## Temporal-relational classifiers for prediction in evolving domains (2008)

Venue: | In Proceedings of the IEEE International Conference on Data Mining |

Citations: | 17 - 5 self |

### BibTeX

@INPROCEEDINGS{Sharan08temporal-relationalclassifiers,

author = {Umang Sharan},

title = {Temporal-relational classifiers for prediction in evolving domains},

booktitle = {In Proceedings of the IEEE International Conference on Data Mining},

year = {2008}

}

### OpenURL

### Abstract

Many relational domains contain temporal information and dynamics that are important to model (e.g., social networks, protein networks). However, past work in relational learning has focused primarily on modeling static “snapshots” of the data and has largely ignored the temporal dimension of these data. In this work, we extend relational techniques to temporally-evolving domains and outline a representational framework that is capable of modeling both temporal and relational dependencies in the data. We develop efficient learning and inference techniques within the framework by considering a restricted set of temporalrelational dependencies and using parameter-tying methods to generalize across relationships and entities. More specifically, we model dynamic relational data with a twophase process, first summarizing the temporal-relational information with kernel smoothing, and then moderating attribute dependencies with the summarized relational information. We develop a number of novel temporal-relational models using the framework and then show that the current approaches to modeling static relational data are special cases within the framework. We compare the new models to the competing static relational methods on three real-world datasets and show that the temporal-relational models consistently outperform the relational models that ignore temporal information—achieving significant reductions in error ranging from 15 % to 70%. 1

### Citations

601 | On the Optimality of the Simple Bayesian Classifier under Zero-One Loss
- Domingos, Pazzani
- 1997
(Show Context)
Citation Context ...ssuming that each value of the multiset is independently drawn from the same multinomial distribution. This approach is designed to mirror the independence assumption of the naive Bayesian classifier =-=[7]-=-. In addition to the conventional assumption of attribute independence, the RBC also assumes attribute value independence within each multiset. More formally, for a class label C, attributes X, and re... |

510 | Learning probabilistic relational models
- Getoor, Friedman, et al.
- 2001
(Show Context)
Citation Context ...large datasets. In the area of Markov decision processes, some recent work has focused on modeling attributes that change over time. Guestrin et al. [13] combine probabilistic relational models (PRM) =-=[10]-=- with dynamic Bayesian networks [12] to model the state transitions in relational Markov decision processes (RMDP). In essence, RMDPs are a series of PRMs, one for each time step, linked together thro... |

384 | Enhanced Hypertext Categorization Using Hyperlinks
- Chakrabarti, Dom, et al.
- 1998
(Show Context)
Citation Context ...ion—achieving significant reductions in error ranging from 15% to 70%. 1 Introduction Recent research has demonstrated the utility of modeling relational information for domains such as web analytics =-=[5]-=-, marketing [8] and fraud detection [19]. This work has demonstrated that incorporating the characteristics of related instances into statistical models improves the accuracy of attribute predictions.... |

347 | Discriminative Probabilistic Models for Relational Data
- Taskar, Abbeel, et al.
(Show Context)
Citation Context ... to use RBCs and RPTs as the relational models in this work due to their simplicity. However, the TVRC framework is flexible enough that it can be used with other statistical relational models (e.g., =-=[10, 22, 24]-=-) as long as the models can be modified to deal with weighted instances. Another strength of our approach is that it is a relatively simple and efficient way of incorporating time into statistical rel... |

327 | Mining the network value of customers
- Domingos, Richardson
- 2001
(Show Context)
Citation Context ...ignificant reductions in error ranging from 15% to 70%. 1 Introduction Recent research has demonstrated the utility of modeling relational information for domains such as web analytics [5], marketing =-=[8]-=- and fraud detection [19]. This work has demonstrated that incorporating the characteristics of related instances into statistical models improves the accuracy of attribute predictions. However, this ... |

266 | On the bursty evolution of blogspace
- Kumar, Novak, et al.
- 2003
(Show Context)
Citation Context ...ristics of the Web by representing its evolution as a series of graph snapshots (see e.g., [2]). Similar data structures have been used to study the evolution of communities and behavior in Blogspace =-=[17]-=-. Temporal changes in link structure have also been modeled by Cortes et al. [6] and Hill et al. [15]. This work was motivated by the problem of analyzing large-scale dynamic networks such as telecomm... |

