CN111598252B

CN111598252B - University computer basic knowledge problem solving method based on deep learning

Info

Publication number: CN111598252B
Application number: CN202010365827.8A
Authority: CN
Inventors: 朱磊; 吕泓瑾; 黑新宏; 冯林林; 张晋源; 刘旭华; 刘尧林; 林泓; 刘瑞
Original assignee: Xian University of Technology
Current assignee: Xian University of Technology
Priority date: 2020-04-30
Filing date: 2020-04-30
Publication date: 2023-08-18
Anticipated expiration: 2040-04-30
Also published as: CN111598252A

Abstract

The invention discloses a method for solving problems by using basic knowledge of a university computer based on deep learning, which uses the existing basic knowledge map of the university computer and an NTN model to generate an inference data set containing all basic knowledge points, then uses PCANET and GRU networks to generate incomplete knowledge maps corresponding to the problems, then uses a PSQUERY algorithm to match approximate subgraphs closest to the incomplete knowledge maps in the knowledge maps, and uses a TransE method to infer the missing parts in the incomplete knowledge maps, and finally the inferred parts are correct results. Reduces the manpower consumption and improves the problem solving efficiency. The invention is easy to realize and can generate answers in batches. The invention improves the efficiency of students to self learn the classes and reduces the burden of teachers in teaching the classes.

Description

University computer basic knowledge problem solving method based on deep learning

Technical Field

The invention belongs to an important direction in the field of artificial intelligence, and particularly relates to a university computer knowledge solving method based on deep learning.

Background

Along with the development of knowledge graph technology, the number of the owned knowledge graphs is increased, and the scale of the knowledge graphs is also increased. The knowledge graph stores massive knowledge in the form of a triplet structure. The extraction of entity relationship is a classical human task in the knowledge graph, and is continuously developed in the past decades, and some staged results are obtained. With the arrival of the deep learning age, various neural network models also provide new ideas for the research of knowledge patterns. Compared with the traditional method, the deep learning model has more efficient learning capability, can face more complex text contexts, and can process larger-scale training data. Deep learning therefore also provides a more novel solution for knowledge-graph research in some aspects. With the rapid development of computer technology and the advent of the IT age, computers have been incorporated into every corner of people's lives. Many students can learn the course by themselves, and the basic knowledge points are not firm, so that the condition of low efficiency can occur. The time of answering questions outside class often cannot meet the needs of a student.

In view of the above, it is very important to know how to learn the course of "university computer foundation". In learning of some students, the situation that the questions are left without standard answers often occurs. Therefore, it is very important to invent a system for obtaining answers according to questions. The systems of this aspect that are currently in use mostly give answers by hand. In the absence of more specialized knowledge, the method is low in speed and the accuracy is not guaranteed. The present invention is a solution to this situation.

Disclosure of Invention

The invention aims to provide a method for solving questions based on deep learning university computer basic knowledge, which solves the problem of low efficiency of manually giving answers in the prior art.

The method is based on a deep learning algorithm and model, an existing university computer basic knowledge graph and an NTN model are used for generating an inference data set containing all basic knowledge points, then PCANET and GRU networks are used for generating incomplete knowledge graphs corresponding to problems, then PSQUERY algorithm is used for matching approximate subgraphs closest to the incomplete knowledge graph in the knowledge graphs, a TransE method is used for reasoning and obtaining missing parts in the incomplete knowledge graphs, and finally the parts obtained through reasoning are correct results.

A method for solving problems by using computer basic knowledge patterns and NTN models is used for generating an inference data set containing all basic knowledge points, PCANET and GRU networks are used for generating incomplete knowledge patterns corresponding to the problems, PSQUERY algorithm is used for matching approximate subgraphs closest to the incomplete knowledge patterns in the knowledge patterns, and a transition method is used for inferring missing parts in the incomplete knowledge patterns, so that correct results of the problems are obtained.

The method specifically comprises the following steps:

step 1, data preprocessing is carried out on a basic knowledge graph of a computer, so that the basic knowledge graph is changed into a triplet form from a visual form such as a neo4j database; meanwhile, screening out knowledge points outside the examination range;

step 2, training the triples in the step 1 by using an NTN network model to generate an inference training set of knowledge points;

step 3, importing a PCANET network model, and training a PCANET network capable of forming a knowledge graph by using the reasoning training set generated in the step 2;

and 4, training by using the data set processed in the step 1 and using a GRU network model to generate a data set capable of identifying knowledge point keywords.

