CN103294791A - Extensible markup language pattern matching method - Google Patents

Abstract

The invention discloses an Extensible Markup Language pattern matching method, which is used to solve the problems of the prior art in terms of pattern representation, discovery of complex matching, matching efficiency, etc. The steps include: inputting an Extensible Markup Language pattern; constructing a pattern tree; Construct a sequence structure; perform name, data type and cardinality constraint matching on all elements to obtain the language similarity value of all element pairs; for complex elements, perform child similarity, leaf similarity, sibling similarity and ancestor similarity matching to obtain The structure and overall similarity of complex element pairs, filter to find matching complex element pairs; for each matched complex element pair, find the corresponding atomic set for each element, and apply non-complex element structure matching to the elements in the atomic set Method, calculate the structure and overall similarity value of atomic elements, filter to find matching element pairs; output all matching element pairs. The invention is fully automated, and the matching efficiency is improved under the premise of ensuring the matching quality.

Description

An Extensible Markup Language Pattern Matching Method

technical field

The invention belongs to the technical field of communication, and further relates to an extensible markup language (eXtensible Markup Language XML) pattern matching method in the technical field of data processing. The present invention can automatically perform Extensible Markup Language pattern matching on two input Extensible Markup Language pattern documents according to the name and structure information of the patterns, find out the mapping between all similar elements in the two documents, and use it to determine different Extensible Markup Language pattern documents. Similarity between markup language data.

Background technique

With the development of Internet, Extensible Markup Language came into being and became the standard of data representation, data analysis and data exchange in the network. Due to the flexibility of extensible markup language data description, the number and scale of extensible markup language documents are increasing day by day, how to efficiently manage large-scale extensible markup language data and integrate a large number of extensible markup language data resources has become very important. Therefore, XML pattern matching technology for identifying element consistency between XML patterns has become a research hotspot.

The XML pattern matching takes two XML patterns as input, and uses different similarity value calculation methods to obtain a mapping between the two XML patterns. Extensible Markup Language pattern matching plays an important role in data sharing applications: in data integration, it can be used to identify and mark the internal schema relationship between multiple schemas; in data warehouse, it can map a data resource to Warehouse mode; in e-commerce, it can realize message mapping between different extensible markup language formats; in semantic network, it can be used to establish semantic correspondence between ontology concepts of different websites; in data migration , it can migrate legacy data from multiple resources into a new data; in data transformation, it can map a source object to a target object; in XML data clusters, it can be used to determine different Extensible Markup Language Semantic similarity between data.

Early pattern matching was usually done manually, and specifying pattern matches manually was a time-consuming, error-prone, and expensive process. At present, a large number of automatic pattern matching algorithms and matching systems have been proposed, such as LSD (Learning Source Descriptions), Cupid, COMA (COmbination of MAtting algorithms), Similarity Flooding, AgreementMaker, ASMOV (Automated Semantic Matching of Ontologies with Verification), OII Harmony, etc. Although a large number of existing pattern matching algorithms and systems have realized semi-automatic or fully automatic pattern matching, and the matching quality is also high, there are still many defects in the automatic pattern matching of XML. First, most matching algorithms only find simple matches (1:1 matches), and there are only a few ways to find complex matches. Secondly, most matching algorithms mainly consider the overall similarity between patterns, ignoring the similarity between independent elements, while the element similarity research of XML patterns can well support semi-automatic and labor-intensive activities, such as Extensible Markup Language schema integration. Finally, and most importantly, most matching systems only focus on matching quality, ignoring matching efficiency, making the matching efficiency of large-scale data extremely low. For example, in element name matching, external dictionaries (WordNet, etc.) are used to perform semantic similarity matching. Although this improves the accuracy of name matching, frequent word searches will greatly increase the matching time.

The patent application "XML document structure and semantic similarity calculation method based on extended adjacency matrix" (application number 201010118060.5 application publication number CN101799825A) filed by Nankai University discloses an Extensible Markup Language document structure and semantic similarity based on extended adjacency matrix Calculation method. The specific steps of the method are as follows: first, input the XML document, and encode the XML document tree; second, for the encoded two documents, generate a schema document node list and a data source document node list ;Thirdly, based on the generated two-node list, generate a pattern extended adjacency matrix and a data source extended adjacency matrix; fourthly, use the cosine theorem to calculate the distance between the two adjacency matrices, and obtain the similarity value of two XML documents. The disadvantages of this patent application are: firstly, this method only measures the similarity of patterns at the document level, but does not go deep into the finer granularity of document elements, which makes this method cannot be used for scalable markup based In data processing applications of mapping between language model elements; secondly, this method only uses limited information such as node labels, node hierarchy information, node encoding information and node parent node information as the basis for measuring the similarity of nodes, which may be used in A large error occurs in the calculation of the similarity value.

The patent application "TOOLS AND METHODS FOR SEMI-AUTOMATIC SCHEMAMATCHING" (application number US20060491167 application publication number US2008021912A1) filed by MITER CORP [US] discloses a semi-automatic Extensible Markup Language pattern matching tool and method. The specific steps are: first , enter the source and target XML patterns to be matched; second, graphically display the source and target XML patterns; third, ask the user whether he wants to manually specify some matches; if yes, let the user select the Manually specify certain matches on the displayed XML schema graph; otherwise, perform steps four through seven; fourth, perform language preprocessing on the source and target XML schemas, and vote by a set of matches Fifth, the voting combiner combines all the scores obtained in the fourth step to generate a matching matrix; sixth, adding structural information to further adjust the score; seventh, graphically display the matching results; eighth , repeat steps 3 to 7 until all matching scores are calculated. The disadvantage of this patent application is that although the semi-automatic Extensible Markup Language pattern matching method can improve the quality of pattern matching to some extent under manual intervention, it can only be adapted to small-scale data processing, and for relatively For large-scale XML schema documents, manually specifying pattern matches is a tedious, time-consuming and error-prone process, thus, this may limit the scale of XML schema data that the method can handle.

