CN114970543A - A Semantic Analysis Method for Crowdsourcing Design Resources - Google Patents

Abstract

The invention discloses a semantic analysis method for crowdsourced design resources, comprising: step 1, performing word separation and part-of-speech tagging on short sentences of crowdsourced design resources; step 2, dividing the short sentences processed in step 1 into multiple independent short sentences; step 3, carry out the following processing for each independent short sentence: analyze the dependency relationship, extract the independent functional components in the independent short sentences with the parallel relationship and independent structure, and build a multi-level semantic model for each independent functional component; The invention combines crowdsourcing design resource short sentence segmentation, part-of-speech tagging, dividing independent short sentences, dividing independent functional components for each independent short sentence, and constructing a multi-level semantic model for each independent functional component, so as to convert unstructured natural language. The short sentence description is converted into a structured relational set, which realizes the unified modeling of crowdsourcing design resources, and is of great significance for subsequent retrieval and matching.

Description

A Semantic Analysis Method for Crowdsourcing Design Resources

technical field

The invention belongs to the technical field of computer data processing, and in particular relates to a semantic analysis method for crowdsourcing design resources.

Background technique

Crowdsourcing design resources are mostly expressed in a mixture of text, numbers, and image data on Internet crowdsourcing platforms, and the self-organization of crowdsourcing services makes the related content between design resources very different. To manage design resources well, it is necessary to The related service content involved is structured.

SUMMARY OF THE INVENTION

The present invention proposes a semantic analysis method for crowdsourcing service resources, which takes the design resources on the Internet crowdsourcing platform as the research object, and divides the crowdsourcing design resources into independent functional components through phrase division and dependency analysis. Multi-level semantic representation is performed to convert unstructured natural language short sentence descriptions into structured relational sets for unified modeling of crowdsourced design resources.

The present invention adopts following technical scheme to realize:

A semantic analysis method for crowdsourced design resources is proposed, including:

Step 1. Perform word separation and part-of-speech tagging on short sentences of crowdsourced design resources;

Step 2. Divide the short sentences processed in step 1 into multiple independent short sentences;

Step 3. Perform the following processing for each independent short sentence:

1) Dependency analysis;

2) Extract independent functional components in independent short sentences with juxtaposition and independent structure;

3) Build a multi-level semantic model for each independent functional component:

(1) Traverse independent functional components to find the core verb predicate, and obtain the first-level semantics with the detected generalized verb-object relationship; the generalized verb-object relationship includes direct object relationship, indirect object relationship, prepositional object relationship, core verb predicate and its domination The noun phrases are adjective-centered, the core verb-predicate and the noun-phrase it dominates are subject-predicate, and when the object of the core verb-predicate is missing, it is a modifier expressed with a fixed-center relationship;

(2) Detect the modifier components of the predicate center word and the object center word in the first-level semantics to obtain the second-level semantics;

(3) Detecting the modified components of the center word of the second-level semantics to obtain the third-level semantics;

(4) Complete the detected semantic components higher than level 3 to the head word they belong to.

Further, in step 2, the short sentence is divided into multiple independent short sentences by spaces, commas, commas and forward and backward slashes.

Further, the independent functional components in the independent short sentences are extracted by the juxtaposition relationship and the independent structural relationship, including:

(1) When the detected juxtaposition is located in the generalized object modification area, it is equally divided into two independent functional components based on the core verb predicate;

(2) When the detected juxtaposition is between the core verb predicates, it is equally divided into two independent functional components;

(3) When a parallel relationship and an independent structure coexist, if the independent structure is a verb predicate and there is a parallel relationship, it is equally divided into two independent functional components based on the verb predicate of the independent structure; if the independent structure is not When it is a verb predicate, no independent functional component is established.

Further, a multi-level semantic model is constructed for each independent functional component, including:

When it is detected that there is no core object in the first-level semantics, its modified part is regarded as its object and built into the first-level semantics.

