1、An Overview of Bayesian Network-based Retrieval Models,Juan Manuel Fernndez Luna Departamento de InformticaUniversidad de Janjmflunaujaen.es,Department of Computing Science, University of GlasgowOctober, 21th - 2002,Bayesian Network-based Retrieval Models,2,Layout,Introduction Introduction to Belief
2、 Networks Bayesian Network-based IR Models Inference Network Model Belief Network Model Bayesian Network Retrieval Model Relevance Feedback Other applications Bibliography,Bayesian Network-based Retrieval Models,3,Introduction,Query and document characterizations are incomplete. The query is a vague
3、 description of the users information need. Computing relevance degree: 1 and 2 +A) different representations that a concept may have, B) these concepts are not independent among them.,Information Retrieval Uncertain process,Bayesian Network-based Retrieval Models,4,Introduction,Probabilistic models
4、 tried to overcome these problems,Researchers focused their attention on Belief networks in order to apply them to IR because:,They show a high performance in actual problems characterised by uncertainty.,Bayesian Network-based Retrieval Models,5,Introduction to Belief Networks,Graphical models able
5、 to represent and efficiently manipulate n-dimensional probability distributions.The knowledge obtained from a problem is encoded in a Belief network by means of the quantitative and qualitative componets:,Bayesian Network-based Retrieval Models,6,Introduction to Belief Networks,Qualitative part: Di
6、rected Acyclic Graph.G=(V,E): V (Nodes) Random variables, and E (Arcs) (In)dependence relationships.,Bayesian Network-based Retrieval Models,7,Introduction to Belief Networks,Quantitative part A set of conditional distributions: Drawn from the graph structure, representing the strength of the relati
7、onships, stored in each node.,Belief Network Bayesian Network (Conditional probability distributions),Bayesian Network-based Retrieval Models,8,Introduction to Belief Networks,Bayesian Network-based Retrieval Models,9,Introduction to Belief Networks,Taking into account these (in)dependences, the joi
8、nt probability distribution could be restored from the network:,Pa(Xi) being the set of parents of the variable Xi. This previous expression implies an important saving in the storage space.,Bayesian Network-based Retrieval Models,10,Introduction to Belief Networks,Construction: Manual, using an exp
9、erts knowledge. Automatic, by means of a learning algorithm.,Inference: Given a set of evidences, E, to obtain the probability with which a variable can take a certain value.p(S=T | W=T)=0.430, p(R=T| W=T)= 0.708,Bayesian Network-based Retrieval Models,11,Bayesian Network-based IR Models,Inference N
10、etwork Model Belief Network Model Peter Bruzas Index Belief Expressions Maria Indrawan et al.s Model Bayesian Network Retrieval Model,Bayesian Network-based Retrieval Models,12,inn,Link Matrices,Inference: Instantiating each document, dj, and computing p(inn | dj).,Inference Network Model,Bayesian N
11、etwork-based Retrieval Models,13,Belief Network Model,Q,2M assigments unfeasible Probabilities are defined in such a way that only one configuration is evaluated,Bayesian Network-based Retrieval Models,14,Bayesian Network Retrieval Model,There are strong relationships among a document and the terms
12、that index it. Document relationships are only present by means of the terms that index them. Documents are conditional independent given the terms by which they were indexed.,Guidelines to build the BNR Model:,Bayesian Network-based Retrieval Models,15,Bayesian Network Retrieval Model,Ti ti, ti,Dj
13、dj, dj,Bayesian Network-based Retrieval Models,16,Bayesian Network Retrieval Model,All the terms are independent among them: Simple Bayesian Network Retrieval Model,Bayesian Network-based Retrieval Models,17,Bayesian Network Retrieval Model,Probability Distributions: Term nodes: p(tj)=1/M, p(tj)=1-p
14、(tj) Document nodes: p(Dj | Pa(Dj), Dj,But. If a document has been indexed by 30 terms, we need to estimate and store 230 probabilities.,Problem!,Bayesian Network-based Retrieval Models,18,Bayesian Network Retrieval Model,Solution:,Probability functions,pa(Dj) being a configuration of the parents of
15、 Dj.,Bayesian Network-based Retrieval Models,19,Bayesian Network Retrieval Model,Retrieval:,Instantiate TQ Q to Relevant. Run a propagation algorithm in the network. Rank the documents according p(dj | Q), Dj,Problem:,Great amount of nodes and existing cycles in the graph,General purpose propagation
16、 algorithms cant be applied due to efficiency considerations.,Bayesian Network-based Retrieval Models,20,Bayesian Network Retrieval Model,Solution: Taking advantage of: The kind of probability function used, and The topology.Propagation is substituted by,Evaluation of the probability function in eac
17、h document node,Bayesian Network-based Retrieval Models,21,Bayesian Network Retrieval Model,Result: An efficient and exact propagation.,Including Query term frequencies:,Bayesian Network-based Retrieval Models,22,Bayesian Network Retrieval Model,Removing the term independency restricction: We are in
18、terested in representing the main relationships among terms in the collection.,Term subnetwork Polytree,Why? There is a set of efficient learning and propagation algorithms available for this topology.,Bayesian Network-based Retrieval Models,23,Bayesian Network Retrieval Model,Bayesian Network-based
19、 Retrieval Models,24,Bayesian Network Retrieval Model,Probability distributions:Marginal Distributions (root term nodes):,(M being the number of terms in the collection),Bayesian Network-based Retrieval Models,25,Bayesian Network Retrieval Model,Conditional Distributions (document nodes): Probabilit
20、y functions,Conditional Distributions (term nodes with parents): (based on Jaccards coefficient),Bayesian Network-based Retrieval Models,26,Bayesian Network Retrieval Model,Retrieval: TqQ Relevant p(dj|Q)?,But. Due to the complexity of the whole network we can not run an exact propagation algorithm.