The patent application "METHODS AND SYSTEMS FOR MODEL MATCHING" (application number US20010028912 application publication number US2003120651A1) proposed by MICROSOFT CORP [US] discloses a method and system for model or pattern matching. The specific steps are: first, input the Source and target XML schemas; second, document object model parsing for the two input XML schemas; third, transform the generated document object model into a common object model; fourth, perform root attribute matching Match the structure; fifth, return the matching result. The disadvantage of this patent application is that structural matching mainly utilizes the similarity of leaf nodes, but does not fully consider other structural related information of nodes such as children and brothers, which may reduce the quality of structural matching; secondly, in structural matching, In order to improve the structure matching effect, it is necessary to repeatedly traverse the subtrees and update the similar value of the nodes for multiple passes, which can improve the matching accuracy to a certain extent, but it may cause great problems when dealing with large-scale Extensible Markup Language patterns. System overhead, thus reducing the matching efficiency.

Contents of the invention

The purpose of the present invention is to address the deficiencies in the prior art above, and propose a method for matching XML patterns, which adopts enhanced Prufer sequences as the intermediate representation of XML patterns, and makes full use of language-related information and structure-related information to find the mapping between all similar elements in the two documents. The method is fully automated, and improves the matching efficiency under the premise of ensuring the matching quality, and solves the problems encountered in the existing pattern matching in terms of pattern representation, finding complex matches, and matching efficiency.

In order to achieve the above object, the concrete steps of the present invention include as follows:

(1) Input two XML schema documents to be matched.

(2) Build a pattern tree:

The document object model parsing is performed on the two extensible markup language pattern documents to be matched, and a pattern tree of the two extensible markup language pattern files to be matched is generated.

(3) Construct sequence structure:

The Prüfer sequences are constructed on the two pattern trees respectively, and two strengthened Prüfer sequences composed of the numbered Prüfer sequences and the marked Prüfer sequences are obtained.

(4) Language matching:

4a) Randomly select an element s and an element t from the marked Prufer sequences of the two strengthened Prufer sequences respectively;

4b) Obtain the name similarity value of element s and element t by adopting the name similarity value calculation method;

4c) Obtain the data type similarity value of element s and element t by adopting a data type similar value calculation method;

4d) Obtain the cardinality-constrained similarity value of element s and element t by using the cardinality-constrained similarity value calculation method;

4e) The weighted average of the name similarity value, data type similarity value, cardinality constraint similarity value of element s and element t is taken as the language similarity value of element s and element t;

4f) Repeat steps 4a) to 4e) until the linguistic similarity values between all elements in the two tagged Prüfer sequences are obtained.

(5) Complex element structure matching:

5a) respectively sorting all the nodes of the numbered Prüfer sequences in the two enhanced Prüfer sequences according to the sequence numbers of the nodes in the pattern tree from small to large;

5b) Randomly select an element i and an element j from the two sorted numbered Prüfer sequences respectively;

5c) Obtain the child similarity value of element i and element j by using the child similarity value calculation method;

5d) using a leaf similarity value calculation method to obtain the leaf similarity values of element i and element j;

5e) Obtain the sibling similarity value of element i and element j by adopting the sibling similarity calculation method;

5f) using the ancestor similarity value calculation method to obtain the ancestor similarity value of element i and element j;

5g) Use the weighted average of the child similarity value, leaf similarity value, sibling similarity value, and ancestor similarity value of element i and element j as the structural similarity value of element i and element j;

5h) taking the weighted average of the structural similarity value of element i and element j and the language similarity value obtained in step (4) as the overall similarity value of element i and element j;

5i) Repeat step 5c) to step 5h), until the overall similarity value between all elements in the two sorted numbered Prufer sequences is obtained;

5j) Filtering the overall similarity values between all elements in the two sorted numbered Prüfer sequences using a threshold method to obtain all matching complex node pairs to form a matching complex node pair set.

(6) Non-complex element structure matching:

6a) Randomly select an element pair from the matching element pairs obtained by complex node structure matching, and record the elements in the element pair as element e and element f respectively:

6b) search for the reinforced Prüfer sequence where the element e and the element f are located, find out all the atoms of the element e and the element f, and form the atomic set of the element e and the element f;

6c) Randomly select an element c from the atomic set of element e, and use the non-complex element structure matching method to obtain the structural similarity value of all elements in the atomic set of element c and element f;

6d) Determine whether there are elements in the atomic set of element e, and if so, perform step 6a); otherwise, consider that the overall similarity value between all elements in the atomic set of element e and element f has been obtained, and perform step 6e ):

6e) Repeat step 6a), step 6b), step 6c), step 6d), until the overall similarity value between all elements in the atomic set corresponding to the matching element pairs obtained by all complex node structure matching is obtained;

6f) Filtering the obtained overall similarity values of all element pairs using a threshold method to obtain matching non-complex node pairs to form a matching non-complex node pair set.

(7) Output matching results:

Output the union of the matched complex node pair set obtained in step (5) and the matched non-complex node pair set obtained in step (6).

