Citation Context

...lgorithm is able to learn from very few data, and compares favorably to similar state of the art approaches. 1 Introduction The class of regular languages is not learnable from positive examples only =-=[8]-=-. One solution for this negative result is defining learnable subclasses. Examples of this approach in the case of string languages are k-reversible languages[2], kcontextual languages[14] and k-testa...

Citation Context

...ement has the highest bounds on its count. For k>1, the refinement is the language [F.k[l]+1,l],howeverfork =1,also[1,l+1]is a refinement. Example 3. Given the data in Figure 1, the languages [1, 5], =-=[4, 3]-=- and [2, 5] are candidates for refinement. Although [4, 3] has the highest count, its refinement [5, 3] has a count bounded by 33 while both refinements of [1, 5] have a count bounded by 48, hence the...

Citation Context

...ion(IE) from structured documents (HTML, XML), aims at extracting specific information from structurally similar documents. Often it is referred to as wrapper induction. Some examples can be found in =-=[12, 15, 20, 3, 10, 11]-=-. In [10] tree automata are learned from positive examples only. Each example is a document, in which one of the elements that is desired, is marked with a special marker. When used to extract informa...

Citation Context

...ion(IE) from structured documents (HTML, XML), aims at extracting specific information from structurally similar documents. Often it is referred to as wrapper induction. Some examples can be found in =-=[12, 15, 20, 3, 10, 11]-=-. In [10] tree automata are learned from positive examples only. Each example is a document, in which one of the elements that is desired, is marked with a special marker. When used to extract informa...

Citation Context

...hat the node containing ’title1’ is marked. Some examples of potential solutions, given Pos are [1; 2] = L(1;2)( ( @:T , @:T a ) ), [1; 3] = L(1;3)( 8>>< >>:@:T , @:T a , @:T a b 9>>= >>;), [1; 4] = L=-=(1;4)-=-( 8>>>>>>< >>>>>>: @:T , @:T a , @:T a b , @:T a b p 9>>>>>>= >>>>>>; ), and [2; 4] = L(2;4)( 8>>>>>>< >>>>>>: @:T , @:T a , @:T a b , @:T a @ b a p 9>>>>>>= >>>>>>; ). Defining S as the set of trees ...

Citation Context

... learnable from positive examples only [8]. One solution for this negative result is defining learnable subclasses. Examples of this approach in the case of string languages are k-reversible languages=-=[2]-=-, kcontextual languages[14] and k-testable languages[7]. Since k-contextual and ktestable languages are equivalent[1], we will refer to them as k-local languages. In the case of tree languages, k-test...

Citation Context

...ion(IE) from structured documents (HTML, XML), aims at extracting specific information from structurally similar documents. Often it is referred to as wrapper induction. Some examples can be found in =-=[12, 15, 20, 3, 10, 11]-=-. In [10] tree automata are learned from positive examples only. Each example is a document, in which one of the elements that is desired, is marked with a special marker. When used to extract informa...

Citation Context

...the parameter that yields the best precision and recall after cross-validation on the training set. The training and test sets used in [10], to compare their algorithm to some string based algorithms =-=[4, 16, 5]-=- are the same as in [5]. Each training set is made up of 5 documents, with all the desired elements marked. Resulting in some hundreds of examples. One can argue that this goes beyond learning from po...

Citation Context

...the parameter that yields the best precision and recall after cross-validation on the training set. The training and test sets used in [10], to compare their algorithm to some string based algorithms =-=[4, 16, 5]-=- are the same as in [5]. Each training set is made up of 5 documents, with all the desired elements marked. Resulting in some hundreds of examples. One can argue that this goes beyond learning from po...

Citation Context

...the parameter that yields the best precision and recall after cross-validation on the training set. The training and test sets used in [10], to compare their algorithm to some string based algorithms =-=[4, 16, 5]-=- are the same as in [5]. Each training set is made up of 5 documents, with all the desired elements marked. Resulting in some hundreds of examples. One can argue that this goes beyond learning from po...

Citation Context

...n for this negative result is defining learnable subclasses. Examples of this approach in the case of string languages are k-reversible languages[2], kcontextual languages[14] and k-testable languages=-=[7]-=-. Since k-contextual and ktestable languages are equivalent[1], we will refer to them as k-local languages. In the case of tree languages, k-testable tree languages are introduced in [6, 9], as an ext...