21、,Solution: PROPAGATION + EVALUATION,Bayesian Network-based Retrieval Models,27,Bayesian Network Retrieval Model,Propagation:Running the exact Pearls propagation algorithm in the polytree (term subnetwork), p(ti|Q), Ti, are computed.Evaluation:Evaluation of a probability function in the Document Subn
22、etwork, computing p(dj|Q), Dj, incorporating p(ti|Q).,Bayesian Network-based Retrieval Models,28,Bayesian Network Retrieval Model,Given a document, Dj:Compute p(dj|di), Di. Select those documents with greatest probability of relevance with respect to Dj. Link Dj with all these documents.,Adding docu
23、ment relationships,Bayesian Network-based Retrieval Models,29,Bayesian Network Retrieval Model,But. Instead of linking the documents in the document subnetwork.,Bayesian Network-based Retrieval Models,30,Bayesian Network Retrieval Model,We dont have to restimate probability distributions in the docu
24、ment nodes. Propagation: Evaluation of a probability function in the second document layer Efficiency.,Advantages of this topology:,Bayesian Network-based Retrieval Models,31,Bayesian Network Retrieval Model,Compute p(dj|Q), Dj (1st document layer) Compute p(dj|Q), Dj (2nd document layer),Where Sj i
25、s a normalising constant,Retrieval?,Bayesian Network-based Retrieval Models,32,Bayesian Network Retrieval Model,Reducing the propagation time in the Term Subnetwork: Representing only the best relationships among terms. Modifying Pearls propagation algorithm. Changing the Term subnetwork topology.,B
26、ayesian Network-based Retrieval Models,33,Bayesian Network Retrieval Model,1. Representing only the best term relationshipsProblems:Automatically learning the relationships among terms could imply that some relationships are not strong enough. Retrieval effectiveness could be damagedIf the number of
27、 terms is very high, the learning stage could be time-consuming.,Bayesian Network-based Retrieval Models,34,Bayesian Network Retrieval Model,Solution:,Selection of best terms,Collection,Bayesian Network-based Retrieval Models,35,Bayesian Network Retrieval Model,Advantages: Reduction of learning time
28、 Representation of the best relationships among terms Faster propagation.,Bayesian Network-based Retrieval Models,36,Bayesian Network Retrieval Model,Classification algorithm: K-means, with Euclidean distance Objects: Terms Attributes: Term discrimination value (tdv) Inverse Document Frequency (idf)
29、 Classes: Good terms: higher tdv, and medium-high idf. Rest of the terms.,Bayesian Network-based Retrieval Models,37,Bayesian Network Retrieval Model,2. Modifying Pearls algorithm.In large polytrees, the belief of a great number of terms, those furthest from query terms, will not be updated after pr
30、opagating.So.Why is the propagation algorithm still running?,Bayesian Network-based Retrieval Models,38,Bayesian Network Retrieval Model,Radial Propagation,r=2,Bayesian Network-based Retrieval Models,39,Bayesian Network Retrieval Model,Linear Propagation,Bayesian Network-based Retrieval Models,40,Ba
31、yesian Network Retrieval Model,3. Changing the Term Subnetwork topology.In certain cases, the polytree topology of the Term subnetwork, even using the term selection approach, could not be very appropriate.,An alternative topology:,Two term layers,Preserving accuracy of term relationships represente
32、d in the graph. Providing an efficient inference mechanism.,Bayesian Network-based Retrieval Models,41,Bayesian Network Retrieval Model,Bayesian Network-based Retrieval Models,42,Bayesian Network Retrieval Model,Relationships ara captured using the coocurrences among terms.The probability of relevan
33、ce in the second term layer is computed by means of:,Bayesian Network-based Retrieval Models,43,Relevance Feedback in B.N. Models,Inference and Belief Network Models: Modifying link matrices and adding new links (and also new document nodes in the second).Bayesian Network Model: Inclusion of new evi