Compared with the prior art, the present invention has the following advantages:

First, the present invention performs XML pattern matching at the level of XML document elements. It overcomes the shortcomings of the existing technology that only measures the similarity of the pattern at the document level, but does not penetrate into the finer granularity of the elements of the document, so that the present invention can be better used for semi-automatic and labor-intensive tasks , such as Extensible Markup Language schema integration, etc.

Second, the present invention makes full use of name, data type and cardinality constraint information in language matching, and fully considers structural related information such as children, siblings, leaves and ancestor nodes of nodes in structural matching. The invention overcomes the disadvantage of calculating the similarity of the XML pattern based on only a small amount of node-related language and structure information in the prior art, so that the present invention improves the quality of the XML pattern matching.

Third, the present invention uses an efficient sequence structure—enhanced Prüfer sequence to represent an extensible markup language pattern, and adopts the principle of decision tree to merge multiple string matching algorithms in the most critical part of language matching—name matching, to improve matching efficiency. In addition, in the structure matching, the structure matching method is only applied to the atomic elements that match complex element pairs, instead of calculating the structural similarity value of all atomic elements, this structure matching method is easy to find complex matches, and at the same time can ensure the matching efficiency . Overcoming the deficiency in the prior art that only pays attention to matching quality but ignores matching efficiency, the present invention achieves a balance between matching quality and performance, making the present invention applicable to wider applications.

Description of drawings

Fig. 1 is a flowchart of the present invention.

Detailed ways

The present invention will be described in further detail below in conjunction with accompanying drawing 1 .

Step 1, input two XML schema documents to be matched.

Enter two XML schema documents to match from the terminal.

Step 2, build the pattern tree.

The document object model analysis is performed on the two XML schema documents to be matched respectively, and the schema trees of the two XML schema files to be matched are generated.

The pattern tree can be expressed as a triplet T={N _T , E _T , Lab _NT }, where N _T ={n ₁ , n ₂ ,...,n _n } is a node set, and each node in the node set The only one represents an object in the schema; E _T = {(n _i , n _j )|n _i , n _j ∈ N _T } is an edge set, n _i is the parent node of n _j , and each edge represents two nodes The parent-child relationship between; Lab _NT is a node label set, and the label is a string describing the attributes of the node; the nodes in the pattern tree are divided into two types, atomic nodes and complex nodes; atomic nodes are leaf nodes without edges, Represents simple elements and attributes in the schema, and complex nodes are internal nodes of the schema tree that represent complex elements.

The structure of Extensible Markup Language mode is very complex, mainly as follows: the number of occurrences of elements and attributes is often a complex regular expression, cardinality constraints are not easy to determine; the existence of shared global declaration elements, attributes and complex types makes Extensible Markup There are loops in the language model. Therefore, you should also include Simplified Extensible Markup Language schemas before building your schema tree. The simplification steps include, in sequence: creating copies to solve the sharing in the schema, limiting the number of recursions to solve the infinite recursion in the schema, and using rules to simplify the cardinality constraints of elements and attributes. A schema tree is a rooted directed markup tree that reflects the hierarchical structure of an Extensible Markup Language schema document.

Step 3, construct the sequence structure.

The Prüfer Sequences (Prüfer Sequences) generation method is used for the two pattern trees to generate the corresponding Consolidated Prüfer Sequence (CPS). Among them, the enhanced Prüfer sequence is composed of Number Prüfer Sequence (Number Prüfer Sequence NPS) and Label Prüfer Sequence (Label Prüfer Sequence LPS), that is, CPS={NPS, LPS}, which respectively represent the Completely different information, wherein the numbered Prüfer sequence represents the structural information of the pattern tree, and the marked Prüfer sequence represents the semantic information of the pattern tree. Therefore, the enhanced Prüfer sequence uniquely represents a pattern tree. The enhanced Prüfer sequence has its own unique advantages, including node characteristics and relationship characteristics. The node characteristics refer to: suppose n _i is a node in the pattern tree, and the post sequence number of n _i in the pattern tree is k, then n _i is an atomic node if and only if k does not belong to NPS; n _i is a complex node if and Only if k belongs to NPS. The relationship characteristics include: parent-child relationship characteristics and brother relationship characteristics. The parent-child relationship feature refers to: Let NPS _i be the element with index i in NPS, and LPS _i be the element with index i in LPS, then NPS _i -node is the parent node of LPS _i node, and LPS _i is the parent node of NPS _i direct child nodes. The characteristics of sibling relationship refer to: Let NPS _i and NPS _j be the elements with index i and j in NPS respectively, and LPS _i and LPS _j are the elements with indexes i and j in NPS respectively, then LPS _i and LPS _j are sibling nodes If and only if NPS _i =NPS _j .

Using the Prüfer sequence construction method to construct the pattern tree of the extensible markup language pattern file to be matched correspondingly strengthens the Prüfer sequence, and the specific implementation process is as follows:

Step 1: Search the schema tree of two XML schema documents to be matched, and find the leaf node with the smallest postorder traversal sequence number;

Step 2, store the found leaf node with the smallest postorder traversal sequence number in the marked Prüfer sequence, and store the parent node of the leaf node in the numbered Prüfer sequence;

Step 3, from the pattern tree of the two extensible markup language pattern files to be matched, delete the found leaf node with the smallest postorder traversal sequence number;

Step 4, judge whether the pattern trees of the two XML pattern files to be matched are empty, and if so, execute step 1; otherwise, complete the construction of the Prüfer sequence.

Step 4, language matching.

The language matching method is based on the similarity among the name of the node, the data type of the node and the cardinality constraint of the node. The specific implementation process is as follows:

4a) An element s and an element t are arbitrarily selected respectively from the two marked Prüfer sequences that strengthen the Prüfer sequences.