Citation Context

...e latter, hence the name (k,l)contextual tree languages. We remark that when we mention k-testable languages we mean locally testable in the strict sense, and not the more general definition given in =-=[13]-=-. In this notion, G will be a set of sets of forks, while the language defined by G will contain every tree, whose forks are a subset of at least one of the sets of forks in G. This Approach is more e...

Citation Context

...xamples of this approach in the case of string languages are k-reversible languages[2], kcontextual languages[14] and k-testable languages[7]. Since k-contextual and ktestable languages are equivalent=-=[1]-=-, we will refer to them as k-local languages. In the case of tree languages, k-testable tree languages are introduced in [6, 9], as an extension of the k-contextual and k-testable string languages, wi...

Citation Context

...xamples only [8]. One solution for this negative result is defining learnable subclasses. Examples of this approach in the case of string languages are k-reversible languages[2], kcontextual languages=-=[14]-=- and k-testable languages[7]. Since k-contextual and ktestable languages are equivalent[1], we will refer to them as k-local languages. In the case of tree languages, k-testable tree languages are int...

Citation Context

...r nodes can be distinguished by the structure alone. In that case the set of contexts can be empty. The set of contexts is learned automatically from the training set. We use the same algorithm as in =-=[10, 11]-=-. This algorithm searches strings that occur at the same distance from each marked node in the training set. The distance we use is defined such that the distance from a node to its parent is one; to ...

Citation Context

...ement has the highest bounds on its count. For k>1, the refinement is the language [F.k[l]+1,l],howeverfork =1,also[1,l+1]is a refinement. Example 3. Given the data in Figure 1, the languages [1, 5], =-=[4, 3]-=- and [2, 5] are candidates for refinement. Although [4, 3] has the highest count, its refinement [5, 3] has a count bounded by 33 while both refinements of [1, 5] have a count bounded by 48, hence the...

Citation Context

...ion(IE) from structured documents (HTML, XML), aims at extracting specific information from structurally similar documents. Often it is referred to as wrapper induction. Some examples can be found in =-=[12, 15, 20, 3, 10, 11]-=-. In [10] tree automata are learned from positive examples only. Each example is a document, in which one of the elements that is desired, is marked with a special marker. When used to extract informa...

Citation Context

...o space restrictions we can not describe this approach in detail here. We present this algorithm in [17]. 6 Experiments We evaluate our approach on the WIEN data sets (available at the RISE repository=-=[19]-=-). Some of the sets are discarded since they do not include labels. The other sets are split in different tasks: a task aiming at the extraction of a n-tuple is splitted in n extraction tasks. We refe...

Citation Context

...able languages[7]. Since k-contextual and ktestable languages are equivalent[1], we will refer to them as k-local languages. In the case of tree languages, k-testable tree languages are introduced in =-=[6, 9]-=-, as an extension of the k-contextual and k-testable string languages, with probabilistic extensions in [18], and in [10] local unranked tree automata are introduced. In Section 3 we define a new subc...

Citation Context

...able languages[7]. Since k-contextual and ktestable languages are equivalent[1], we will refer to them as k-local languages. In the case of tree languages, k-testable tree languages are introduced in =-=[6, 9]-=-, as an extension of the k-contextual and k-testable string languages, with probabilistic extensions in [18], and in [10] local unranked tree automata are introduced. In Section 3 we define a new subc...

Citation Context

.... In the case of tree languages, k-testable tree languages are introduced in [6, 9], as an extension of the k-contextual and k-testable string languages, with probabilistic extensions in [18], and in =-=[10]-=- local unranked tree automata are introduced. In Section 3 we define a new subclass for the class of regular unranked tree languages, called (k,l)-contextual tree languages and we present an inference...

Citation Context

...two points, where the number of extractions stays constant. This is detected by the heuristic. Due to space restrictions we can not describe this approach in detail here. We present this algorithm in =-=[17]-=-. 6 Experiments We evaluate our approach on the WIEN data sets (available at the RISE repository[19]). Some of the sets are discarded since they do not include labels. The other sets are split in diff...

Citation Context

...cal languages. In the case of tree languages, k-testable tree languages are introduced in [6, 9], as an extension of the k-contextual and k-testable string languages, with probabilistic extensions in =-=[18]-=-, and in [10] local unranked tree automata are introduced. In Section 3 we define a new subclass for the class of regular unranked tree languages, called (k,l)-contextual tree languages and we present...