CN112668719A - Knowledge graph construction method based on engineering capacity improvement - Google Patents
Knowledge graph construction method based on engineering capacity improvement Download PDFInfo
- Publication number
- CN112668719A CN112668719A CN202011226817.2A CN202011226817A CN112668719A CN 112668719 A CN112668719 A CN 112668719A CN 202011226817 A CN202011226817 A CN 202011226817A CN 112668719 A CN112668719 A CN 112668719A
- Authority
- CN
- China
- Prior art keywords
- engineering
- entity
- knowledge graph
- knowledge
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a knowledge graph construction method based on engineering ability improvement, and a series of effective ways for students to improve engineering ability are obtained by constructing a knowledge graph in the field of engineering education. In the process of constructing the knowledge graph, a named entity identification method based on the combination of Bilstm and CRF, a relation extraction method based on a BERT model and a negative triple potential correct probability knowledge inference algorithm based on a pre-training TransE model are provided. The invention extends the knowledge graph to each measure by taking engineering education as a starting point for visual display, and can more clearly and accurately display the content of each path.
Description
Technical Field
The invention belongs to the field of engineering education, and particularly relates to a knowledge graph construction method based on engineering capacity improvement.
Background
Currently, under the background of meeting global challenges and strengthening national core competitiveness, great culture of superior scientific and technological engineers has become a necessary requirement for rapid development of countries in the world, and the importance of engineering education as a core field for culturing superior engineering talents has increasingly emerged. The engineering education development and the globalization development are mutually promoted and mutually influenced. The method enhances the research strength in the field of domestic and foreign engineering education, quickly masters the hotspots, trends and characteristics of various researches, and has great significance for promoting the construction and reformation of the national engineering education.
In the field of engineering education, students can be helped to obtain ways for improving engineering ability through the knowledge graph, and the engineering ability level of the students is improved. And because the knowledge map is a reticular knowledge base, the knowledge map has strong expression capability, and can help students to quickly find an effective way for improving engineering capability and improve learning efficiency. Therefore, the knowledge graph is effectively utilized to help students to effectively learn independently, and the engineering ability is improved.
Through the analysis, in order to enhance the comprehensive strength of the country and cultivate more high-quality talents in the engineering field, an effective way for improving the engineering capacity of students needs to be obtained to cultivate more high-quality talents. The effective engineering capacity improving method needs knowledge base or knowledge map in specific field to extract specific information as basic knowledge support, just like experts need knowledge in professional field as support, so the invention introduces the knowledge map into engineering education field, constructs the knowledge map of engineering education and uses the knowledge map to find effective way for improving engineering capacity. The construction of knowledge maps by extracting information in specific fields is one of the key technologies of artificial intelligence, and is widely applied in the fields of automatic question answering, semantic search, personalized recommendation systems, content distribution and the like. The temperature rise of the related research and application of information extraction is a necessary result generated by the urgent need of artificial intelligence for deeply understanding data and processing data, and the development of the temperature rise depends on the intersection of multiple technologies, and relates to natural language processing, semantic networks, databases, text analysis and the like. The knowledge graph of the specific field is a relational network structure which analyzes, constructs and draws knowledge of the specific field through processes of text analysis, concept extraction, relation mining, ontology construction and reasoning, visualization and the like, is presented in a friendly graphic mode, and has the effect of one graph surpassing thousand languages. The knowledge graph is an improved knowledge organization mode and is an effective carrier for realizing knowledge network visualization. It is more intuitive and easy to understand, and more conforms to the cognitive rule of human from the outside to the inside and from the shallow to the deep. Through the structured knowledge in the knowledge map, people can avoid submerging in complicated and complicated junk information, avoid information overload, run directly on the theme, master core knowledge and relation networks thereof, and are favorable for deeply understanding knowledge and essence thereof. Therefore, the effective way of improving the engineering ability can be inquired through the knowledge map, and the invention visually displays the effective path of improving the ability, thereby facilitating students to intuitively find an effective method and improving the comprehensive quality.