4b) Obtain the name similarity value of the element s and element t by using the name similarity value calculation method.

The node name is an important information for matching without considering the data instance. Similar node names can be semantically similar, such as People and Staff, or structurally similar, such as Staff and TechnicalStaff. The structural similarity of the name can use the string matching method to calculate the similarity value of two name strings. Semantic similarity requires the help of external dictionaries, and frequent lookup of external dictionaries will increase the matching time. Therefore, considering the matching efficiency and automatic matching, name matching only includes structural similarity of names. The specific implementation process of the name matching method is as follows:

4b1) According to the name tokenization rules, divide the names of element s and element t to obtain token set 1 and token set 2.

In the XML mode, some nodes have longer names, some nodes represent the same information but often have different numerical serial numbers to distinguish them, and some node names have special symbols. In order to make node names better for string matching algorithms, node names are first normalized as a collection of many substrings—the token set, and each substring is called a token. The tokenization rule refers to: using special symbols such as "_", spaces, numbers, uppercase letters, etc. Symbol tokens, etc.

4b2) Randomly select an element a from the token set 1, and use the decision tree method to obtain the string similarity values between the element a and all elements in the token set 2. The specific implementation process of this step is as follows:

Step 1, randomly select an element b from token set 2.

Step 2: Compare the string values of element a and element b. If they are identical, the string similarity value of element a and element b is 1; otherwise, calculate the edit distance similarity value of element a and element b.

Step 3: Determine whether the similarity value of the edit distance is greater than or equal to the threshold 0.58. If yes, the string similarity value of element a and element b is the calculated similarity value of the edit distance; otherwise, use the Jaro-Winkler algorithm to calculate the similarity value of element a and element b The string similarity value of b, while using the 3-gram algorithm to calculate another string similarity value of element a and element b, and use the weighted average of the two similarity values as the string similarity value of element a and element b.

Step 4, determine whether there are elements in token set 2, and if so, go to step 1; otherwise, consider that the string similarity between element a and all elements in token set 2 has been obtained, and perform step 4b3 ).

4b3) Find the maximum value among the similarity values obtained in step 4b2), and use the maximum value as the string similarity value between element a and token set 2.

4b4) Judging whether there are elements in token set 1, if so, then go to step 4b2); otherwise, think that all elements of token set 1 and the string similarity values of token set 2 have been obtained, and perform step 4b5) .

4b5) Accumulate all the elements of token set 1 and the character string similarity values of token set 2 to obtain the cumulative sum.

4b6) Randomly select an element b from token set 2, and use the decision tree method to calculate the string similarity between element b and all elements in token set 1.

4b7) Find the maximum value among the similarity values obtained in step 4b6), and use the maximum value as the string similarity value between element b and token set 1.

4b8) Judging whether there are elements in the token set 2, if so, then go to step 4b6); otherwise, think that all elements of the token set 2 and the string similarity values of the token set 1 have been obtained, and perform step 4b9) .

4b9) Accumulate all the elements of token set 2 and the character string similarity values of token set 1 to obtain the cumulative sum.

4b10) adding up the sums obtained in step 4b5) and step 4b9) to obtain a sum.

4b11) Divide the sum by the total number of elements in the two token sets, and the resulting quotient is the linguistic similarity value of element s and element t.

4c) Obtain the data type similarity value of the element s and the element t by using the data type similarity value calculation method.

Although node names are important information for language matching, there are still many false matches in the mapping elements obtained by name matching. In order to improve the matching quality, data type becomes another pattern information that can be used in language matching. The data types of Extensible Markup Language schema include built-in types and custom types. Built-in types include string, int, bool, etc., and custom types include complex types and simple types, and simple types are also built-in types in the final analysis. The similarity of two built-in type nodes is greater than that of a built-in type node and a complex type node, and the type similarity value of two complex type nodes is determined by the structure of the nodes. The specific implementation process of the data type matching method is as follows:

Determine whether the data types of element s and element t are both built-in types, if so, look up the type similarity table to find out the data type similarity value of element s and element t; otherwise, determine whether the data types of element s and element t are the same is a complex type, if yes, the data type similarity value of element s and element t is 1; otherwise, the data type similarity value of element s and element t is 0.

4d) The cardinality-constrained similarity value calculation method is used to obtain the cardinality-constrained similarity value of the element s and the element t.

Cardinality constraint information of nodes is another important information that can be used in language matching. XML schema uses "minOccurs" and "maxOccurs" to define the occurrence times of elements or attributes in the schema. There are four basic cardinality constraint values for node cardinality representation in Document Type Definition DTD: "*", "?", "+" and "none". These four basic cardinality constraint values correspond to the scalable In the markup language schema definition (XML Schema Definition XSD), "none" means minOccurs=1 and maxOccurs=1, "?" means mihOccurs=0 and maxOccurs=1, "*" means minOccurs=0 and maxOccurs=unbounded, "+ " indicates that minOccurs=1 and maxOccurs=unbounded. If the cardinality constraints of two nodes can be expressed as these four basic cardinality constraint values, then the constraint similarity value of the two nodes only needs to look up the cardinality constraint similarity table, otherwise, perform the following steps to calculate the element s and element t Cardinality constraint similarity values for :

4d1) Calculate the minimum cardinality-constrained similarity value of element s and element t using the following formula:

$\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}$

Among them, u represents the minimum cardinality constraint similarity value of element s and element t, x represents the minimum constraint occurrence number of element s, and y represents the minimum constraint occurrence frequency of element t.

