal+ social latent factor model for social recommendation | |
US7680746B2 (en) | Prediction of click through rates using hybrid kalman filter-tree structured markov model classifiers | |
US8484083B2 (en) | Method and apparatus for targeting messages to users in a social network | |
US20110082824A1 (en) | Method for selecting an optimal classification protocol for classifying one or more targets | |
Kazienko et al. | AdROSA—Adaptive personalization of web advertising | |
Yang et al. | Mining social networks for targeted advertising | |
Raghu et al. | Dynamic profiling of consumers for customized offerings over the Internet: A model and analysis | |
US11823291B2 (en) | Predictive segmentation of customers | |
US10853730B2 (en) | Systems and methods for generating a brand Bayesian hierarchical model with a category Bayesian hierarchical model | |
US9727653B2 (en) | System and method for identifying and ranking user preferences | |
US20080208674A1 (en) | Targeting advertising content in a virtual universe (vu) | |
Yang et al. | Discovering cohesive subgroups from social networks for targeted advertising | |
Liu et al. | Large-scale behavioral targeting with a social twist | |
Yang et al. | An improved collaborative filtering method for recommendations' generation | |
Yang et al. | A unified view of social and temporal modeling for B2B marketing campaign recommendation | |
Nepal et al. | A behaviour-based trust model for service web | |
Li et al. | Corals: Who are my potential new customers? tapping into the wisdom of customers' decisions | |
Zhang et al. | Mining target users for mobile advertising based on telecom big data | |
Lacerda et al. | Improving daily deals recommendation using explore-then-exploit strategies | |
US20170330221A1 (en) | Systems and methods for integration of universal marketing activities | |
Ngonmang et al. | Monetization and services on a real online social network using social network analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 08861847 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 08861847 Country of ref document: EP Kind code of ref document: A1 |