And 5, performing data preprocessing on the computer basic knowledge question base to obtain a data set which can be trained by the PCANET network model.

And 6, training the data set which can identify the knowledge point keywords and is formed in the step 4 and the computer basic knowledge graph data set which is formed in the step 5 by using the PCANET network model generated in the step 3, and generating a incomplete knowledge graph for describing the problem.

And 7, matching approximate subgraphs of incomplete knowledge maps of the description problem in the step 6 in the basic knowledge map of the university computer by using a PSQUERY method.

And 8, obtaining a missing part in the knowledge graph for describing the problem by using a TransE method.

And 9, determining the incomplete part in the knowledge graph as a correct answer, wherein the incomplete part comprises the text concepts and the topological relations.

In step 2, the advantages of using NTN and triplets are: the NTN is mainly trained by representing entities in the database as a vector, and a triplet may also be considered as a vector, i.e. each datum in the triplet is considered as a value in the vector. Therefore, the two can be combined, thereby achieving the effect of training the triples by using NTN; specifically, NTN is used for training the triples; NTN represents each object or individual in the dataset as a vector, which is directly manipulated. The main steps of NTN are: and writing a custom layer, initializing tensor shape, activating functions and tensor parameters, defining and comparing a maximum edge loss function, and finally summarizing data training to obtain a result. These vector vectors can capture the fact about the entity and whether it is part of a relationship, each defined by parameters of a new neural tensor network; as NTN may explicitly refer to two entity vectors. Training triples using NTN would therefore be better.

In step 4, training and generating a data set of knowledge point keywords by using the GRU network model; the GRU can transmit the current data to be used at the next moment; the GRU is composed of innumerable structures called memory blocks, each of which includes an input gate, an output gate, a forget gate and a memory unit; the input gate judges the data passing through the selection by activating the output data of the data input to the input gate by the function; a commonly used activation function is a sigmoid function sigmiod (x) =1/(1+e ∈ (-x)), and a number having a value (- ≡infinity) can be mapped to (0, 1); if the number of mappings is greater than a threshold set in advance, then this data can be output, otherwise it is not;

the activation function has a sigmoid activation function, a tanh activation function, and a Relu activation function.

In the step 7, the PSQUERY algorithm is used for matching the optimal approximate subgraph of the generated result, firstly, the PSQUERT algorithm extracts the characteristics of the query graph, and then codes are respectively carried out according to the extracted characteristics to form node and graph codes; constructing an index tree based on graph coding for filtering; and finally, generating a candidate graph set, and then carrying out sub-graph isomorphism verification to obtain a final result set.

In step 9, a incomplete part of the knowledge graph is obtained by using the TransE; the TransE treats the relationships in each triplet instance as a transformation from a head entity to a tail entity based on a distributed vector representation of the entities and relationships.

The beneficial effects of the invention are as follows:

aiming at the problem that part of question banks in basic knowledge of university computers do not have standard answers, the invention adopts an NTN network model to train a knowledge graph, uses GRU network model training to generate a data set of knowledge point keywords, uses a PSQUERY algorithm to match and generate an optimal approximate subgraph of a result, and uses a TransE to obtain the incomplete part of the knowledge graph. Compared with the traditional algorithm, the method has the advantages that the running efficiency of the algorithm and the accuracy of answers are improved.

Drawings

FIG. 1 is a general flow chart of a university computer basic problem solution method based on deep learning, which is a university computer knowledge solution method based on deep learning;

FIG. 2 is a process of matching approximate subgraphs based on PSQUERY algorithm in the method for solving problems of university computer knowledge based on deep learning;

FIG. 3 is a process of generating an inference data set of knowledge points based on NTN in the method for solving the problem of the university computer knowledge based on deep learning;

fig. 4 is a process of generating a incomplete knowledge graph corresponding to a problem based on pcante according to the college computer knowledge solving method based on deep learning of the present invention.