Disclosure of Invention
The rapid development of the economic society puts higher requirements on the engineering capacity of students, and the difference between the assimilative childbearing goals of colleges and universities and the engineering reserve military cultured under a course system is widened and the social requirements are met. Therefore, the engineering capacity culture is strengthened, and an effective way for realizing engineering improvement is found, so that the method plays an extremely important role in the development of the modern society. In order to solve the problem, the invention provides a knowledge map construction system based on engineering capacity improvement, and provides a more intuitive and accurate effective way for the nation to cultivate high-quality engineering talents by constructing a knowledge map in the field of engineering education. The construction of the knowledge graph mainly comprises the steps of obtaining knowledge elements such as entities, attributes and relations from some open structured, semi-structured and unstructured data by using a knowledge extraction technology, and forming a structured semantic knowledge base by combining knowledge fusion, knowledge reasoning and other technologies. Knowledge extraction is the basis for constructing a knowledge graph and is mainly divided into two subtasks: entity extraction and entity relationship extraction. After the knowledge extraction work is completed, the acquired triple data is supplemented and perfected by adopting knowledge reasoning, and the complete knowledge map construction work is completed. The invention provides a method for combining a BILSTM-CRF model and a BERT model, extracting the data entity relation in the engineering education field and constructing a knowledge map in the engineering education field. The BILSTM can capture more information before and after sentences and solve the problem of gradient dispersion or gradient disappearance in training, all entities can be identified in the named entity identification process, and the CRF layer can automatically learn some constraint conditions in the training data, so that the validity of a final prediction result is ensured, namely whether all identified entities are effective entities can be judged according to weight values, and the accuracy of named entity identification is improved. The BERT model can increase the generalization capability of a word vector model, fully describe character-level, word-level, sentence-level and even sentence-level relation characteristics, is easy to perform migration operation, can capture a complete relation chain according to context in the project of entity relation extraction, and is very suitable for data in the field of engineering education.
Because some extracted data are always lost and the extracted data cannot be absolutely complete data, the invention provides a method for supplementing and perfecting the extracted knowledge graph data by using a knowledge inference model TransE, and realizes the continuous updating of the knowledge graph. Because the head entity is replaced randomly by the traditional TransE model, the positive triple problem may be included in the process of generating the negative triple by the tail entity or the relationship entity, so that the model cannot make correct reasoning. Therefore, the invention provides that the potential correct probability of the negative triples is used, the negative triples generated by randomly replacing the head entities, the tail entities or the relationship entities are scored according to the correct probability, and the training weights of the models are different by the negative triples with different scores. And when the model is trained, the weight of the triad in the negative triad, which is close to the positive triad score, is reduced, so that the model has an inference effect.
After the extracted knowledge graph data is supplemented and perfected through a knowledge inference model TransE, the knowledge graph data is stored by using a graph database Neo4j and visually displayed, so that students can visually check various promotion capacity paths, and the method plays a vital role in the promotion of the engineering capacity of the students and the talent culture plan of the country.
Drawings
FIG. 1 is a schematic diagram of the general construction of a knowledge graph according to the present invention.
FIG. 2 is a diagram of named entity identification and entity relationship extraction according to the present invention.
FIG. 3 is a schematic diagram of the knowledge inference of the present invention.
FIG. 4 is an example of the construction of the knowledge-graph based engineering capacity enhancement of the present invention.
Detailed Description
The present invention will be described in detail below with reference to specific embodiments shown in the drawings.
Fig. 1 is a general flow chart of construction of a knowledge graph according to the present invention, and as shown in fig. 1, in order to obtain data related to engineering capacity improvement in the engineering education field, web crawls are performed using a web crawler tool to crawl summary content of an article in the engineering education field and content in the encyclopedia engineering education field in the known network, and the known network data are first sorted and merged to obtain entity relationship data related to engineering capacity improvement. And then, combining a BILSTM-CRF model and a BERT model, carrying out model training on entity relationship coefficient data related to engineering capacity improvement in the knowledge network, carrying out named entity recognition on the Baidu encyclopedia engineering field data through the BILSTM-CRF model, then entering the BERT model for entity relationship extraction, and finally obtaining a series of triple data capable of realizing student engineering capacity improvement to complete construction work of the knowledge graph.