4d2) Use the following formula to calculate the maximum cardinality constraint similarity value of element s and element t:

$\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; no</span>}}$

Among them, v represents the maximum cardinality constraint similarity value of element s and element t, m represents the maximum constraint occurrence number of element s, and n represents the maximum constraint occurrence frequency of element t.

4d3) Calculate the average value of the minimum cardinality constraint similarity value and the maximum cardinality constraint similarity value of element s and element t, and the average value is the cardinality constraint similarity value of element s and element t.

4e) Take the weighted average of the name similarity value, data type similarity value, and cardinality constraint similarity value of element s and element t as the language similarity value of element s and element t.

4f) Repeat steps 4a) to 4e) until the linguistic similarity values between all elements in the two tagged Prüfer sequences are obtained.

Step 5, complex element structure matching.

The structure matching method of complex nodes is based on four structures of nodes: children, leaves, brothers and ancestors. The specific implementation process of this step is as follows:

5a) Sorting all the nodes of the numbered Prüfer sequences in the two enhanced Prüfer sequences according to the descending sequence numbers of the nodes in the pattern tree.

5b) Randomly select an element i and an element j from the two sorted numbered Prüfer sequences respectively.

5c) Using the child similarity value calculation method to obtain the child similarity values of element i and element j.

As the most important part of the element structure similarity value, the child similarity value directly reflects the basic structure of the element. The specific implementation process of this step is as follows:

5c1) Using the parent-child relationship characteristics contained in the enhanced Prüfer sequence, search for the enhanced Prüfer sequence corresponding to element i, find out all the children of element i, and form the child set of element i; search for the enhanced Prüfer sequence corresponding to element j Rueff sequence, find out all the children of element j, and form the child set of element j;

5c2) From the child set of element i, randomly select an element p, and obtain the overall similarity value of element p and all elements in the child set of element j. The overall similarity value of the two elements here is the weighted average of the language similarity value of the two elements and the structural similarity value of the two elements; because the structural similarity value of the two elements is in the order of the node's post-sequence number in the pattern tree from small to large Calculated, so, at this time, the structural similarity value of element p and element j's child set has been calculated, and its linguistic similarity value has been calculated in step (4).

5c3) Find the maximum value among the similarity values obtained in step 2, and use the maximum value as the similarity value of the child set of element p and element j.

5c4) Determine whether there are elements in the child set of element i, if so, go to step 5c2); otherwise, consider that the overall similarity value of all elements in the child set of element i and the child set of element j has been obtained, and perform step 5c5) .

5c5) Accumulate all similar values of all elements in the child set of element i and all similar values in the child set of element j to obtain a cumulative sum.

5c6) Divide the cumulative sum obtained in step 5c5) by the maximum number of elements contained in the two child sets, and the obtained quotient is the child similarity value of element i and element j.

5d) Obtain the leaf similarity value of element i and element j by adopting the leaf similarity value calculation method.

The specific implementation process of this step is as follows:

5d1) Using the parent-child relationship and sibling relationship characteristics contained in the enhanced Prüfer sequence, search for the enhanced Prüfer sequence corresponding to element i, find out all the leaves of element i, and form the leaf set of element i; search for the corresponding element j The strengthened Prufer sequence, find all the leaves of element j, and form the leaf set of element j.

5d2) The difference between the post-sequence number of the element in the pattern tree and the post-sequence number of each leaf node in the pattern tree in the leaf set of the element is used as the component of the digital vector of the element, and the digital vectors of element i and j are respectively constructed.

5d3) Using the law of cosines, calculate the similarity value of the numerical vectors of element i and element j, and the similarity value is the leaf similarity value of element i and element j.

5e) Obtain the sibling similarity value of element i and element j by adopting the sibling similarity calculation method.

The specific implementation process of this step is as follows:

5e1) Utilize the characteristics of the sibling relationship included in the enhanced Prüfer sequence, search for the enhanced Prüfer sequence corresponding to element i, find out all brothers of element i, and form a sibling set of element i; search for the enhanced Prüfer sequence corresponding to element j Prufer sequence, find out all siblings of element j, and form the sibling set of element j.

5e2) From the sibling set of element i, randomly select an element q, and obtain the language similarity value of all elements in the sibling set of element q and element j.

5e3) Find the maximum value among the similarity values obtained in step 5e2), and use the maximum value as the similarity value of the sibling set of element q and element j.

5e4) Determine whether there are elements in the sibling set of element i, and if so, go to step 5e2); otherwise, consider that the linguistic similarity values of all elements in the sibling set of element i and element j have been obtained, and perform step 5e5).

5e5) Accumulate the similarity values of all elements in the sibling set of element i and all the similarity values in the sibling set of element j to obtain an accumulation sum.

5e6) Divide the cumulative sum obtained in step 5e5) by the maximum number of elements contained in the two sibling sets, and the obtained quotient is the sibling similarity value of element i and element j.

5f) Using an ancestor similarity value calculation method to obtain the ancestor similarity values of element i and element j.

The specific implementation process of this step is as follows:

5f1) Use the parent-child relationship characteristics contained in the enhanced Prüfer sequence to search for the enhanced Prüfer sequence corresponding to element i, find out all the ancestors of element i, and connect all the ancestors of element i according to the search sequence Constitute the ancestor path of element i; search the enhanced prufer sequence corresponding to element j, find out all the ancestors of element j, and connect all the ancestors of element j according to the search order to form the ancestor path of element j.