Compared with the prior art, the advantages and positive effects of the present invention are: in the method for semantic analysis of crowdsourced design resources proposed by the present invention, the crowdsourced design resource short sentences are segmented into words, part-of-speech tagging is performed, independent short sentences are divided, and each short sentence is divided into separate sentences. Each independent short sentence is divided into independent functional components, and a multi-level semantic model is constructed for each independent functional component, so as to convert the unstructured natural language short sentence description into a structured relational set, and realize the unified modeling of crowdsourcing design resources. , which is of great significance to subsequent retrieval and matching.

Other features and advantages of the present invention will become more apparent upon reading the detailed description of the embodiments of the present invention in conjunction with the accompanying drawings.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a schematic diagram of analysis steps of a semantic analysis method for crowdsourcing design resources proposed by the present invention;

Fig. 2 is one of the division schematic diagrams of dividing independent short sentences in the present invention;

Fig. 3 is the second schematic diagram of dividing independent short sentences in the present invention;

4 is the third schematic diagram of the division of independent short sentences based on the dependency analysis tree in the present invention;

5 is one of the schematic diagrams of the division of independent functional components based on a dependency analysis tree in the present invention;

6 is the second schematic diagram of the division of independent functional components based on the dependency analysis tree in the present invention;

7 is the third schematic diagram of the division of independent functional components based on the dependency analysis tree in the present invention;

8 is a schematic diagram of the construction steps for constructing a multi-level semantic model in the present invention;

9 is one of the schematic diagrams of constructing a multi-level semantic model in the present invention;

10 is the second schematic diagram of constructing a multi-level semantic model in the present invention;

11 is the third schematic diagram of constructing a multi-level semantic model in the present invention;

FIG. 12 is the fourth schematic diagram of constructing a multi-level semantic model in the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

In the description of the present invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", The orientation or positional relationship indicated by "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying The device or element referred to must have a particular orientation, be constructed and operate in a particular orientation, and therefore should not be construed as limiting the invention.

In the description of the present invention, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense, unless otherwise expressly specified and limited, for example, it may be a fixed connection or a detachable connection connected, or integrally connected. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood in specific situations. In the foregoing description of the embodiments, the particular features, structures, materials or characteristics may be combined in any suitable manner in any one or more of the embodiments or examples.

The terms "first" and "second" are only used for descriptive purposes, and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may expressly or implicitly include one or more of that feature. In the description of the present invention, unless otherwise specified, "plurality" means two or more.

The semantic analysis method for crowdsourcing design resources proposed by the present invention, as shown in Figure 1, includes:

Step S1: Perform word separation and part-of-speech tagging on short sentences of crowdsourced design resources.

In Chinese, a character is the basic unit of a word. To understand the meaning of a phrase and a sentence, it must be divided into words. This requires the computer to firstly mark the sentence when processing a text, that is, automatically identify each word and put it in the Boundary markers are added to separate words.

In practical engineering, the commonly used method uses rule algorithm for word segmentation, and then uses statistical method to assist, which can efficiently and accurately segment the text, and can also take into account the recognition of new words and unentered words, which is the hybrid word segmentation method.

For example, the word segmentation result for the short sentence "writing and designing various brush fonts" describing the function of a crowdsourced resource is: [various, writing brush, font, writing, , , design].

After participle, part of speech needs to be tagged. Part of speech tagging is to determine and tag each word in the sentence, and divide each word into nouns, verbs, adjectives, and auxiliary words.

In Chinese, many parts of speech are not single, and different parts of speech are expressed in different sentences. On the other hand, most words often only have one or two parts of speech, and one part of speech is used in two (or more) parts of speech frequency is much higher than others. There is no unified standard for Chinese part-of-speech tagging. The two mainstream tags are the Peking University POS tagging set and the Pennsylvania POS tagging set. The following are examples of tagging parts of speech:

The result of the part-of-speech division of "various brush fonts to write and design" is: [various/rz, brush/n, font/n, writing/v, , /w, design/vn].