However, in the process of constructing the knowledge graph, many entities or relations of the triples are missing, so that the knowledge reasoning is adopted to complete the entity relations of the constructed knowledge graph. The TransE model has few parameters and low complexity and shows the characteristics of simplicity and high efficiency in constructing a large-scale knowledge graph, so that the constructed knowledge graph triple information is supplemented and perfected by selecting the TransE model. The invention provides a method for calculating the probability of potential positive triples in negative triples for the knowledge reasoning process aiming at the probability of generating positive triples in the process of randomly replacing the negative triples by a TransE model.
FIG. 2 is a schematic diagram of entity relationship extraction process based on BILSTM-CRF and BERT fusion, as shown in FIG. 2, including:
and step 21, mapping each word in the sentence into a low-dimensional dense word vector from one-hot vectors by using a pre-trained or randomly initialized embedding matrix. Before inputting the next layer, dropout is set to relieve overfitting;
and step 23, mapping the vector to a dimension k, wherein k is the number of labels of the label set, so as to obtain the sentence characteristics which are automatically extracted and record the sentence characteristics as a matrix P. Each dimension pi of P can be regarded as a score value for classifying the word xi to the jth label, then Softmax is carried out on P to independently carry out k-class classification on each position, and then the P is accessed to a CRF layer to be labeled;
FIG. 3 is a flow chart of knowledge inference of the present invention, as shown in FIG. 3, comprising:
in the process of manually constructing the knowledge graph, the knowledge representation form of the triples cannot effectively measure and utilize the semantic association relationship between the entities, so knowledge reasoning is needed to complement the semantic relationship between the entities. The TransE model has few parameters and low complexity and has the characteristics of simplicity and high efficiency in constructing a large-scale knowledge graph, so that the constructed knowledge graph triple information is supplemented and perfected by selecting the TransE model. The core of the TransE model used by the invention is that in a triplet (h, r, t), the vector representation of the relation r is regarded as a translation from a head entity vector to a tail entity vector. Positive and negative triples may be separated by a triple scoring function. The TransE algorithm score function adopted by the invention is as follows: f (h, r, t) | | | h + r-t | | non-conducting hair1/2Where h, t, r are vector representations of head, relationship and tail entities, respectively, and distances are measured by L1 or L2 norm. The score function may measure the correctness of the triples to the extent that a positive triplet should score close to 0, while a negative triplet score is greater, indicating that the positive triplet before the replacement entity may score closer to 0.
Because the data size of the knowledge graph obtained in the invention is small, the knowledge graph cannot be called a complete knowledge graph, and certain triples are missing. Therefore, the method for randomly replacing the head entity or the tail entity to form the negative triplet in the algorithm has certain defects, because the negative triplet formed in this way may actually be a positive triplet, and the potential correctness probability of such a negative triplet is particularly high, so that the negative triplet is particularly close to the positive triplet and cannot be distinguished. The score formula of the TransE model is as follows: f (h, r, t) | | | h + r-t | | non-conducting hair1/2Where h, r, t represent the vector representation of head, relationship and tail entities, respectively, 1 and 2 represent the L1 norm and L2 norm, respectively, the positive triplet score is close to 0, the larger the negative triplet score the better. The text modifies the potential correct probability of the negative triple in the TransE algorithm, and the potential correct probability of the negative triple in the new algorithm is defined by (CH) is calculated as follows:
wherein P (P)i) Representing the probability of occurrence of a relationship path in the knowledge graph, and an operator representing the relationship rzAnd relation path piCo-linear. Alpha is a hyper-parameter. By modifying the linear function, negative triples with potentially large correct probability values can be reduced to a certain extent, clearly distinguished from positive triples.