5f2) Treat the ancestor path as a sequence of strings, each node name in the path as a whole, use the language matching method to calculate the language similarity value of each node, based on the language similarity value of all nodes in the ancestor path, calculate the element The edit distance between the ancestor path of i and the ancestor path of element j; when calculating edit distance, only consider whether the node names are similar in language and not required to be identical. For example, assuming that the ancestor paths of two nodes are PO/Orders/shipTo and PO/POrders/buyer respectively, among them, PO is exactly the same as PO, Orders is similar to POrders in language, and shipTo and buyer are not similar in language. Therefore, the two The edit distance between the ancestor paths of nodes is 1.

5f3) Divide the edit distance between the ancestor path of element i and the ancestor path of element j obtained in step 5f2) by the length of the ancestor path of element i (the number of ancestor nodes contained) and the length of the ancestor path of element j Maximum value, to get a quotient.

5f4) Unit 1 minus the quotient obtained in step 5f3) is the ancestor similarity value of element i and element j.

5g) Take the weighted average of the child similarity value, leaf similarity value, sibling similarity value, and ancestor similarity value of element i and element j as the structural similarity value of element i and element j.

5h) The weighted average of the structural similarity value and language similarity value of element i and element j is taken as the overall similarity value of element i and element j.

5i) Repeat steps 5c) to 5h) until the overall similarity values between all elements in the two sorted Prüfer sequences are obtained.

5j) Filtering the overall similarity values between all elements in the two sorted numbered Prüfer sequences using a threshold method to obtain all matching complex node pairs to form a matching complex node pair set.

Step 6, non-complex element structure matching.

For each matching element pair obtained by complex node structure matching, the structural similarity value between all nodes in the atomic set corresponding to the element pair is calculated. In addition to improving matching efficiency, this matching method can also identify complex matching. The specific implementation process of this step is as follows:

6a) Randomly select an element pair from the matching element pairs obtained by complex node structure matching, and denote the elements in the element pair as element e and element f respectively.

6b) Search the enhanced Prüfer sequence of element e and element f respectively, find out all atoms of element e and element f, and form the atomic set of element e and element f.

6c) From the atomic set of element e, randomly select an element c, and obtain the structural similarity value of all elements in the atomic set of element c and element f, the specific steps are as follows:

6c1) Randomly select an element d from the atomic set of element t.

6c2) Using the calculation method of sibling similarity value described in 5e), obtain the sibling similarity value of element c and element d.

6c3) Using the ancestor similarity value calculation method described in 5f), obtain the ancestor similarity values of elements c and d.

6c4) The weighted average of sibling similarity values and ancestor similarity values of element c and element d is taken as the structural similarity value of element c and element d.

6c5) Add the structural similarity value and linguistic similarity value of element c and element d, and the obtained sum is used as the overall similarity value of element c and element d.

6c6) Determine whether there are elements in the atomic set of element f, and if so, go to step 6c1); otherwise, consider that the overall similarity value of all elements in the atomic set of element c and element f has been obtained, and perform step 6d).

6d) Determine whether there are elements in the atomic set of element e, and if so, perform step 6a); otherwise, consider that the overall similarity value between all elements in the atomic set of element e and element f has been obtained, and perform step 6e ).

6e) Repeat step 6a), step 6b), step 6c), and step 6d), until the overall similarity value between all elements in the atomic set corresponding to the matched element pairs obtained by all complex node structure matches is obtained.

6f) Filtering the obtained overall similarity values of all element pairs using a threshold method to obtain matching non-complex node pairs to form a matching non-complex node pair set.

(7) Output matching results:

Output the union of the matched complex node pair set obtained in step (5) and the matched non-complex node pair set obtained in step (6).

Claims (10)

1. expandable mark language mode matching process comprises following concrete steps:

(1) two expandable mark language mode documents to be matched of input;

(2) make up scheme-tree:

Two expandable mark language mode documents to be matched are carried out DOM Document Object Model resolve, generate the scheme-tree of two expandable mark language mode files to be matched;

(3) tectonic sequence structure:

Respectively two scheme-trees are carried out the Pu Lvfu sequence structure, strengthen the Pu Lvfu sequence for two that obtain to be formed by numbering Pu Lvfu sequence and mark Pu Lvfu sequence;

(4) language coupling:

4a) from the mark Pu Lvfu sequence of two reinforcement Pu Lvfu sequences, choose an element s and element t arbitrarily respectively;

4b) adopt title similar value computing method, obtain the title similar value of element s and element t;

4c) adopt data type similar value computing method, obtain the data type similar value of element s and element t;

4d) adopt constraint base similar value computing method, obtain the constraint base similar value of element s and element t;

4e) with the weighted mean of the title similar value of element s and element t, data type similar value, the constraint base similar value language similar value as element s and element t;

4f) repeated execution of steps 4a) to step 4e), all elements language similar value between any two in obtaining two mark Pu Lvfu sequences;

(5) complicated element structure coupling:

5a) according to back sequence number from small to large the order of node in scheme-tree, respectively two all nodes of strengthening the numbering Pu Lvfu sequence in the Pu Lvfu sequence are sorted;

5b) from the numbering Pu Lvfu sequence after two orderings, choose an element i and element j respectively arbitrarily;

5c) adopt child's similar value computing method, obtain child's similar value of element i and element j;

5d) adopt leaf similar value computing method, obtain the leaf similar value of element i and element j;

5e) adopt fraternal similar value computing method, obtain the fraternal similar value of element i and element j;

5f) adopt ancestors' similar value computing method, obtain ancestors' similar value of element i and element j;