127 |
Citation analysis as a tool in journal evaluation
- Garfield
- 1972
(Show Context)
Citation Context ...ional changes in a concise way that evolves smoothly through time and has been utilized for fraud detection analysis. Temporal locality has been studied for many years in citation analysis (see e.g., =-=[9, 11]-=-)—it is well-known that a scientific paper gets the majority of its citations soon after it is published and as time passes, receives fewer citations. Amitay et al. [1] have also identified aspects of... |

124 | Learning dynamic bayesian networks
- Ghahramani
- 1998
(Show Context)
Citation Context ...v decision processes, some recent work has focused on modeling attributes that change over time. Guestrin et al. [13] combine probabilistic relational models (PRM) [10] with dynamic Bayesian networks =-=[12]-=- to model the state transitions in relational Markov decision processes (RMDP). In essence, RMDPs are a series of PRMs, one for each time step, linked together through temporal slot chains following a... |

116 | Learning relational probability trees
- Neville, Jensen, et al.
- 2003
(Show Context)
Citation Context ... Xm∈X G(i) P (X i m|C) · Y Y j∈R Xk∈XG(j) where wt ij graph GS t on the path from node i to the related node j. is the product of the weights in the summary Weighted Relational Probability Trees RPTs =-=[20]-=- extend standard probability estimation trees to a relational setting in which data instances are heterogeneous and interdependent. The RPT algorithm searches over a space of relational features that ... |

104 | Markov chain Monte Carlo estimation of exponential random graph models
- Snijders
(Show Context)
Citation Context ...sequential linear models [14]. This work represents temporal sequences of network structures as first-order Markov chains where each network instance is generated by an exponential random graph model =-=[23]-=-. The probability distribution for a graph at time t only depends on the graph at time (t−1): P (G t |G t−1 ) = 1 Z(θ,G t−1 ) exp{θ′ ψ(G t , G t−1 )}. However, this model was designed for link predict... |

95 | Linkage and autocorrelation cause feature selection bias in relational learning
- Jensen, Neville
- 2002
(Show Context)
Citation Context ...ato grapher Past Movie earnings year Past Movie earnings year Past Movie earnings year Figure 6. IMDb query with attributes used for summarization and classification. can result in statistical biases =-=[16]-=-, we conjecture that i.i.d. cross-validation is effective in this situation because we are focused on selecting a single parameter—so all choices of that parameter value are biased uniformly and thus,... |

94 | N.: Generalizing plans to new environments in relational MDPs
- Guestrin, Koller, et al.
- 2003
(Show Context)
Citation Context ... transforming temporal-varying links and objects into static aggregated features [19] and other work has focused on modeling the temporal dynamics of time-varying attributes in static link structures =-=[13]-=-. There have been some recent efforts to model temporally-varying links to improve automatic discovery of relational communities or groups [4, 15] but this work has not attempted to exploit temporal l... |

77 | Who links to whom: Mining linkage between web sites
- Bharat, Henzinger, et al.
- 2001
(Show Context)
Citation Context ...e (e.g., in the World Wide Web or citation graphs). In particular, many researchers have studied the characteristics of the Web by representing its evolution as a series of graph snapshots (see e.g., =-=[2]-=-). Similar data structures have been used to study the evolution of communities and behavior in Blogspace [17]. Temporal changes in link structure have also been modeled by Cortes et al. [6] and Hill ... |

76 | A machine learning approach to building domain-specific search engines
- McCallum, Nigam, et al.
- 1999
(Show Context)
Citation Context ...ely to indicate a stronger underlying relationship than an event isolated in time. As illustration, consider the Cora database of computer science research papers extracted automatically from the web =-=[18]-=-. Each paper has an associated topic and citations to other papers that have been published in the past. Figure 1 shows the autocorrelation between the topics of papers published in the year 1996 with... |

60 | Simple estimators for relational Bayesian classifiers
- Neville, Jensen, et al.
- 2003
(Show Context)
Citation Context ... In this work, we extend the relational Bayes classifier (RBC) and the relational probability tree (RPT) because of their relative simplicity and efficiency. Weighted Relational Bayes Classifier RBCs =-=[21]-=- extend naive Bayes classifiers to relational settings by treating heterogeneous relational subgraphs as homogenous sets of attribute multisets. For example, when modeling the dependencies between the... |