The part-of-speech tags of the words involved mainly fall into the following categories: 1) noun (n); 2) verb (v); 3) punctuation (w); 4) adjective (a); 5) conjunction (c); 6 ) pronoun (r); 7) quantifier (q); 8) other (o). There are more detailed divisions: gerunds (vn); intransitive verbs (vi); food (nf); demonstrative pronouns (rz).

Step S2: Divide the short sentence processed in step S1 into a plurality of independent short sentences.

In the design resource information of crowdsourcing websites, punctuation marks are very common, such as "I can design logo drawings, product shapes, etc.", "red wine, food, physical photography", "homepage design/professional PS cutout/watermark removal/advertising image" Design/baby description", and the functions of these punctuation marks are mostly to divide short sentences, so the present invention divides short sentences into multiple independent short sentences by spaces, commas, commas, forward and backward slashes, etc., without punctuation Short sentences separated by symbols do not need to be divided.

E.g:

1) Phrases separated by slashes: "Website construction/graphic design/Baidu ranking/Weibo marketing speak with quality!", the segmentation result is shown in Figure 2.

2) Phrases separated by spaces: "Food dishes, desserts, desserts, beverages, beverages, and sodas, and still life products." The result of segmentation is shown in Figure 3.

3) Phrases separated by commas: "undertake various brochures, menu layout design", the segmentation result is shown in Figure 4.

Step S3: Perform the following processing for each independent short sentence:

1. Dependency analysis.

Dependency is the relationship between words, that is, a binary asymmetric relationship between a head word and its subordinate words, the head word of a sentence is usually a verb (Verb), and all other words depend on the head word.

The following syntactic dependencies are mainly used in crowdsourced resources: 1) Collocation (COO); 2) Compound Nouns (FM); 3) Other Nouns (QM); 4) Direct Object (VOB); 5) Indirect Object (IOB); 6) Prepositional Object (FOB); 7) Adjective relationship (ADV); 8) Subject-predicate relationship (SBV); 9) Definitive relationship (ATT); (10) Independent structure (IC).

The various relationships are defined as follows:

Co-ordinate relationship (COO) represents the word relationship connected with "and, and, or", which is used for the next processing of its coordinating noun or verb, such as, "detail page and home page design": {(detail page /n->homepage/ n, COO).

Compound nouns (FM) represent the relationship between nouns and nouns, and are used to describe the object of the verb after splitting. Dessert/n -> Dessert/n, FM).

Other nouns (QM) represent the relationship of other part-of-speech modified nouns other than nouns, describe the attributes of nouns, and are used for sub-category, such as, "small program development": {(small/a->program/n, QM).

The direct object (VOB) represents the direct object followed by the predicate verb. Both the direct object and the prepositional object are classified into a verb-object structure, which is the basic relationship of dependencies in crowdsourced corpus, for example, "send her a bunch of flowers": { (send/v–>flowers/n, VOB)}.

The Prepositional Object (FOB) represents the object before the predicate verb, eg, "small program development": {(program/n – development/v, FOB).

Indirect Object (IOB) represents the personal pronoun after the predicate verb, which occurs occasionally in crowdsourced corpora, such as, "send her a bouquet of flowers": {(send/v -> her/n, IOB)}, but this structural meaning Not much, won't show up in the final dependencies.

The subject-verb relationship (SBV) represents the relationship between the agent and the action in the sentence, for example, "I can help you...": {(i/r –> help/n, SBV)}.

An adjective relationship (ADV) represents a relationship in which a verb modifies a verb, for example, "I can help you...": {(can/v –> help/v, ADV)}.

Attributive relationship (ATT) represents the relationship between attributive and head, for example, "system security": {(system/n -> security/adj, ATT)}.

Independent Structure (IC), which means that two independent semantic components are connected, such as "Website Development Website Construction": {(Website Development -> Website Construction, IC)}.

2. Extract independent functional components in independent short sentences with juxtaposition and independent structure extraction.