After the scores of all the positive and negative triples and the potential correct probabilities of the negative triples are calculated, the vector representations of all the entities and the relations need to be updated by minimizing the target function. The invention adds the potential correct probability of the negative sample into the objective function of the TransE algorithm:
wherein σ is the potential positive probability of the negative triplet, β is the hyper-parameter of the model, the smaller the difference between the positive triplet and the negative triplet is, the larger the potential positive probability of the negative triplet is, and the smaller the objective function score of the negative triplet is, so the score of the negative triplet on the TransE model is used as the standard for measuring the potential positive probability thereof.
Therefore, the invention uses a TransE model to randomly replace a head entity and a tail entity or a relation entity in the knowledge graph to generate a negative triple, then calculates the potential correct probability of the negative triple, removes the triple with higher potential correct probability in the negative triple, and finally uses the corrected negative triple data to perform knowledge inference training. .
FIG. 4 is an example of construction of the knowledge graph based on engineering capability improvement according to the present invention, as shown in FIG. 4, including:
the engineering ability of students is improved from engineering education, talent culture is taken as the research center of gravity, teaching reform is taken as the culture way, and an exploration type teaching method is adopted, wherein the method comprises the steps of establishing an experimental situation and exciting the autonomous exploration desire of the students; an open classroom is created to stimulate students to independently explore the potential; timely dialing is carried out to guide the students to explore the direction; promoting students to realize cooperative exploration and training the active learning ability of the students; promote students to complete innovative homework and stimulate the students to realize active learning. By inquiring knowledge map information, an effective culture way can be found, and the engineering capacity of students is improved. The knowledge map can enable people to intuitively obtain required information, and provides powerful support for cultivating students to improve self engineering capacity.
The method can find a way for improving the engineering capacity of students in all aspects through the knowledge map, and provides an effective way for meeting global challenges of China, strengthening the core competitiveness of China, and vigorously culturing the requirements of excellent scientific and technological engineering talents.
It is to be understood that while the specification has been described in terms of embodiments, it is not intended that each embodiment be construed as a separate embodiment, but rather that the descriptions be provided in a manner that is solely for purposes of clarity, and that all embodiments shown herein are to be considered in all respects as illustrative and not restrictive.
The above-listed series of detailed descriptions are merely specific illustrations of possible embodiments of the present invention, and they are not intended to limit the scope of the present invention, and all equivalent embodiments or modifications that do not depart from the technical spirit of the present invention should be included within the scope of the present invention.
Claims (4)
1. The knowledge graph construction method based on engineering capacity improvement is characterized by comprising the following steps: the method comprises the following steps:
step 1, in order to obtain data related to the improvement of engineering capacity in the engineering education field, crawling knowledge network engineering education field thesis abstract content and Baidu encyclopedia engineering education field content by using a web crawler tool;
step 2, sorting and combining the known network data to obtain entity relation data related to engineering capacity improvement; combining a BILSTM-CRF model and a BERT model, performing model training by using entity relationship data related to engineering capacity improvement in a knowledge network, performing named entity recognition on data in the field of encyclopedic engineering education by using the BILSTM-CRF model, and entering the BERT model to perform entity relationship extraction;
step 3, obtaining a series of triple data for realizing the improvement of the engineering ability of students, and primarily finishing the construction work of the knowledge graph;
step 4, a plurality of triples of entities or relations are lost in the construction process of the knowledge graph, and the entity relations of the constructed knowledge graph are perfected by adopting a negative sample potential correct probability knowledge inference algorithm based on a pre-training model;
and 5, storing the knowledge graph data by using a graph database Neo4j, and performing visual display.
2. The engineering capability elevation-based knowledge graph building system of claim 1, wherein: in order to obtain data related to the engineering capacity improvement in the engineering education field, web crawler tools are used for crawling the content of the thesis abstract in the engineering education field and the content of the encyclopedia engineering education field in the known network, the known network data are sorted and combined to obtain entity relation data related to the engineering capacity improvement, and the entity relation data are used as a training set for fusing an entity naming recognition model and an entity relation extraction model.