5g) with the weighted mean of child's similar value of element i and element j, leaf similar value, fraternal similar value, the ancestors' similar value structural similarity value as element i and element j;

5h) weighted mean of the language similar value that the structural similarity value of element i and element j and step (4) are obtained are as the global similarity value of element i and element j;

5i) repeated execution of steps 5c) to step 5h), all elements global similarity value between any two in obtaining two numbering Pu Lvfu sequences after the ordering;

5j) to all elements global similarity value between any two in the numbering Pu Lvfu sequence after two orderings, use threshold method to filter, the complex node that obtains all couplings is right, forms the complex node of coupling to collection;

(6) non-complex element structure coupling:

6a) appoint from the resulting coupling element of complex node structure matching centering that to get an element right, the element of element centering be designated as element e and element f respectively:

6b) the reinforcement Pu Lvfu sequence at searching element e and element f place is respectively found out all atoms of element e and element f, the former subclass of component e and element f;

6c) concentrate from the atom of element e, appoint and get an element c, adopt non-complex element structure matching process, the atom that obtains element c and element f is concentrated the structural similarity value of all elements;

6d) atom of judging element e concentrates whether to also have element, if having, and execution in step 6a then); Otherwise, think that the atom that has obtained element e and element f concentrates all elements global similarity value between any two, execution in step 6e);

6e) repeated execution of steps 6a), step 6b), step 6c), step 6d), up to obtaining the resulting coupling element of all complex node structure matching corresponding atom is concentrated all elements global similarity value between any two;

6f) the global similarity value right to resulting all elements uses threshold method to filter, and the non-complex node that obtains mating is right, forms the non-complex node of coupling to collection;

(7) output matching result:

The union of non-complex node to collecting of the coupling that the complex node of the coupling that obtains of output step (5) obtains collection and step (6).

2. a kind of expandable mark language mode matching process according to claim 1 is characterized in that, the concrete steps of the Pu Lvfu sequence structure described in the step (3) are as follows:

The 1st goes on foot, and searches for the scheme-tree of two expandable mark language mode documents to be matched, therefrom finds the leaf node with minimum postorder traversal serial number;

The 2nd step with the leaf node with minimum postorder traversal serial number that finds, was stored in the mark Pu Lvfu sequence, and the father node with this leaf node is stored in the numbering Pu Lvfu sequence simultaneously;

The 3rd step, from the scheme-tree of two expandable mark language mode files to be matched, the leaf node with minimum postorder traversal serial number that deletion is found;

In the 4th step, judge whether the scheme-tree of two expandable mark language mode files to be matched is empty, if then carried out for the 1st step; Otherwise, finished the structure of Pu Lvfu sequence.

3. a kind of expandable mark language mode matching process according to claim 1 is characterized in that step 4b) described in the performing step of title similar value computing method as follows:

The 1st step, according to title token rule, the title of element s and element t is cut apart, obtain token collection 1 and token collection 2;

The 2nd step, from token collection 1, appoint and get an element a, adopt traditional decision-tree, obtain the character string similar value of all elements in element a and the token collection 2;

In the 3rd step, find out the maximal value in the resulting similar value in the 2nd step, with the character string similar value of this maximal value as element a and token collection 2;

The 4th step, judge in the token collection 1 whether also have element, if having, then forwarded for the 2nd step to; Otherwise, think the character string similar value that has obtained token collection 1 all elements and token collection 2, carried out for the 5th step;

The 5th step, the character string similar value of token collection 1 all elements and token collection 2 is added up, obtain adding up with;

The 6th step, from token collection 2, appoint and get an element b, adopt traditional decision-tree, calculate the character string similar value of all elements in element b and the token collection 1;

In the 7th step, find out the maximal value in the resulting similar value in the 6th step, with the character string similar value of this maximal value as element b and token collection 1;

The 8th step, judge in the token collection 2 whether also have element, if having, then forwarded for the 6th step to; Otherwise, think the character string similar value that has obtained token collection 2 all elements and token collection 1, carried out for the 9th step;

The 9th step, the character string similar value of token collection 2 all elements and token collection 1 is added up, obtain adding up with;

In the 10th step, with resulting adding up and addition in the 5th step, the 9th step, obtain summation;

In the 11st step, with the total number of summation divided by the concentrated element of two tokens, the gained merchant is the language similar value of element s and element t.

4. a kind of expandable mark language mode matching process according to claim 1, it is characterized in that, step 4c) the data type similar value computing method described in refer to, whether the data type of judging element s and element t is built-in type, if then from the similar table of type, find out the data type similar value of element s and element t; Otherwise, judge whether the data type of element s and element t is complicated type, if then the data type similar value of element s and element t is 1: otherwise the data type similar value of element s and element t is 0.

5. a kind of expandable mark language mode matching process according to claim 1 is characterized in that step 4d) described in constraint base similar value computing method refer to,

The 1st step, according to the different values of the constraint base of element s and element t, judge whether the constraint base of element s and element t is basic constraint base value, if, then search the similar table of constraint base, draw the constraint base similar value of element s and element t; Otherwise, carried out for the 2nd step to the 4th step, calculate the constraint base similar value of element s and element t:

In the 2nd step, use following formula to calculate the minimum cardinality constraint similar value of element s and element t:

$u=1-\frac{|x-y|}{x+y}$

Wherein, u represents the minimum cardinality constraint similar value of element s and element t, and x represents the least commitment occurrence number of element s, and y represents the least commitment occurrence number of element t;

In the 3rd step, use following formula to calculate the maximum constraint base similar value of element s and element t:

$v=1-\frac{|m-n|}{m+n}$

Wherein, v represents the maximum constraint base similar value of element s and element t, and m represents the maximum constrained occurrence number of element s, and n represents the maximum constrained occurrence number of element t;

The 4th step, calculate the minimum cardinality constraint similar value of element s and element t and the mean value of maximum constraint base similar value, this mean value is the constraint base similar value of element s and element t.

