CN113297089B - Knowledge graph-based mass measurement assistant implementation method - Google Patents

Abstract

The invention discloses a knowledge graph-based mass measurement assistant implementation method. A data acquisition stage, namely determining keywords related to the testing field; in the data preprocessing stage, corresponding rules are designed to extract triples in the content, aliases and foreign names are used for entity alignment and combined with a general knowledge graph CN-DBpedia; the data storage stage, which is to store data in a graph database Neo4j, and embed triples into the graph database by using a Cypher query statement; in the question analysis stage, from the input of the user, a slot value pair, a slot corresponding entity, a value corresponding relation or attribute are extracted from the input by using template matching, a corresponding result is queried by using py2neo embedded Cypher sentences, and the result is returned to the user in a chat format. The invention takes the graph database as a storage structure, meets the maximum delay required by the intelligent assistant, and the knowledge graph in the field can enable the intelligent assistant to provide better help for the testers.

Description

Knowledge graph-based mass measurement assistant implementation method

Technical Field

The invention belongs to the field of crowdsourcing test, and particularly relates to a knowledge graph-based crowdsourcing assistant implementation method.

Background

With the development of internet technology, it is possible to hire different platforms and test persons in different locations. Crowd-sourced testing is an emerging trend in the testing field, and the advantages of the crowd-sourced and cloud platforms are fully utilized. The test pool can be expanded, the prejudice of internal testers is avoided, and the product company only pays for the reported effective loopholes, so that the test cost is reduced.

In the traditional testing field, testers are required to have more specialized ability quality, including testing technology, field knowledge, even related experience, and the like. The crowding platform reduces the test admission threshold, but brings greater uncertainty, and the capability of the crowding members is uneven, so that the high-quality completion of the crowding tasks is challenged. How to balance the testing skills of testers, find more effective loopholes, improve the quality of software and face great challenges.

The current solutions focus on scoring the testers, drawing capability records of the testers, credit records. Judging the testing technology according to the historical completion condition of the user; meanwhile, the test difficulty of each test task is divided, a specific scheduling algorithm is selected, and the proper task is distributed to the proper test object. However, this does not address the vulnerability of non-professional testers to often commit invalidity and repetition. By constructing an automatic test and crowd-sourced test model of the fusion field characteristics, a man-machine cooperation-feedback mechanism is realized, a manual cooperation test technology with machine efficiency and manual field characteristics is formed, and the connotation of crowd-sourced test is generalized and extended deeply, so that comprehensive support is provided for obtaining high-quality test results.

Disclosure of Invention

The invention aims to provide a knowledge graph-based mass measurement assistant implementation method, which aims to solve the technical problems of uneven capability of testers in the mass measurement field and low quality of test results.

In order to solve the technical problems, the specific technical scheme of the invention is as follows:

a knowledge graph-based mass measurement assistant implementation method is characterized by comprising the following steps:

step 1, data acquisition, namely acquiring keywords related to a test;

step 2, preprocessing data, extracting a webpage sentence pattern by using a regular expression, extracting triples in the webpage sentence pattern by using a rule, aligning aliases and foreign names with entities, and merging the aliases and the foreign names with a universal knowledge graph DBpedia to obtain a knowledge graph of the test field;

step 3, data storage, namely storing the data in a graph database Neo4j by using a Cypher query statement;

and 4, analyzing the question, analyzing the slot value pairs in the question of the user, querying a graph database, and returning the result to the user according to a fixed sentence pattern.

Further, the step 1 of obtaining the keyword includes the following steps:

step 1.1, searching a concept and a method of testing from books;

step 1.2, answering questions related to the test and completing keywords in an open answer webpage;

and 1.3, crawling the keywords, judging whether the entity of the triplet obtained by analysis belongs to the field of testing by using a manual mode, and if so, adding the entity into a keyword library.

Further, the step 2 specifically includes the following steps:

step 2.1, obtaining a rule of a triplet comprises: extracting triples from the structured information bar, extracting triples from specific sentence patterns meeting the requirements, and manually supplementing the triples;

step 2.2, entity alignment: aligning the aliases and English names in the information column to the keywords, and automatically aligning to the names when inquiring the aliases and English names;

and 2.3, combining the atlas, and combining the knowledge atlas of the testing field with the knowledge atlas of the DBpedia of the general field.