3. The knowledge graph construction system based on engineering capability improvement according to claim 1, wherein an entity relationship extraction method based on fusion of BILSTM-CRF and BERT is used; the method captures more information before and after the sentence, and ensures the effectiveness of the final prediction result; comprises the steps of (a) preparing a mixture of a plurality of raw materials,
step 21, mapping each word in the sentence into a low-dimensional dense word vector from one-hot vectors by using a pre-trained or randomly initialized embedded matrix; before inputting the next layer, dropout is set to relieve overfitting;
step 22, automatically extracting sentence characteristics, using a word embedding sequence of each word of a sentence as the input of each time step of the bidirectional LSTM, and splicing the state sequence output by the forward LSTM and the state sequence output by the reverse LSTM at each position according to the position;
step 23, mapping the vector to k dimensions, wherein k is the number of labels of the label set, so as to obtain the sentence characteristics which are automatically extracted and recording as a matrix P; taking each dimension pi of P as a score value for classifying the word xi into the jth label, then performing Softmax on the P to independently perform k-class classification on each position, and then accessing a CRF layer for labeling;
step 24, sentence-level sequence labeling is carried out, and the parameter of the CRF layer is a matrix A, A of (k +2) × (k +2)ijThe expression is the transition score from the ith label to the jth label, and further utilizes the label which is labeled before when labeling a position, and 2 is added because a starting state is added to the head of the sentence and an ending state is added to the tail of the sentence; the score of the whole sequence is equal to the sum of the scores of all the positions, the score of each position is obtained by two parts, one part is determined by pi output by the LSTM, the other part is determined by a transfer matrix A of the CRF, and the sentence is subjected to entity marking according to the score result;
step 25, using a BILSTM-CRF labeling entity as the input of a BERT model, converting the entity vocabulary into vocabulary id through BERT processing, and then mapping the vocabulary id to a vocabulary embedding vector;
step 26, obtaining a coding vector of a sentence through encoding of an encoder of a plurality of layers of transformers, taking a first result of a coding sequence as a semantic vector extracted by a relation, obtaining a semantic expression vector of the sentence at BERT, and then passing through a full connection layer;
and 27, calculating the relation probability by using a softmax function to obtain a final extraction relation result.
4. The knowledge graph construction method based on engineering capacity improvement according to claim 1, characterized in that a knowledge inference method is used, a TransE model is selected to supplement and perfect the constructed knowledge graph triple information; the method comprises the following steps of using a negative sample potential correct probability knowledge inference algorithm based on a pre-training model, wherein the specific operation process comprises the following steps: randomly replacing a head entity in the knowledge graph by using a TransE model, generating a negative triple by using a tail entity or a relationship entity, calculating the potential correct probability of the negative triple, removing the triple with higher potential correct probability in the negative triple, and finally performing knowledge inference training by using the corrected negative triple data;
the score formula of the TransE model is as follows: f (h, r, t) | | | h + r-t | | non-conducting hair1/2Wherein h, r, t represent the vector representation of head entity, relationship and tail entity, respectively, and 1 and 2 represent the L1 norm and L2 norm, respectively; the negative triplet latent correct probability definition (CH) calculation is as follows:
wherein P (P)i) Representing the probability of occurrence of a relationship path in the knowledge graph, and an operator representing the relationship rzAnd relation path piCollinear; α is a hyperparameter; adding a scoring formula and the potential correct probability of the negative sample into an objective function of a TransE algorithm:
where σ is the potential correct probability of the negative triplet and β is the hyper-parameter of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011226817.2A CN112668719A (en) | 2020-11-06 | 2020-11-06 | Knowledge graph construction method based on engineering capacity improvement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011226817.2A CN112668719A (en) | 2020-11-06 | 2020-11-06 | Knowledge graph construction method based on engineering capacity improvement |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112668719A true CN112668719A (en) | 2021-04-16 |