6. a kind of expandable mark language mode matching process according to claim 1 is characterized in that step 5c) described in the performing step of child's similar value computing method as follows:

The 1st step, utilize to strengthen the set membership characteristic that the Pu Lvfu sequence comprises, corresponding reinforcements of searching element i Pu Lvfu sequence is found out all children of element i, and the child of component i collects; The corresponding reinforcement of searching element j Pu Lvfu sequence is found out all children of element j, child's collection of component j;

The 2nd step, concentrated from the child of element i, appoint and get an element p, the child who obtains element p and element j concentrates the global similarity value of all elements;

In the 3rd step, find out the maximal value in the resulting similar value in the 2nd step, with the similar value of this maximal value as child's collection of element p and element j;

The 4th step, judge the child of element i concentrates whether also have element, if having, then forwarded for the 2nd step to; Otherwise, think to have obtained the global similarity value that element child i concentrates the child of all elements and element j to collect, carried out for the 5th step;

The 5th step, concentrate all elements and all similar value of child's collection of element j to add up the child of element i, obtain one add up with;

The 6th step, the 5th step was resultingly added up and collects the maximal value of contained number of elements divided by two children, the gained merchant is child's similar value of element i and element j.

7. a kind of expandable mark language mode matching process according to claim 1 is characterized in that step 5d) described in the performing step of leaf similar value computing method as follows:

The 1st step, utilize to strengthen set membership and brotherhood characteristic that the Pu Lvfu sequence comprises, the corresponding reinforcement of searching element i Pu Lvfu sequence is found out all leaves of element i, the leaf collection of component i; The corresponding reinforcement of searching element j Pu Lvfu sequence is found out all leaves of element j, the leaf collection of component j;

In the 2nd step, with the difference of the concentrated back sequence number of each leaf node in scheme-tree of leaf of back sequence number and the element of element in the scheme-tree component as the digital vectors of element, make up the digital vectors of element i and element j respectively;

The 3rd step, use the cosine law, calculate the similar value of the digital vectors of element i and element j, this similar value is the leaf similar value of element i and element j.

8. a kind of expandable mark language mode matching process according to claim 1 is characterized in that step 5e) described in the performing step of fraternal similar value computing method as follows:

The 1st step, utilize to strengthen the brotherhood characteristic that comprises in the Pu Lvfu sequence, corresponding reinforcements of searching element i Pu Lvfu sequence is found out all brothers of element i, and the brother of component i collects; The corresponding reinforcement of searching element j Pu Lvfu sequence is found out all brothers of element j, brother's collection of component j;

The 2nd step, concentrated from the brother of element i, appoint and get an element q, the brother who obtains element q and element j concentrates the language similar value of all elements;

In the 3rd step, find out the maximal value in the gained similar value in the 2nd step, with the similar value of this maximal value as brother's collection of element q and element j;

The 4th step, judge the brother of element i concentrates whether also have element, if having, then forwarded for the 2nd step to; Otherwise, think that the brother who has obtained element i concentrates the language similar value of all elements and element j, carries out for the 5th step;

The 5th step, concentrate all elements and all similar value of brother's collection of element j to add up the brother of element i, obtain one add up with;

The 6th step is with resulting add up and divided by the maximal values of two contained number of elements of brother collection, the gained merchant is the fraternal similar value of element i and element j of the 5th step.

9. a kind of expandable mark language mode matching process according to claim 1 is characterized in that step 5f) described in the performing step of ancestors' similar value computing method as follows:

The 1st step, utilize the set membership characteristic that comprises in the reinforcement Pu Lvfu sequence, the corresponding reinforcement of searching element i Pu Lvfu sequence is found out all ancestors of element i, and according to the sequencing of search all ancestors of element i is coupled together the ancestors path that constitutes element i; The corresponding reinforcement of searching element j Pu Lvfu sequence is found out all ancestors of element j, and according to the sequencing of search all ancestors of element j is coupled together the ancestors path that constitutes element j;

The 2nd step, regard the ancestors path as a character string sequence, each nodename in the path is regarded an integral body as, the language similar value of utilizing the language matching process to obtain, the editing distance between the ancestors path of calculating element i and the ancestors path of element j;

In the 3rd step, the editing distance between the ancestors path of the ancestors path of the element i that obtains in the 2nd step and element j divided by the maximal value in ancestors' path of ancestors' path (contained ancestor node number) of element i and element j, is obtained a merchant;

In the 4th step, unit 1 deducts resulting merchant of the 3rd step, is ancestors' similar value of element i and element j.

10. a kind of expandable mark language mode matching process according to claim 1 is characterized in that step 6c) described in the performing step of non-complex element structure coupling as follows:

In the 1st step, get an element d from concentrated of the atom of element t;

The 2nd step, adopt the described fraternal similar value computing method of claim 8, obtain the fraternal similar value of element c and element d;

The 3rd step, adopt the described ancestors' similar value of claim 9 computing method, obtain ancestors' similar value of element c and element d;

The 4th step is with the weighted mean value of fraternal similar value and the ancestors' similar value of element c and element d, as the structural similarity value of element c and element d;

The 5th step, with structural similarity value and the addition of language similar value of element c and element d, resulting and as the global similarity value of element c and element d.