The functional description of crowdsourced design resources can be short or long, and complex descriptions usually include multiple core business functions, which can also be referred to as having multiple independent functional components without any punctuation between each component. For example: " Penetration testing, vulnerability scanning, website security detection, intrusion detection security testing”, “Mobile phone mall website construction, online shopping, online payment, and shopping website development”, etc. This description shows that it is a collection of multiple business functions. Analysis and extraction of independent functional components of independent short sentences.

Specifically, based on the analysis of the dependency relationship, the independent functional components are divided according to the parallel relationship (COO) and the independent structure (IC). Due to the different positions of the COO, IC and other relationships, different semantic relationships will result. The present invention is based on the following three situation to divide independent functional components.

(1) When the detected co-relationship (COO) is located in the generalized object modification region, it is equally divided into two independent functional components based on the core verb predicate.

Taking "WeChat and system security test" as an example, the dependency analysis tree is shown in Figure 5. According to the understanding of semantics, based on the core verb "test", the parallel objects "WeChat" and "system" are combined with the core verb "test". "Test" is equally divided into two separate functional components:

Independent functional component 1: (wherein the construction of multi-level semantics will be detailed in the following content)

First-level semantics: {(test, safe, (att))}

Secondary semantics: {(WeChat, Security, (ATT))}

Independent functional ingredient 2:

First-level semantics: {(test, safe, (att))}

Secondary semantics: {(system, security, (att))}

(2) When the detected juxtaposition is between the core verb predicates, it is equally divided into two independent functional components.

Taking "penetration testing vulnerability scanning website security detection intrusion detection security testing" as an example, this description is long and unclear, but the dependency analysis tree can express the semantic logical relationship. Based on the dependency analysis tree shown in Figure 6, it can Three core verb predicates "test 1", "test", and "test 2" were detected, among which tests 1 and 2 are distinguished by the same name:

Pairs and two core verb predicates form a parallel relationship, and are equally divided into two independent functional components. In this embodiment, three independent functional components are divided: (The construction content of multi-level semantics will be described in detail in the following content)

Independent functional ingredient 1:

Level 1 Semantics: {(Test1, Penetration, (ATT)), (Test1, Scan, (ATT))}

Secondary semantics: (vulnerability, scan, (ATT))

Independent functional ingredient 2:

Level 1 Semantics: {(Detect, Site, (ATT), (Detect, Intrusion, (ATT)), (Detect, Secure, (ATT))}

Independent functional ingredient 3:

First-level semantics: {(test2, safe, (att))}

(3) When a parallel relationship and an independent structure coexist, if the independent structure is a verb predicate and there is a parallel relationship, it is equally divided into two independent functional components based on the verb predicate of the independent structure; if the independent structure is not When it is a verb predicate, no independent functional component is established.

Taking "Shanghai company's corporate social security account opening professional registered accounting and tax declaration agent with rich experience and value-added special offer" as an example, based on the dependency analysis relationship tree shown in Figure 7, one core verb "open an account" and four independent structures "register" can be detected. ", "bookkeeping", "agent", "experienced" and "special offer", where the independent structures "registration", "bookkeeping" and "agent" are all verb predicates and the two are in a parallel relationship, then the "Registration", "Accounting" and "Agent" are equally divided into three independent structures based on the same basis, and a total of four independent functional components are divided. Independent functional ingredients:

Independent functional ingredient 1:

Level 1 semantics: {(open account, social security, (SBV))}

Secondary semantics: {(Enterprise, Social Security, (ATT))}

Level 3 semantics: {(Shanghai Company, Enterprise, (ATT))}

Independent functional ingredient 2:

First-level semantics: {(register, ∅, (∅))}

Secondary semantics: {(professional, registered, (ADV))}

Independent functional ingredient 3:

First-level semantics: {(bookkeeping, ∅, (∅))}

Independent functional ingredient 4:

First-level semantics: {(agent, ∅, (∅))}

Secondary semantics: {(tax, agent, (adv))}

3. Build a multi-level semantic model for each independent functional component.

The present invention builds a multi-level semantic model based on the following predefined definitions:

The first-level semantics takes the verb predicate as the core word, and the noun phrase as the subordinate word, expresses the basic semantics of the sentence through the direct object relationship (VOB), indirect object relationship (IOB), front object relationship (FOB), etc., generally representing service A primary function provided by the resource.