Family
ID=75402995
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011226817.2A Pending CN112668719A (en) | 2020-11-06 | 2020-11-06 | Knowledge graph construction method based on engineering capacity improvement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112668719A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113312496A (en) * | 2021-05-31 | 2021-08-27 | 四川大学 | Method for selecting parameters of micro-lubricating device by fusing knowledge graph |
| CN113672693A (en) * | 2021-08-23 | 2021-11-19 | 东北林业大学 | Label recommendation method of online question and answer platform based on knowledge graph and label association |
| CN114490884A (en) * | 2021-12-21 | 2022-05-13 | 北京三快在线科技有限公司 | Method and device for determining entity association relationship, electronic equipment and storage medium |
| CN115422369A (en) * | 2022-08-30 | 2022-12-02 | 中国人民解放军国防科技大学 | Knowledge graph completion method and device based on improved TextRank |
-
2020
- 2020-11-06 CN CN202011226817.2A patent/CN112668719A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113312496A (en) * | 2021-05-31 | 2021-08-27 | 四川大学 | Method for selecting parameters of micro-lubricating device by fusing knowledge graph |
| CN113672693A (en) * | 2021-08-23 | 2021-11-19 | 东北林业大学 | Label recommendation method of online question and answer platform based on knowledge graph and label association |
| CN114490884A (en) * | 2021-12-21 | 2022-05-13 | 北京三快在线科技有限公司 | Method and device for determining entity association relationship, electronic equipment and storage medium |
| CN115422369A (en) * | 2022-08-30 | 2022-12-02 | 中国人民解放军国防科技大学 | Knowledge graph completion method and device based on improved TextRank |
| CN115422369B (en) * | 2022-08-30 | 2023-11-03 | 中国人民解放军国防科技大学 | Knowledge graph completion method and device based on improved TextRank |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106776581B (en) | Subjective text emotion analysis method based on deep learning | |
| CN112668719A (en) | Knowledge graph construction method based on engineering capacity improvement | |
| CN104318340B (en) | Information visualization methods and intelligent visible analysis system based on text resume information | |
| CN108009285B (en) | Forest Ecology man-machine interaction method based on natural language processing | |
| CN111444351A (en) | Method and device for constructing knowledge graph in industrial process field | |
| CN111538835B (en) | Social media emotion classification method and device based on knowledge graph | |
| CN106980608A (en) | A kind of Chinese electronic health record participle and name entity recognition method and system | |
| CN110990590A (en) | Dynamic financial knowledge map construction method based on reinforcement learning and transfer learning | |
| CN105512209A (en) | Biomedicine event trigger word identification method based on characteristic automatic learning | |
| CN107798624A (en) | A kind of technical label in software Ask-Answer Community recommends method | |
| CN116070602B (en) | PDF document intelligent labeling and extracting method | |
| CN115526236A (en) | Text network graph classification method based on multi-modal comparative learning | |
| CN116383352A (en) | Knowledge graph-based method for constructing field intelligent question-answering system by using zero samples | |
| CN112559734A (en) | Presentation generation method and device, electronic equipment and computer readable storage medium | |
| CN116186237A (en) | Entity relationship joint extraction method based on event cause and effect inference | |
| CN113901224A (en) | Knowledge distillation-based secret-related text recognition model training method, system and device | |
| CN116306653A (en) | Regularized domain knowledge-aided named entity recognition method | |
| CN114943216B (en) | Case microblog attribute level view mining method based on graph attention network | |
| CN114579706B (en) | Automatic subjective question review method based on BERT neural network and multi-task learning | |
| CN113590745B (en) | Interpretable text inference method | |
| CN113378571A (en) | Entity data relation extraction method of text data | |
| Wang et al. | Construction of a Knowledge Map Based on Text CNN Algorithm for Maritime English Subjects | |
| Zuo et al. | College English teaching evaluation model using natural language processing technology and neural networks | |
| Li | An English Writing Grammar Error Correction Technology Based on Similarity Algorithm | |
| CN114996407B (en) | Remote supervision relation extraction method and system based on packet reconstruction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |