The aspect Bernoulli model: multiple causes of presences and absences

Bingham, Ella; Kabán, Ata

doi:10.1007/s10044-007-0096-4

Cited by 25 publications

(33 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the case of MAP, we should expect to see a peak for the optimal value of K, indicating that relevant documents are ranked higher. Furthermore, previous work has suggested that the optimum number of topics will be close to the cardinality of the concepts underlying the data [4,11,12]. In Fig.…”

Section: Document Rankingmentioning

confidence: 91%

“…The PLSA algorithm was written by Peter Gehler. 4 Convergence for PLSA was satisfied after 500 EM iterations or when the log-likelihood change was less than 1e-4. For LDA, following a preliminary experiment, we set the Dirichlet priors to α = 0.1 and β = 0.01 and the number of Gibbs sampling iterations to 1000.…”

Section: Methodsmentioning

confidence: 99%

“…In the merged view, LDA climbs steeply, while PLSA remains distanced from orig in the merged view. PLSA has previously shown poor performance in the factorisation of sparse 0-1 data [4] where this has been attributed to the fact that the zero entries do not contribute to the loglikelihood. This may explain why PLSA shows slightly better performance on the less sparse merged view.…”

Section: Document Rankingmentioning

confidence: 99%

See 2 more Smart Citations

Topic modelling of clickthrough data in image search

Morrison

Tsikrika

Hollink

et al. 2012

Multimed Tools Appl

View full text Add to dashboard Cite

In this paper we explore the benefits of latent variable modelling of clickthrough data in the domain of image retrieval. Clicks in image search logs are regarded as implicit relevance judgements that express both user intent and important relations between selected documents. We posit that clickthrough data contains hidden topics and can be used to infer a lower dimensional latent space that can be subsequently employed to improve various aspects of the retrieval system. We use a subset of a clickthrough corpus from the image search portal of a news agency to evaluate several popular latent variable models in terms of their ability to model topics underlying queries. We demonstrate that latent variable modelling reveals underlying structure in clickthrough data and our results show that computing document similarities in the latent space improves retrieval effectiveness compared to computing similarities in the original query space. These results are compared with baselines using visual and textual features. We show performance substantially

show abstract

Section: Document Rankingmentioning

confidence: 91%

Section: Methodsmentioning

confidence: 99%

Section: Document Rankingmentioning

confidence: 99%

See 1 more Smart Citation

Topic modelling of clickthrough data in image search

Morrison

Tsikrika

Hollink

et al. 2012

Multimed Tools Appl

View full text Add to dashboard Cite

show abstract

“…Methods using normal algebra [19] or probabilistic modeling [2,18], for example, have been proposed. The characteristics and behaviour of such methods are very different to Boolean matrix factorization, though.…”

Section: Related Workmentioning

confidence: 99%

Fully dynamic quasi-biclique edge covers via Boolean matrix factorizations

Miettinen

2013

Proceedings of the Workshop on Dynamic Networks Management and Mining

View full text Add to dashboard Cite

An important way of summarizing a bipartite graph is to give a set of (quasi-) bicliques that contain (almost) all of its edges. These quasi-bicliques are somewhat similar to clustering of the nodes, giving sets of similar nodes. Unlike clustering, however, the quasi-bicliques are not required to partition the nodes, allowing greater flexibility when creating them. When we identify the bipartite graph with its bi-adjacency matrix, the problem of finding these quasi-bicliques turns into the problem of finding the Boolean matrix factorization of the bi-adjacency matrix -a problem that has received increasing research interest in data mining in recent years. But many real-world graphs are dynamic and evolve over time. How can we update our bicliques without having to re-compute them from the scratch?An algorithm was recently proposed for this task (Miettinen, ICMD 2012). The algorithm, however, is only able to handle the case where the new 1s are added to the matrix -it cannot handle the removal of existing 1s. Furthermore, the algorithm cannot adjust the rank of the factorization. This paper extends said algorithm with the capability of working in fully dynamic setting (with both additions and deletions) and with capability of adjusting its rank dynamically, as well. The behaviour and performance of the algorithm is studied in experiments conducted with both real-world and synthetic data.

show abstract

“…Methods using normal algebra [13] or probabilistic modeling [14], [15], for example, have been proposed. The characteristics and behaviour of such methods are very different to Boolean matrix factorization, though.…”

Section: Related Workmentioning

confidence: 99%

Dynamic Boolean Matrix Factorizations

Miettinen

2012

2012 IEEE 12th International Conference on Data Mining

View full text Add to dashboard Cite

Abstract-Boolean matrix factorization is a method to decompose a binary matrix into two binary factor matrices. Akin to other matrix factorizations, the factor matrices can be used for various data analysis tasks. Many (if not most) real-world data sets are dynamic, though, meaning that new information is recorded over time. Incorporating this new information into the factorization can require a re-computation of the factorization -something we cannot do if we want to keep our factorization up-to-date after each update.This paper proposes a method to dynamically update the Boolean matrix factorization when new data is added to the data base. This method is extended with a mechanism to improve the factorization with a trade-off in speed of computation. The method is tested with a number of real-world and synthetic data sets including studying its efficiency against off-line methods. The results show that with good initialization the proposed online and dynamic methods can beat the stateof-the-art offline Boolean matrix factorization algorithms.

show abstract

The aspect Bernoulli model: multiple causes of presences and absences

Abstract: We present a probabilistic multiple cause model for the analysis of binary (0-1) data. A distinctive feature of the aspect Bernoulli (AB) model is its

Cited by 25 publications

References 38 publications

Topic modelling of clickthrough data in image search

Topic modelling of clickthrough data in image search

Fully dynamic quasi-biclique edge covers via Boolean matrix factorizations

Dynamic Boolean Matrix Factorizations

Contact Info

Product

Resources

About