On the basis of the first-level semantics, the second-level semantics modifies and supplements the respective central words through the adjective-median relationship (ADV), the verb-complement relationship (CMP), and the prepositional-object relationship (POB).

On the basis of the second-level semantics, the third-level semantics continues to modify and supplement the central words of the second-level semantics, making the semantics richer and fuller.

The natural language of the service resource expressed in phrases can basically express the full meaning of the resource through the third-level semantics, and the grammatical components higher than the third-level can be processed by truncation or completion to the third-level central word.

When constructing first-level semantics, considering the error of dependency analysis, many relationships that should be verb-object classes from the actual semantics will be identified as attentive (ATT) relationships. Therefore, for first-level semantics, if a core verb is detected When the gender predicate and its dominant noun phrase are in the ATT relation, it is also regarded as an action-object relation. The core verb predicate and its dominant noun phrase are subject-verb (SBV) relationship, which is also regarded as verb-object relationship. When the object of the core verb predicate is missing, it may be caused by the expression of the language itself or the error of the dependency analysis model itself. In this case, the semantics is mainly expressed by the core predicate and its modified parts, and the present application In , the modifications expressed in the attentive (ATT) relationship are also regarded as objects, located in the first-level position.

Therefore, when constructing the first-level semantics, the present application takes the direct object relationship, indirect object relationship, pre-object relationship, core verb predicate and its dominant noun phrase as adjective-in-the-middle relationship, and core verb predicate and its dominant noun phrase as main predicate. The relation and the modifiers expressed by the central relation when the object of the core verb predicate is missing are collectively referred to as the generalized verb-object relation. The reason for this design is that in both cases, the dominant noun can be regarded as the subject role, and the verb predicate can be combined to fully express the semantics, so as to obtain as general a semantic representation model as possible.

Then based on the above-mentioned predefined methods, the present invention proposes a method for constructing a multi-level semantic model, as shown in Figure 8, including:

1. Traverse the independent functional components to find the core verb predicate, and obtain the first-level semantics with the detected generalized verb-object relationship.

The generalized verb-object relationship here includes direct-object relationship, indirect-object relationship, pre-object relationship, central verb-predicate and the noun phrase governed by it as adjective, the core verb-predicate and its governed noun phrase as subject-predicate relationship, when the core verb predicate A modifier expressed in a definite relation when the object of is missing.

2. Detect the modifier components of the predicate center word and the object center word in the first-level semantics to obtain the second-level semantics.

Modification and supplementation of predicate head words and object head words are carried out through adjective-median relation (ADV), verb-complement relation (CMP) and prepositional-object relation (POB).

3. Detect the modified components of the central word of the second-level semantics to obtain the third-level semantics.

The center word of the second-level semantics is different from the predicate-centered word and the object-centered word in the first-level semantics, and is the modified supplementary part of the two. If there are no other modifiers in the modified supplementary part, the third-level semantics will not be constructed.

4. Complete the semantic components higher than the third level to the head word to which it belongs.

The entire extraction process is performed by traversing the dependency analysis tree from the bottom to the top in the opposite direction of the dependency analysis tree relationship arc.

Taking "homemade handmade top-level cleansing oil for skin care products" as an example, the dependency analysis tree is shown in Figure 9. The core verb "self-made" and its object "cleansing oil" constitute the first-level semantics, and the predicate head word "self-made" in the second-level semantics There is no modified part, the modified parts of the object head word "cleansing oil" "top" and "skin care product" are sufficient to form secondary semantics, and the central word "skin care product" in the secondary semantics has a modified part, then "skin care product" Three-level semantics are constructed for the central word, and the multi-level semantic model obtained according to the preceding steps is as follows:

Level 1 semantics: {(homemade, cleansing oil, (VOB))}

Secondary semantics: {(Top, Cleansing Oil, (ATT)), (Skincare, Cleansing Oil, (ATT))}

Level 3 semantics: {(handmade, skincare, (att))}

It can be seen that the first-level semantics reflect the main function of the service resource, and the second- and third-level semantics are the modified parts of the core words.

In practical applications, although the resource description of a single independent functional component has only one core business function, it is also divided into the following two situations:

(1) There is a core object at the first level

Taking "low-cost rental of temporary network space" as an example, the syntax tree obtained through dependency analysis is shown in Figure 10, and its multi-level semantic model is given as:

First-level semantics: {(rent, cyberspace, (VOB))}

Secondary semantics: {(low, rent, (ADV)), (temporary, cyberspace, (ATT))}

(2) There is no core object at the first level

As mentioned above, the lack of the object of the core verb predicate may be caused by the expression of the language itself or the error of the dependency analysis itself. In this case, the semantics are mainly expressed by the core predicate and its modified parts, and the modification expressed in the attentive (ATT) relationship in the present invention is still regarded as an object, which is located at the position of the first-level semantics.

Taking "website security detection" as an example, its dependency relationship analysis tree is shown in Figure 11, and its multi-level semantic model is:

First-level semantics: {(detect, secure, (att))}

Secondary semantics: {(site, security, (att))}

Based on the semantic analysis method for crowdsourced design resources proposed by the present invention, the crowdsourced design resources are segmented into short sentences, tagged with part of speech, divided into independent short sentences, divided into independent functional components for each independent short sentence, and for each independent functional component A multi-level semantic model is constructed to convert the unstructured natural language short sentence description into a structured relational set, which realizes the unified modeling of crowdsourcing design resources, and is of great significance for subsequent retrieval and matching.

Finally, taking "mobile phone mall website construction online shopping and online payment shopping website development" as an example, the semantic analysis method of crowdsourcing design resources proposed by the present invention is applied to obtain a structured relationship set, and the dependency relationship analysis tree is shown in Figure 12. Structured processing include:

1. Perform word segmentation and part-of-speech tagging on the short sentence.

2. Divide the short sentence after processing 1 into multiple independent short sentences through spaces, commas, commas and forward and backward slashes;

The phrase has no spaces, commas, commas, and forward and backward slashes, so skip this step and treat it as a single phrase as a whole.

3. Carry out dependency analysis on the independent short sentence.

4. Extract the independent functional components in the independent short sentence with juxtaposition and independent structure.

Traversing the independent short sentence, it is retrieved that its core verb predicates are "construction" and "development", and it is divided into two independent functional components with "construction" and "development" as the core verb predicates according to the parallel relationship (COO): Independent functional component 1 (the core verb predicate is "construction") and independent functional component 2 (the core verb predicate is "development").

There are no independent structures such as independent verbs and independent nouns in the independent short sentence, so there is no need to divide independent functional components according to independent structures.

5. Build a multi-level semantic model for each independent functional component.

For the independent functional component 1, the first-level semantics {(construction, website1, (ATT))} is obtained according to the generalized verb-object relationship, and the predicate-centered word "construction" and the object-centered word "website" in the first-level semantics are searched for The modified part gets the second-level semantics {(mall, website 1, (ATT))}, and the modified part is retrieved for the central word "mall" in the second-level semantics to obtain the third-level semantics {(mobile phone, mall, (ATT))}, The multi-level semantic model is obtained as:

Independent functional ingredient 1:

Level 1 semantics: {(build, site1, (att))}

Secondary semantics: {(Mall, Site1, (ATT))}

Level 3 semantics: {(mobile, mall, (ATT))}

For independent functional component 2, obtain its first-level semantics {{(Development, Website2, (ATT))} according to the generalized verb-object relationship, and search for the predicate-centered word "Development" and the object-centered word "Website" in the first-level semantics Its modified part gets secondary semantics {(online shopping, website 2, (ATT)), (shopping, website 2, (ATT))}, and the modified part is retrieved for the central word "shopping" in the secondary semantics to get the third level Semantic {{(online payment, shopping, (ADV))}, its multi-level semantic model is obtained as:

Independent functional ingredient 2:

First-level semantics: {(dev, site2, (att))}

Secondary Semantics: {(Online Shopping, Site2, (ATT)), (Shopping, Site2, (ATT))}

Three-level semantics: {(online payment, shopping, (ADV))}

The above websites 1 and 2 are distinguished by the same name.

