CN108717405B - The complementing method of the default subject of staircase design specification based on mind map - Google Patents

Abstract

The invention provides a method for complementing the default subject of a staircase design specification based on a mind map. Through the inheritance relationship between ontology in the mind map, it is possible to determine the missing subject of the subordinate clause with incomplete syntactic structure in the complex sentence and complete it, and to extract the subject-predicate-object in the concurrent sentence and the conjunction sentence, in order to be able to better Natural language processing is carried out, and finally the knowledge graph of the staircase building code is constructed to realize automatic drawing review; the present invention effectively avoids uncertain factors that may occur in the drawing review process, has low false detection rate, simple operation, saves manpower, and is extremely Greatly improved the completion efficiency of engineering projects in the construction industry.

Description

A Completion Method for the Default Subject of Stair Design Specifications Based on Mind Map

technical field

The invention belongs to the technical field of computer natural language processing technology, and in particular relates to a method for complementing the default subject of a staircase design specification based on a mind map.

Background technique

With the development of computer science and technology, various engineering fields have introduced the concept of knowledge graphs. Knowledge graphs can describe the entities in the world and the connections between entities, and provide an ability to analyze problems from the perspective of relationships. co-construction, query, sharing and reuse. Industry fields that currently use knowledge graphs include medicine, military, education, and more. However, the information construction of the construction industry is still in its infancy. A project in the construction industry first needs to be planned and designed by the design institute, and then the designed models and drawings are reviewed, and then provided to the construction unit for construction, and finally put into use. Accurate drawings and models will greatly reduce the phenomenon of rework caused by design changes in the construction stage. However, due to the lack of technology at this stage, most projects in our country are manually reviewed by experts after design. Comparing the designed drawings with the design specifications of the construction industry, this will cause problems such as high false detection rate, high missed detection rate, low inspection intensity, and many uncertain factors.

Contents of the invention

The purpose of the present invention is to provide a method for complementing the default subject of the staircase design specification based on the mind map, which solves the problem that the subject of some sentences in the staircase design specification is incomplete and difficult to identify.

The invention provides a method for completing the default subject of the staircase design specification based on a mind map. By completing the text in the specification, it can better perform natural language processing, and finally build a knowledge map of the staircase construction specification to realize automatic review. The graph helps to solve the problem of high false detection rate in the prior art when people review the graph.

The technical solution adopted in the present invention is a method for complementing the default subject of the staircase design specification based on the mind map, comprising the following steps:

Step 1: Obtain the specifications about staircase design from the residential building design specifications as the processing corpus, and use the dictionary-based forward maximum matching algorithm to segment the original text, and use the statistical method to tag the words after word segmentation , to get the preprocessed text.

Step 2: Referring to the description format of the staircase specification in the IFC standard, comb the ontology related to the staircase design and the relationship between the ontology into a mind map, and construct the corresponding index tree.

Step 3: Use context-free grammar to parse the preprocessed text. In the case of a concurrent sentence or a conjunction sentence, determine the sentence type components in the sentence by searching the ontology related to the object in the index tree. This builds a syntax tree and analyzes whether the syntactic structure of a sentence is complete. Finally, the corpus with missing subject is filtered out.

Step 4: By searching in the index tree, find the parent node and the only path to the root node for the object ontology with incomplete sentence structure. The parent node is the default subject of the sentence, except Except the parent node, all the nodes on this path are the modifiers of the subject, and then add the default subject to the original sentence, and output the complete staircase design specification of the subject, predicate and object.

In step 1: the method adopted for word segmentation is the dictionary-based forward maximum matching algorithm. Part-of-speech tagging uses a part-of-speech tagging method based on a hidden Markov model.

In step 3: when performing grammatical analysis, determine its component in the sentence according to the part of speech of the word and the position where it appears in the sentence.