54 | Aggregation-Based Feature Invention and Relational Concept Classes. Paper presented at the
- Perlich, Provost
(Show Context)
Citation Context ... to use RBCs and RPTs as the relational models in this work due to their simplicity. However, the TVRC framework is flexible enough that it can be used with other statistical relational models (e.g., =-=[10, 22, 24]-=-) as long as the models can be modified to deal with weighted instances. Another strength of our approach is that it is a relatively simple and efficient way of incorporating time into statistical rel... |

48 | Communities of interest
- Cortes, Pregibon, et al.
- 2002
(Show Context)
Citation Context ...see e.g., [2]). Similar data structures have been used to study the evolution of communities and behavior in Blogspace [17]. Temporal changes in link structure have also been modeled by Cortes et al. =-=[6]-=- and Hill et al. [15]. This work was motivated by the problem of analyzing large-scale dynamic networks such as telecommunications call networks. In order to process massive volumes of data efficientl... |

45 | Using relational knowledge discovery to prevent securities fraud
- Neville, Simsek, et al.
- 2005
(Show Context)
Citation Context ... error ranging from 15% to 70%. 1 Introduction Recent research has demonstrated the utility of modeling relational information for domains such as web analytics [5], marketing [8] and fraud detection =-=[19]-=-. This work has demonstrated that incorporating the characteristics of related instances into statistical models improves the accuracy of attribute predictions. However, this work has focused primaril... |

26 | Recovering temporally rewiring networks: A model-based approach
- GUO, HANNEKE, et al.
- 2007
(Show Context)
Citation Context ...porating hyperlink timestamps into link-based page-ranking algorithms can improve retrieval accuracy. In the area of graph models, recent work has modeled dynamic graphs with sequential linear models =-=[14]-=-. This work represents temporal sequences of network structures as first-order Markov chains where each network instance is generated by an exponential random graph model [23]. The probability distrib... |

20 | Building an effective representation for dynamic networks
- Hill, Agarwal, et al.
- 2006
(Show Context)
Citation Context ...ics of time-varying attributes in static link structures [13]. There have been some recent efforts to model temporally-varying links to improve automatic discovery of relational communities or groups =-=[4, 15]-=- but this work has not attempted to exploit temporal link information in a classification context. The goal of this work is to improve attribute prediction in dynamic domains by incorporating the infl... |

16 | Trend Detection Through Temporal Link Analysis
- Amitay, Carmel, et al.
(Show Context)
Citation Context ...itation analysis (see e.g., [9, 11])—it is well-known that a scientific paper gets the majority of its citations soon after it is published and as time passes, receives fewer citations. Amitay et al. =-=[1]-=- have also identified aspects of temporal locality in Web domains, showing that incorporating hyperlink timestamps into link-based page-ranking algorithms can improve retrieval accuracy. In the area o... |

16 | Community mining from multi-relational networks
- Cai, Shao, et al.
- 2005
(Show Context)
Citation Context ...ics of time-varying attributes in static link structures [13]. There have been some recent efforts to model temporally-varying links to improve automatic discovery of relational communities or groups =-=[4, 15]-=- but this work has not attempted to exploit temporal link information in a classification context. The goal of this work is to improve attribute prediction in dynamic domains by incorporating the infl... |

10 | A Visual Query Language for Relational Knowledge Discovery TITLE2
- Blau, Immerman, et al.
- 2001
(Show Context)
Citation Context ...e set of 16,153 papers published in the time window 19811998, along with their associated references and authors. Figure 4 shows the relational query we used for classification, using modified QGraph =-=[3]-=- notation. The topic and area attributes are supplied to the relational classification model; the summary weights on reference and coauthor links are computed using the year publication dates on paper... |

3 |
A noninformetric analysis of the relationship between citation age and journal productivity
- Egghe
- 2001
(Show Context)
Citation Context ...ional changes in a concise way that evolves smoothly through time and has been utilized for fraud detection analysis. Temporal locality has been studied for many years in citation analysis (see e.g., =-=[9, 11]-=-)—it is well-known that a scientific paper gets the majority of its citations soon after it is published and as time passes, receives fewer citations. Amitay et al. [1] have also identified aspects of... |