Learning Stochastic Tree Edit Distance

Abstract

Abstract. Trees provide a suited structural representation to deal with complex tasks such as web information extraction, RNA secondary structure prediction, or conversion of tree structured documents. In this context, many applications require the calculation of similarities between tree pairs. The most studied distance is likely the tree edit distance (ED) for which improvements in terms of complexity have been achieved during the last decade. However, this classic ED usually uses a priori fixed edit costs which are often difficult to tune, that leaves little room for tackling complex problems. In this paper, we focus on the learning of a stochastic tree ED. We use an adaptation of the ExpectationMaximization algorithm for learning the primitive edit costs. We carried out series of experiments that confirm the interest to learn a tree ED rather than a priori imposing edit costs.

Learning Metrics Between Tree Structured Data: Application to Image Recognition

“…These probabilities are the parameters of a generative model describing a joint distribution over (input,output) pairs of trees. In [8], we proposed a first solution to this problem, in a restrictive case of tree edit distance, when a deletion (resp. an insertion) implies the removal (resp.…”

“…Recently, the Pascal network of excellence funded a pump priming project on the learning of a stochastic tree ED for musical recognition. A first learning algorithm, where deletions and insertions only concern entire subtrees, has been proposed in [8]. Although this type of tree ED is costless from an algorithmic standpoint (quadratic complexity [9] rather than a polynomial complexity of order 4 for a more general case [10]), it is not the most used in the literature because of a clear loss of generality.…”

Learning probabilistic models of tree edit distance

“…A parametric approach has also been presented in [6] in the 2 A c c e p t e d m a n u s c r i p t context of graph ED, where each edit operation is modeled by a Gaussian Mixture Density. With the exception of our preliminary work [7], as far as we know, no method has been proposed to directly learn edit costs for a stochastic tree ED. The aim of this paper is to fill this gap by a non parametric stochastic method specifically adapted to trees.…”

SEDiL: Software for Edit Distance Learning

“…This joint work has lead to publications in the previous conferences ECML'06 [7] and ECML'07 [4], and in Pattern Recognition [3,8]. This research has also received funding from the RedEx PASCAL in the form of a pump-priming project in 2007.…”

“…However, in many real world applications, such a strategy clearly appears insufficient. To overcome this drawback and to capture background knowledge, supervised learning has been used during the last few years for learning the parameters of edit distances [1,2,3,4,7,8,9], often by maximizing the likelihood of a learning set. The learned models usually take the form of state machines such as stochastic transducers or probabilistic automata.…”