The specific steps of the method for constructing the syntax tree in step 3 are as follows: first, define the context-free grammar G={N, Σ, X, S}. Among them, N represents a set of labels for non-leaf nodes; Σ represents a set of labels for leaf nodes, that is, the words that make up a sentence; X represents a set of syntactic rules, which is the production formula of N, and each rule can be expressed as X =Y ₁ Y ₂ ...Y _n , X∈N, Y _i ∈(N∪Σ); the α of at least one production in X must be acted by S. And S indicates the label at the beginning of the syntax tree.

Using the bottom-up method, starting from the string to be analyzed, use the string to be analyzed to match the right character of the X arrow in the context-free grammar rule, and replace it with the left character after the match is successful, until S appears, and the syntax tree is constructed complete.

In step 4: the traversal algorithm is used when finding the parent node.

The beneficial effects of the present invention are:

A Completion Method for the Default Subject of Stair Design Specification Based on Mind Map. Through the inheritance relationship between ontology in the mind map, the missing subject of the subordinate clause with incomplete syntactic structure in the compound sentence can be determined and completed, and the subject-predicate-object in the concurrent sentence and the conjunction sentence can be extracted to facilitate natural language Process and construct the knowledge map, and finally help to automatically review the drawings; the present invention effectively avoids uncertain factors that may appear in the process of manual review, has low false detection rate, simple operation, saves manpower, and greatly improves the quality of engineering projects in the construction industry. completion efficiency.

Description of drawings

Fig. 1 is the main flow of the completion method of the default subject of the stair design specification based on mind map in the present invention;

Figure 2 is a mind map model defined with reference to the IFC standard.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

The present invention is based on the method for complementing the default subject of the staircase design specification based on the mind map, comprising the following steps:

Step 1: Obtain the specifications about staircase design from the residential building design specifications as the processing corpus, and use the dictionary-based forward maximum matching algorithm to segment the original text, and the method based on the hidden Markov model to segment the text Word strings are part-of-speech tagged to obtain the preprocessed text.

Step 2: Referring to the description format of the staircase specification in the IFC standard, comb the ontology related to the staircase design and the relationship between the ontology into a mind map, and construct the corresponding index tree.

Step 3: Use context-free grammar to parse the preprocessed text. In the case of a concurrent sentence or a conjunction sentence, determine the sentence type components in the sentence by searching the ontology related to the object in the index tree. This builds a syntax tree and analyzes whether the syntactic structure of a sentence is complete. Finally, the corpus with missing subject is filtered out.

Step 4: By searching in the index tree, find the parent node and the only path to the root node for the object ontology with incomplete sentence structure. The parent node is the default subject of the sentence, except Except the parent node, all the nodes on this path are the modifiers of the subject, and then add the default subject to the original sentence, and output the complete staircase design specification of the subject, predicate and object.

In step 1: the method adopted for word segmentation is the dictionary-based forward maximum matching algorithm. Part-of-speech tagging uses a part-of-speech tagging method based on a hidden Markov model.

In step 3: when performing grammatical analysis, determine its component in the sentence according to the part of speech of the word and the position where it appears in the sentence.

The specific steps of the method for constructing the syntax tree in step 3 are as follows: first, define the context-free grammar G={N, Σ, X, S}. Among them, N represents a set of labels for non-leaf nodes; Σ represents a set of labels for leaf nodes, that is, the words that make up a sentence; X represents a set of syntactic rules, which is the production formula of N, and each rule can be expressed as X =Y ₁ Y ₂ ...Y _n , X∈N, Y _i ∈(N∪Σ); the α of at least one production in X must be acted by S. And S indicates the label at the beginning of the syntax tree.

Using the bottom-up method, starting from the string to be analyzed, use the string to be analyzed to match the right character of the X arrow in the context-free grammar rule, and replace it with the left character after the match is successful, until S appears, and the syntax tree is constructed complete.

In step 4: the traversal algorithm is used when finding the parent node.