It should be noted that the dependency analysis tree mentioned in the above content of the present invention is another existing analysis means for semantic analysis, and is not a technical means limited by the present invention, and can support the semantic analysis method of the present invention by using existing methods to realize .

It should be noted that, in the specific implementation process, the above-mentioned control part may be implemented by a processor in the form of hardware executing computer-executed instructions in the form of software stored in the memory, which will not be repeated here, and the actions performed by the above-mentioned control circuit The corresponding programs can be stored in the computer-readable storage medium of the system in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

The computer-readable storage medium above can include volatile memory, such as random access memory; can also include non-volatile memory, such as read-only memory, flash memory, hard disk or solid-state hard disk; can also include the above-mentioned types combination of memory.

The processor mentioned above may also be a collective term for multiple processing elements. For example, the processor may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. . A general purpose processor may be a microprocessor or it may be any conventional processor, etc., but also a special purpose processor.

It should be pointed out that the above description is not intended to limit the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by those of ordinary skill in the art within the essential scope of the present invention, It should also belong to the protection scope of the present invention.

Claims (4)

1. A semantic analysis method for crowdsourced design resources is characterized by comprising the following steps:

step 1, carrying out word separation and part-of-speech tagging on crowdsourced design resource short sentences;

step 2, segmenting the short sentence processed in the step 1 into a plurality of independent short sentences;

and 3, aiming at each independent short sentence, carrying out the following processing:

1) analyzing the dependence relationship;

2) extracting independent functional components in the independent short sentences in a parallel relation and an independent structure;

3) constructing a multilevel semantic model for each independent functional component:

(1) traversing the independent functional components to find out the core dynamic predicates, and obtaining primary semantics according to the detected generalized dynamic guest relationship; the generalized verb relationship comprises a direct object relationship, an indirect object relationship, a preposed object relationship, a core predicate and noun phrases dominated by the core predicate as a shape-middle relationship, a core predicate and noun phrases dominated by the core predicate as a main-predicate relationship, and modifiers expressed in a middle relationship when the object of the core predicate is absent;

(2) detecting the modification components of predicate central words and object central words in the primary semantics to obtain secondary semantics;

(3) detecting the modified components of the central words of the secondary semantics to obtain tertiary semantics;

(4) and completing the detected semantic components higher than three levels to the subordinate central words.

2. The method for semantic analysis of a crowdsourced design resource of claim 1, wherein in step 2, the phrases are segmented into a plurality of independent phrases by spaces, pause signs, commas and forward and backward slashes.

3. The method for semantic analysis of crowdsourced design resources according to claim 1, wherein extracting independent functional components in independent phrases in a parallel relationship and an independent structural relationship comprises:

(1) when the detected parallel relation is located in the generalized object modification area, equally dividing the detected parallel relation into two independent functional components on the basis of the core dynamic predicate;

(2) when the detected parallel relation is positioned between the core dynamic predicates, equally dividing the core dynamic predicates into two independent functional components;

(3) when the parallel relationship and the independent structure exist at the same time, if the independent structure is a verb predicate and the parallel relationship exists, the independent structure is equally divided into two independent functional components based on the verb predicate of the independent structure; if the independent structure is not a verb predicate, no independent functional component is created.

4. The method for semantic analysis of crowdsourced design resources of claim 1, wherein building a multilevel semantic model for each independent functional component comprises:

when the core object is detected not to exist in the primary semantic, the modified part of the core object is regarded as the object and is built into the primary semantic.

Images