Further, the data storage in step 3 specifically includes the following steps:

step 3.1, adding a Cypher statement into py2Neo, and storing the triples into a graph database Neo4 j;

and 3.2, constructing an inverted index table, taking the alias of each entity as a key, taking the name of the entity as a value, and storing the name in the table.

Further, the question parsing in the step 4 specifically includes the following steps:

step 4.1, searching out a slot value pair by matching in a template-based mode, wherein the slot corresponds to an entity, a value corresponding relation or an attribute;

step 4.2, word segmentation is carried out on input by using jieba, parts of speech are obtained, keywords and relations are submitted to a jieba word stock as special words, confidence is given, and correct word segmentation of the jieba is ensured;

step 4.3, after word segmentation, acquiring the part of speech of each word;

step 4.4, a questioning form is carried out on the user, corresponding templates are matched, the entity appears in the form of nouns and proper nouns, the relation/attribute generally appears in the form of nouns and proper verbs, and the middle is connected by a conjunctive, namely, the first type of templates are considered: n|\vn+c+ \v\n;

step 4.5, the second class of questions are different from the first class, namely questions are asked by replacing specific attributes with query words; converting the query words into attributes, capturing the entities therein, and establishing a corresponding second class template as follows:

step 4.5.1, replace "what" with "definition" or "content";

step 4.5.2, replacing "how\how\with" content "or" purpose ";

step 4.5.3, replacing "why" with "meaning" or "purpose";

step 4.6, the replacement words have sequence, the replacement word placed in front is searched earlier, and if the attribute does not exist, the replacement words are pushed backwards in sequence;

and 4.7, searching the map database and returning the result to the front end.

The knowledge graph-based mass measurement assistant implementation method provided by the invention has the following advantages:

by using the graph database Neo4j to store the triples, compared with the traditional relational database, the query result can be given in ten million data in one second, and the real-time requirement of the public test assistant is met. By using the inverted index table, the time complexity of entity alignment is O (1); question analysis converts the user input into slot value pairs which can be identified by the graph database, so that the accuracy of assistant identification is improved.

Drawings

FIG. 1 is a flowchart of a knowledge graph-based crowd measurement assistant implementation method of the present invention;

FIG. 2 is a flow chart of knowledge graph construction of the present invention;

FIG. 3 is a flow chart of question parsing according to the present invention.

Detailed Description

In order to better understand the purpose, structure and function of the present invention, a knowledge-graph-based public measurement assistant implementation method of the present invention is described in further detail below with reference to the accompanying drawings.

As shown in fig. 1, the present invention includes the steps of:

step 1, data acquisition, namely acquiring enough keywords related to testing, wherein the keywords are used as keywords for inquiring hundred degrees encyclopedia and wikipedia. As much as possible, all keywords in the domain. The method for determining the keywords comprises the following steps:

step 1.1, searching test related concepts, methods and the like from test related books;

step 1.2, answering questions related to the test in an open answer webpage, and completing the keywords possibly missing in the step 1.1;

step 1.3, crawling a keyword, often obtaining more keywords, judging whether the entity of the analyzed triplet belongs to the testing field or not by using a manual mode, and if so, adding the entity into a keyword library;

step 2, preprocessing data, extracting webpage content by using a regular expression, extracting triples in the webpage content by using a rule, simultaneously aligning aliases and Chinese names with entities, and merging the aliases and Chinese names with a universal knowledge graph DBpedia to obtain a knowledge graph of the test field, wherein the detailed method is as follows:

step 2.1, extracting text contents of the webpage by using the regular expression;

step 2.2, obtaining the rule of the triplet includes: taking the structured information column as a triplet, extracting the triplet from the specific sentence pattern meeting the requirements, and manually supplementing the triplet. Two specific sentence patterns meeting the requirements exist, and the following are adopted:

and 2.2.1, taking the catalogue as an attribute, and taking sentences of each calibration serial number under the catalogue as values. For example, the black box test, and (1) whether there is a functional error or not, and whether there is a functional omission. (2) Whether the input data can be correctly received and produce the correct output result. (3) Whether there is a data structure error or an external information access error. Thus, three triples (black box test, function, whether there is a functional error, whether there is a functional omission), (black box test, function, whether input data can be correctly received and a correct output result can be generated), (black box test, function, whether there is a data structure error or an external information access error);

step 2.2.2, second sentence pattern: b of A includes C, D, E or B of A has C, D, E. For example, the test case design method of the black box test includes equivalence class classification, boundary value analysis, misspeculation, causal graph, and the like. Forming four triplets (black box test, design method, equivalence class division method), (black box test, design method, misspeculation method), (black box test, design method, causal graph method), (black box test, design method, boundary value analysis method);

step 2.3, entity alignment: aligning the aliases and English names in the information column to the keywords, and automatically aligning to the names when inquiring the aliases and English names;

2.4, combining the atlas, namely combining the knowledge atlas of the testing field with the knowledge atlas of the general field DBpedia to obtain a field knowledge atlas with general knowledge;

and 3, storing the data in a graph database Neo4j by using a Cypher query statement. The specific method comprises the following steps:

step 3.1, adding a Cypher insert sentence into py2Neo, and storing the triplet into map data Neo4 j;

and 3.2, constructing an inverted index table, taking the alias of each entity as a key, taking the name of the entity as a value, and storing the name in the table. For example, the alias function test for the black box test is stored in dic with "function test" as a key and "black box test" as a value. Thus, the name can be obtained through dic [ 'functional test' ]. The inverted index table can find the name in the time of O (1), so that the searching speed is increased;

and 4, analyzing the question, as shown in fig. 3. And analyzing the slot value pairs in the user question, querying a graph database, and returning the result to the user according to the fixed sentence pattern. The specific method comprises the following steps:

and 4.1, searching out a slot value pair by matching in a template-based mode. The slot corresponds to an entity, value correspondence or attribute;

step 4.1.1, word segmentation is carried out on input by using jieba, word parts are obtained, and as the jieba has poor word segmentation effect on the test field, keywords and relations are submitted to a jieba word stock as special words, and higher confidence is given to ensure that the jieba performs word segmentation correctly;

step 4.1.2, after word segmentation, obtaining the part of speech of each word, wherein the common part of speech of jieba is \n is noun, \v is dynamic, \vn is proper noun, \x is non-morpheme word,_v is structure aid word and the like;

step 4.1.3, consider the form of multi-user questions as much as possible, and match the corresponding templates, in general, the entity appears in terms of nouns and proper nouns, the relationship/attribute appears in terms of nouns and proper verbs, and the middle is connected by a conjunctive, i.e. we can consider the first template: n|\vn+c+ \v\n. For example: what is the definition of the white box test?

Step 4.2, the second class of questions is different from the first class, i.e. questions are asked by replacing specific attributes with query words, for example, what is a white box test, at this time, the relation or the attributes cannot be extracted from the question sentences, we need to convert the query words into the attributes, capture the entities therein, and build a corresponding second class template. The following are provided:

step 4.2.1, replace "what" with "definition" or "content", such as "what is the definition of the white box test" replaced to "white box test";

step 4.2.2, replacing "how\how\with" content "or" purpose ", such as" how test for white box test "with" purpose for white box test ";

step 4.2.3, replacing "why" with "meaning" or "purpose", such as "why the white box test" is replaced with "meaning of white box test";

step 4.3, the replacement words have a certain sequence, the replacement words placed in front are searched earlier, if the attribute does not exist, the replacement words are pushed backwards in sequence, for example, what is a white box test, the query (white box test, definition) is performed, and if the white box test does not exist, the attribute called definition is performed, the query (white box test, content) is performed;

step 4.4, searching the graph database and returning the result to the front end, wherein the searching is divided into several cases, the first case is that the result is unique, for example, the definition of the white box test is only one node, and the +attribute/relation +of the returned entity +is a +value. For example, the definition of a white-box test is a test performed by a user in a development environment. If multiple results are returned, the +relationship/attribute +having +a+b+c … … in the format of entity +is returned. For example, the α test method is a black box test, a white box test, a pressure test, or a stress test.