Since the Chinese text is composed of words, and there is no obvious dividing mark between words, the first step in processing the text is to input the corpus to segment it. The word segmentation method adopted in this patent is based on the positive maximum matching of the dictionary The algorithm:

First, a string s ₁ is taken from the text to be segmented, and the ontology in the staircase specification is formed into a dictionary, and a hash table is constructed. According to the maximum word length maxlen defined in the dictionary, starting from the left side of the string s ₁ , extract a substring w whose length is not greater than maxlen. And: search whether w is a word in the hash table, if w is output as a word. If it is not a word, subtract a word from the end of w, and continue to iteratively judge whether w is in the hash table until w is empty or s ₁ is empty.

In order to determine the syntactic structure of a sentence, it is first necessary to determine the sentence composition. The sentence composition in Chinese is determined by the part of speech of the word and the order in which it appears in the sentence. Part-of-speech tagging uses a method based on the hidden Markov model:

This method is divided into three modules: Initialise, Induction, Back tracing the besttagging.

First, in the Initialize step, the probability of each part of speech appearing at the beginning of a sentence in the corpus is counted, and multiplied by the probability of the part of speech ejecting a word to obtain a word score. Then in the Induction step, the Viterbi algorithm is used to calculate the score of every two adjacent words, which is equal to the initial score of the part of speech multiplied by the probability of conversion between parts of speech and the probability of injecting the part of speech into this word. Select the part-of-speech value with a large score from the final score and record it in the Backpointer. Finally, in the step of Back tracing the best tagging, trace back from back to front to obtain the sequence string str ₂ composed of parts of speech.

The part of speech alone cannot determine whether the sentence structure of this sentence is complete, and it is necessary to further judge by constructing a grammar tree. The specific steps of the method for constructing the syntax tree are as follows:

First define a context-free grammar as follows:

(a) N represents a set of labels for non-leaf nodes, such as original sentences, noun phrases, verb phrases, etc.

(b) Σ represents the labeling of a set of leaf nodes, that is, the words that make up the sentence.

(c) X represents a set of syntactic rules, that is, a production of N, and each rule can be expressed as X=Y ₁ Y ₂ ...Y _n , X∈N, Y _i ∈(N∪Σ).

(d) S indicates the label at the beginning of the syntax tree.

Define its quadruples for a specific sentence through context-free grammar, and a syntax tree can be obtained through bottom-up derivation. If there are two or more nouns before the predicate, and the subject cannot be judged, you need to use the mind map to determine the ontology that has an inheritance relationship with the object to determine who is the noun in these nouns. The subject and other nouns are only attributives, and the modification limits the scope of the subject. The corpus with incomplete syntactic structure can be easily judged and extracted through the syntax tree.

After extracting the compound sentence without the subject, it is necessary to find the object from the clauses, and determine the parent node that has an inheritance relationship with it based on the position of the object in the mind map. In order to fill in the missing subject, the process is to input the object ontology, output the subject that has a direct inheritance relationship with this object as its subject, and an attributive phrase composed of all nodes on the unique path from the subject to the root node, which is used to modify Limit the domain of the subject.

What shown in accompanying drawing 1 is the flow process of the whole completion subject. At the same time of word segmentation and part-of-speech marking, refer to the IFC standard to construct the relationship of ontology in the specification of residential stairs into a mind map model, use this model to construct a grammar tree from the corpus, and filter out the corpus with incomplete syntactic structure, through query The inheritance relationship of the ontology used as the object component in the mind map finds its parent node as the subject, and the only path from it to the root node is to modify and limit the scope of this specification.

According to the forward maximum matching algorithm based on the dictionary, firstly, the residence rules 4.1.2 are extracted from the "Residential Design Code" as a corpus sample: "The net width of the residential stair section should not be less than 1.1m. When there is a railing on one side, no It should be less than 1m.".

Take a character string to be segmented from the original text and perform word segmentation. At this time, s ₁ = "The net width of the residential staircase should not be less than 1.1m, and if there is a railing on one side, it should not be less than 1m." Determine maxlen according to the constructed dictionary as =10, s ₂ is initialized as empty. Construct a dictionary as a hash table.

Select the substring w whose length is not greater than maxlen from the left side of _s1 = "the net width of the residential staircase is not", judge whether w is empty or not, judge whether w is a word in the hash table, and traverse the hash table, no matches found. Decrease the right side of w by one word, w="clean width of residential stairs", and continue to iterate. Until w is reduced to "house", the search is successful in the hash table, and the house is added to s ₂ , s ₂ = "house/". s ₁ = "The net width of the stair section should not be less than 1.1m, and if there is a railing on one side, it should not be less than 1m." When iterating until s ₁ is empty, the output s ₂ = "residential/stairs/section/net The width /should not/be less than /1.1m/, /one side/has/railing/when/, /should not/be less than /1m/.”

After the word segmentation, the part of speech of the word is marked, and the process of marking the part of speech is shown in Algorithm 2. First, the part of speech of a part of the corpus is manually marked, and then the viterbi algorithm is used to train the parameters, and the remaining corpus is automatically marked by machine learning. Among them, the unregistered words are smoothed, and the marked unregistered words are smoothed. Add the correct corpus to the training set to continue training more reliable parameters. The final output corpus is {noun: residence}{noun: stairs}{noun: stairs}{noun:net width}{adv:should not}{v:less than}{num:1.1m}, {adv:side} {v:with}{noun:railing}{adv:when}, {adv:should not}{v:less than}{num:1m}.

According to the inter-ontology relationship in the residential staircase design specification defined by the IFC standard, the mind map model defined with reference to this format is shown in Figure 2.

Construct a grammar tree based on the corpus output by preprocessing, where S represents a sentence; NP, VP, and PP are nouns, verbs, and prepositional phrases (phrase level); A, V, and P are adverbs, verbs, and prepositions respectively; Noun is a noun, and Num is a numeral. Define a context-free grammar production as follows:

1) S → NPVP

2) NP→NP|NounNoun|ε

3) VP→AVN|AVNA

4) A→Adv

5) V→Verb

6) N→Noun|Num

7) Noun→residential|stairs|stairs|net width|railing

8) Adv → should not

9)Num→1.1m|1m

10) Verb→less than |one side|when|has

The definition of production X: α→β needs to meet the following conditions:

1) α can be any label of leaf nodes and non-leaf nodes, and cannot be ε;

2) β can be any label of leaf nodes and non-leaf nodes, and can be ε;

3) The α of at least one production in X must be acted by S.

Using a bottom-up, specification-based method, starting from the string to be analyzed, use the string to be analyzed to match the right character of the arrow of the context-free grammar rule, and replace it with the left character after the match is successful until S.

Record the constructed syntax tree as the following format:

a)[S[NP[NP[Noun residence][Noun stairs]][NP[Noun steps][Noun net width]]][VP[A[Adv should not]][V[Verb less than]][N [Num 1.1m]]]]

b)[S[NP[ε][VP[A[Adv side]][V[Verb set]][N[Noun railing]][A[Adv]]]]

c)[S[NP[ε]][VP[A[Adv should not]][V[Verb less than]][N[Num 1m]]]]

Analyze the syntactic structure, take the second clause as an example, center on the predicate, analyze the subject and adverbial before the predicate, and the participle object and complement after the predicate. After analysis, it can be judged that the adverb "side" is used as an adverbial, and the verb "have" As a predicate, the noun "railing" is used as an object, and the last adverb "shi" means that this clause is used as an adverbial in the whole sentence, but this clause is incomplete, and it is impossible to judge what is "with railings" without contacting the context ".

Sentences found to have an incomplete syntactic structure include s ₁ = "when there is a railing on one side", and s ₂ = "should not be less than 1m".

The process of finding the subject takes s ₁ as an example. First, search for the node "railing" in the B index tree, and then find a path from this node to the root node: find the parent node of the railing as "ladder section", "ladder section "'s parent node is "stairs", the parent node of "stairs" is "house", and the parent node of "house" is "building". That is to say, the missing subject of this clause is the stair section. Complete s2 as "when there is a railing _on one side of the stair section", the path to the root node is the attributive that modifies the subject, namely: "the staircase of the building's residence""When there is a railing on one side of the ladder". Since attributives do not make components in the sentence structure, it is not necessary to complete them.

In addition to filling in missing subjects, this method can be used to determine the subject in specifications where the composition of the sentence is unclear. For example, in the first clause of the corpus sample "the net width of the residential staircase should not be less than 1.1m", when constructing the syntax tree, it is found that the subject is not missing, but it is not as easy to extract triples from this specification as the subject-predicate sentence , because without semantic analysis, it is impossible to judge what should not be less than 1.1m, even using semantic analysis is not easy. Therefore, when analyzing the syntactic structure of this sentence, we must first determine the subject. In the mind map, determine which one of the house, stairs, and stairs contains the attribute of net width. By traversing the index tree, it can be determined that the stairs are the subject component, and the residential stairs are the attributive, modifying and limiting the scope of the subject.

Finally, the clauses are completely supplemented as follows: "The clear width of the steps of the residential stairs should not be less than 1.1m, and when there is a railing on one side of the steps, the clear width of the steps should not be less than 1m.".

According to the method of the present invention, the default staircase design specification text of the subject is successfully completed. The purpose of the present invention is to provide a method for completing the default subject of the staircase design specification based on a mind map. By completing the text in the specification, natural language processing can be better performed, and finally a knowledge map of the staircase construction specification can be constructed. , to achieve automatic drawing review, effectively avoiding uncertain factors that may occur in the manual review process, low false detection rate, simple operation, while saving manpower, it greatly improves the completion efficiency of construction projects in the construction industry.

Claims (4)

1. the complementing method of the default subject of staircase design specification based on mind map, which comprises the following steps:

Step 1: obtaining corpus of the specification as processing in relation to staircase design from house building design standard, and use Forward Maximum Method algorithm based on dictionary segments urtext, and based on the method for hidden Markov model to point Word after word carries out part-of-speech tagging, obtains pretreated text；

Step 2:, will be relevant to staircase design referring to the descriptor format of staircase design specification in industrial foundation class IFC standard Relationship is combed into mind map between body and ontology, and constructs corresponding index tree；

Step 3: pretreated text being carried out syntax parsing using context-free grammar, is encountering pivotal sentence or even meaning sentence The case where, the sentence pattern ingredient in sentence is determined by searching ontology relevant to object in index tree, and syntax tree is constructed with this, And whether the syntactic structure for analyzing a sentence is complete；Finally therefrom filter out the corpus of default subject；

Step 4: by scanning in index tree, its father node is searched to the object ontology with imperfect sentence pattern structure, And object ontology, to unique paths of root node, father node is this default subject, in addition to father node, this All nodes on path are the modification attribute of subject, after default subject is added to prototype statement, output Subject, Predicate and Object is complete Staircase design specification.

2. the complementing method of the staircase design specification default subject according to claim 1 based on mind map, feature It is, in the step 3: hereafter Grammars exist after carrying out syntax parsing according to the part of speech and word of word in use The position occurred in sentence determines the ingredient in sentence.

3. the complementing method of the staircase design specification default subject according to claim 1 based on mind map, feature Be, in the step 3 construct syntax tree method specific step is as follows: define first context-free grammar G=N, Σ, X, S }, wherein N indicates the mark of one group of n omicronn-leaf child node；Σ indicates the mark of one group of leaf node, that is, forms the word of sentence；X Indicate the rule of one group of syntax, the as production of N, every rule is expressed as X=Y₁Y₂…Y_n,X∈N,Y_i∈(N∪Σ)；X In the α of at least one production served as by S；α is any mark of leaf node and n omicronn-leaf child node, and S indicates syntax tree The mark of beginning；

Using bottom-up method, from character start of string to be analyzed, matching context-free is gone with character string to be analyzed Matching character is replaced with left character after successful match by the character of arrow right part in grammar rule X, until S occurs, syntax tree Building finishes.

4. the complementing method of the staircase design specification default subject according to claim 1 based on mind map, feature It is, in the step 4: using ergodic algorithm when searching father node.