Distant Supervision for Relation Extraction with Matrix Completion

Abstract

The essence of distantly supervised relation extraction is that it is an incomplete multi-label classification problem with sparse and noisy features. To tackle the sparsity and noise challenges, we propose solving the classification problem using matrix completion on factorized matrix of minimized rank. We formulate relation classification as completing the unknown labels of testing items (entity pairs) in a sparse matrix that concatenates training and testing textual features with training labels. Our algorithmic framework is based on the assumption that the rank of item-byfeature and item-by-label joint matrix is low. We apply two optimization models to recover the underlying low-rank matrix leveraging the sparsity of feature-label matrix. The matrix completion problem is then solved by the fixed point continuation (FPC) algorithm, which can find the global optimum. Experiments on two widely used datasets with different dimensions of textual features demonstrate that our low-rank matrix completion approach significantly outperforms the baseline and the state-of-the-art methods.

Extracting Databases from Dark Data with DeepDive

“…DeepDive 's model of KBC is motivated by the recent attempts of using machine learning-based technique for KBC [3,4,24,38,46,52,56] and the line of research that aims to improve the quality of a specific component of KBC system [7,12,15,21,26,27,31–33,35,39,42,47,48,51,53,54]. When designing DeepDive , we used these systems as test cases to justify the generality of our framework.…”

“…When designing DeepDive , we used these systems as test cases to justify the generality of our framework. In fact, we find that DeepDive is able to model thirteen of these popular KBC systems [15,23,26,27,31–33,35,39,42,51,53,54]. …”

Distant Supervision for Relation Extraction with Ranking-Based Methods

“…References [14,15] are extensions to [13] with additional layers or add-in penalty factors. Recent work also included the embedding models that transfer the relation extraction problem into a translation model like h`r « t [20][21][22], and the probability matrix factorization (PMF) models from [11,23] in which training and testing are carried out jointly. Besides, Fan et al [24] presented a novel framework by integrating active learning and weakly supervised learning.…”

“…Ritter et al [15] added two penalty factors to model the missing of texts and the missing of KB, and they also considered some additional information such as popularity of entities. Angeli et al [23] added a bias factor b to their PMF algorithm to model the noise. Xiang et al [30] computed the value of two types of biases to model the correctness and incorrectness for each group-level label.…”

ArabRelat: Arabic Relation Extraction using Distant Supervision

“…Their system shows good accuracy for extracting 52 types of relations which suggests the applicability of distant supervision for general RE. In (Fan et al, 2014), the distantly supervised relation extraction was solved as a matrix completion problem. (Yao et al, 2012) uses an unsupervised approach to handle the problem of Polysemy where the same pattern can have several meanings.…”