In summary, the method for implementing the public testing assistant based on the knowledge graph provided by the invention can better answer the problems in the testing field, help the testers to work, control the time cost and return the result in tens of millions of triples in lower time. The overhead incurred by alias lookup is controlled to be O (1) by using the inverted index table.

It will be understood that the invention has been described in terms of several embodiments, and that various changes and equivalents may be made to these features and embodiments by those skilled in the art without departing from the spirit and scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from the essential scope thereof. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.

Claims (3)

1. A knowledge graph-based mass measurement assistant implementation method is characterized by comprising the following steps:

step 1, data acquisition, namely acquiring keywords related to a test;

step 2, preprocessing data, extracting a webpage sentence pattern by using a regular expression, extracting triples in the webpage sentence pattern by using a rule, aligning aliases and foreign names with entities, and merging the aliases and the foreign names with a universal knowledge graph DBpedia to obtain a knowledge graph of the test field;

the step 2 specifically comprises the following steps:

step 2.1, obtaining a rule of a triplet comprises: extracting triples from the structured information bar, extracting triples from specific sentence patterns meeting the requirements, and manually supplementing the triples;

two specific sentence patterns meeting the requirements exist, and the following are adopted:

first sentence: taking the catalogue as an attribute, and taking sentences of each calibration sequence number under the catalogue as a value;

the second sentence pattern: b of A comprises C, D, E or B of A has C, D, E;

step 2.2, entity alignment: aligning the aliases and English names in the information column to the keywords, and automatically aligning to the names when inquiring the aliases and English names;

2.3, combining the atlas, namely combining the knowledge atlas of the testing field with the knowledge atlas of the DBpedia of the general field;

step 3, data storage, namely storing the data in a graph database Neo4j by using a Cypher query statement;

the data storage in the step 3 specifically comprises the following steps:

step 3.1, adding a Cypher statement into py2Neo, and storing the triples into a graph database Neo4 j;

step 3.2, constructing an inverted index table, taking the alias of each entity as a key, taking the name of the entity as a value, and storing the name in the table;

and 4, analyzing the question, analyzing the slot value pairs in the question of the user, querying a graph database, and returning the result to the user according to a fixed sentence pattern.

2. The knowledge graph-based crowd-sourced assistant implementation method of claim 1, wherein the keyword obtaining in step 1 comprises the following steps:

step 1.1, searching a concept and a method of testing from books;

step 1.2, answering questions related to the test and completing keywords in an open answer webpage;

and 1.3, crawling the keywords, judging whether the entity of the triplet obtained by analysis belongs to the field of testing by using a manual mode, and if so, adding the entity into a keyword library.

3. The knowledge graph-based crowd-test assistant implementation method of claim 2, wherein the question parsing in the step 4 specifically includes the following steps:

step 4.1, searching out a slot value pair by matching in a template-based mode, wherein the slot corresponds to an entity, a value corresponding relation or an attribute;

step 4.2, word segmentation is carried out on input by using jieba, parts of speech are obtained, keywords and relations are submitted to a jieba word stock as special words, confidence is given, and correct word segmentation of the jieba is ensured;

step 4.3, after word segmentation, acquiring the part of speech of each word;

step 4.4, the form of asking questions to the user is matched with corresponding templates, the entity appears in the form of nouns and proper nouns, the relation/attribute appears in the form of nouns and proper verbs, and the middle is connected by a conjunctive, namely, a first type of template is used: n|\vn+c+ \v\n;

n is a noun, v is dynamic, vn is a proper noun;

step 4.5, the second class of questions are different from the first class, namely questions are asked by replacing specific attributes with query words; converting the query words into attributes, capturing the entities therein, and establishing a corresponding second class template as follows:

step 4.5.1, replace "what" with "definition" or "content";

step 4.5.2, replacing "how\how\with" content "or" purpose ";

step 4.5.3, replacing "why" with "meaning" or "purpose";

step 4.6, the replacement words have sequence, the replacement word placed in front is searched earlier, and if the attribute does not exist, the replacement words are pushed backwards in sequence;

and 4.7, searching the map database and returning the result to the front end.

Images