UMR CNRS 7253

Antoine Bordes
Antoine Bordes
Antoine Bordes
Antoine Bordes
Antoine Bordes

Site Tools


en:tensor_fact

Tensor Factorization Resources

Papers (and code)

[1] Many papers by Andrei Cichoki. webpage

[2] Non-Negative Tensor Factorization with Applications to Statistics and Computer Vision. ICML05. paper

[3] Learning Systems of Concepts with an Infinite Relational Model. AAAI06.  paper code (C)

[4] Relational Learning via Collective Matrix Factorization. KDD08. paper code (Matlab)

[5] TripleRank: Ranking Semantic Web Data By Tensor Decomposition. ISWC09.  paper

[6] Tensor Sparse Coding for Region Covariances. ECCV 2010.  paper

[7] Modelling Relational Data using Bayesian Clustered Tensor Factorization. NIPS10.  paper

[8] Multichannel Nonnegative Tensor Factorization With Structured Constraints For User-Guided Audio Source Separation. ICASSP11.  paper

[9] A Three-Way Model for Collective Learning on Multi-Relational Data. ICML11.  paper code (Python)

[10] Sparse Non-Negative Tensor Factorization Using Columnwise Coordinate Descent. Pattern Recognition Jan12.  paper

[11] Statistical predicate invention. ICML07. paper

Data

Benchmarks

Three datasets used in [3,7,9,11]:

  • UMLS: Entities: 135, Relation types: 49, Sparsity: 0.7% (i.e. non-zeros) umls.dat.gz
  • Kinships: Entities: 104, Relation types: 26, Sparsity: 3.84% (i.e. non-zeros) kinships.dat.gz
  • Cora: Entities: 2497, Relation types: 7, Sparsity: 0.32% (i.e. non-zeros) cora.dat.gz

Data is in gzipped python cPickle format, a dictionary contains the tensor (numpy array) and lists of entity and relation names (not for kinships)
In papers [3,7,9,11], people perform 10-fold cross-validation by using (subject, predicate, object) triples as the statistical unit.

Homemade data

* Text: word triples extracted from 50k articles of Wikipedia following the scheme Subject-Verb-Direct Object s-v-cod_nouns_only.dat.gz.

* WordNet: preprocessed version of WordNet3.0 wordnet-3.0.tar.gz


User Tools