34、dences from the inspection of the document ranking using partial evidences. (Advantage: neither graph structure modification nor probability matrix re-estimation).,Bayesian Network-based Retrieval Models,44,Other applications:,Indexing Hypertext User profiling WWW Structured documents Image retrieva
35、l Document classification Filtering,Bayesian Network-based Retrieval Models,45,Bibliography,Bruza, P. Fernndez-Luna, J.M. & Huete, J.F. (2002). A layered Bayesian Network Model for Document Retrieval. Proceedings of the ECIR2002 Colloquium. Lecture notes in Computer Science, 2291, 169 182.,Bayesian
36、Network-based Retrieval Models,46,Bibliography,Luis M. de Campos, Juan M. Fernndez-Luna, Juan F. Huete. Reducing term to term relationships in an extended Bayesian network retrieval model. Proceedings of the Ninth International IPMU Conference (Information Processing and Mangement of Uncertainty in
37、Knowledge-based Systems) Conference, Vol. 2, 1195-1202 (ISBN Vol. 2: 2-9516453-2-5), 2002. ESIA Universit de Savoie (Editor). Luis M. de Campos, Juan M. Fernndez-Luna, Juan F. Huete. Two terms layer: An alternative topology for representing term relationships in the Bayesian Network Retrieval Model.
38、 Electronic Proceeding of the Seventh Online World Conference on Soft Computing in Industrial Applications (wsc7.ugr.es). Luis M. de Campos, Juan M. Fernndez-Luna, Juan F. Huete. Reducing Propagation Effort in Large Polytree: An application to Information Retrieval. To appear in Proceedings of the W
39、orkshop on Probabilistic and Graphical Models. Cuenca (SPAIN), 2002. Crestani, F., Lalmas, M., van Rijsbergen, C.J., Campbell, L. (1998). Is this Document Relevant? Probably: A Survey of Probabilistic Models in Information Retrieval. Computing Survey. 30(4). 528-552.,Bayesian Network-based Retrieval
40、 Models,47,Bibliography,Fernndez-Luna, J.M. (2001). Modelos de Recuperacin de Informacin basados en Redes de Creencia. Ph.D. Thesis (in Spanish). University of Granada. Frisse M. & Cousins, S.B. (1989). Information Retrieval from Hypertext: Update on the Dynamic Medical Handbook Project. Proceedings
41、 of the Hypertext89 Conference. 199-212. Ghazfan , D., Indrawan, M. & Srinivasan, B. (1996). Towards meaningful Bayesian networks. IPMU96 Conference. 841-846. Haines, D. & Croft W.B. (1983). Relevance Feedback and Inference Networks. 20th ACM-SIGIR Conference. 119-128. Pearl, J. (1988). Probabilisti
42、c Reasoning in Intelligent Systems: Networks of Plausible Inference. Morgan and Kaufmann. San Mateo, California. Reis, I. (2000). Bayesian Networks for Information Retrieval. Ph.D. Thesis. Universidad Federal de Minas Gerais. van Rijsbergen, C.J. (1971). Information Retrieval. 2nd Edition. Butter Wo
43、rths. van Rijsbergen, C.J., Harper, D.J., & Porter, M.F. (1981). The selection of good search terms. Information Processing & Management. 17, 77-91.,Bayesian Network-based Retrieval Models,48,Bibliography,Sahami, M. (1998). Using Machine Learning to Improve Information Access. Ph.D. Thesis. Stanford
44、 University. Savoy, J. & Desbois, D. (1991). Information Retrieval in Hypertext Systems: An Approach using Bayesian Networks. Electronic Publishing. 42(2), 87-108. Turtle, H.R., & Croft, W.B. (1991). Evaluation of an Inference Network-based Retrieval Model. Information Systems. 9(3), 189-224. Turtle
45、, H.R., & Croft, W.B. (1997). Uncertainty in Information Systems. In Uncertainty Management in Information System: From needs to solutions. Kluver Academic. 189-224. Tzeras K. & Hartman, S. (1993). Automatic Indexing Based on Bayesian Inference Netoworks. 16th ACM-SIGIR Conference. 22-35.,The end.,Thank you very much,