Fig. 5 is a process of generating a keyword data set including knowledge points based on a GRU according to the method for solving problems of college computer knowledge based on deep learning of the present invention.

Fig. 6 is a process of obtaining a missing part in a knowledge graph based on a transition method in the university computer knowledge solving method based on deep learning.

Detailed Description

Specific embodiments of the present invention will be described below with reference to the accompanying drawings.

The invention relates to a method for solving problems of basic knowledge of a university computer based on deep learning, which is shown in figure 1 and specifically comprises the following steps:

and 1.1, carrying out data preprocessing on a basic knowledge graph of a university computer to convert the basic knowledge graph from a neo4j database and other forms into a data form of a triplet.

Step 1.2, the knowledge points which are not required by the examination in the data set and only need to be known by students are removed.

And step 2, training the triples by using the NTN network model to generate an inference data set of knowledge points. As shown in fig. 3, the tensor shape, activation function, tensor parameters, custom layers in the NTN model are fixed, or may be slightly modified in practice. The maximum edge loss function is also determined. The pseudocode is as follows:

and 3, as shown in fig. 4, importing a PCANET network model, and training a PCANET network model capable of generating a knowledge graph by using the reasoning training set generated in the step 2.

And 4, training by using the data set processed in the step 1 and using a GRU network model to generate a data set containing knowledge point keywords. As shown in fig. 5.

And 5.1, reading corpus data of a basic knowledge question bank of a university computer, and dividing all data by taking a complete question as a basis unit.

And 5.2, marking all the titles for searching.

And 6.1, inputting the data set generated in the step 4 into the PCANET network model generated in the step 3.

And 6.2, generating a incomplete knowledge graph for describing the problem by using the PCANET network model. As shown in fig. 4.

And 7.1, extracting the marking information of the nodes to generate node marking codes, and combining all the node marking codes to generate the node marking codes of the graph.

And 7.2, extracting edge characteristics as adjacent edge information. The encoding of the adjacent edge can be obtained by traversing the hash mapping of the adjacent edge.

And 7.3, extracting the shortest weight path as a third characteristic. May be implemented using the dijkstra algorithm. Where list represents the weight matrix of the current path. An initial state is a weight between two connected points, which is considered infinite if the two points are not connected. Paths represents the shortest path between two points, with initial state being the same as list. The pseudo code is as follows:

and 7.4, taking the minimum value from the weight value of each node to obtain the weight value path code of the graph D.

And 7.5, generating an N-layer generation diagram for each node, calculating the partial topology information of each node by using the diagram to represent, and finally combining to obtain the topology information of the undirected weighted diagram. I.e. Laplacian map. The pseudocode is as follows:

and 7.6, solving corresponding characteristic values according to a Jacobian algorithm, and taking the largest two characteristic values as a node map. And combining the maps of all the nodes to generate a map of the graph.

And 7.7, constructing an index tree by referring to the GCoding method.

And 7.8, filtering the graphs and the nodes, and clipping most of the graphs by using PSQUERY to generate candidate sets, as shown in FIG. 2.

And 7.9, performing subgraph isomorphism judgment by adopting a VF2 algorithm. The pseudo code is as follows:

step 8, as shown in fig. 6, a method based on the transition is used to obtain a missing part of the knowledge graph describing the problem. In this algorithm, the parameters such as super-parameters, learning rate, etc. are generally determined and can be modified slowly by practice. The pseudocode of the TransE algorithm is as follows:

the TransE treats the relationships in each triplet instance as a transformation from a head entity to a tail entity based on a distributed vector representation of the entities and relationships. For example, the phrase "the creator of the Linux operating system is Lin Nasi ·benna kth·towatts" may be expressed as "Linux operating system+creator= Lin Nasi ·benna kth·towatts". Then in the incomplete knowledge graph, by analogy, the entity "Linux operating system" and the relation "creator" are known, and the entity "Lin Nasi-benna kth-tova" to be found can be deduced.

And 9, converting the incomplete part in the knowledge graph into a correct answer described by using the characters in a manual mode.

The invention reasonably utilizes the characteristics of the basic knowledge graph of the computer and the question bank of the university computer, reduces the manpower consumption and improves the question solving efficiency. Firstly, converting the question into a incomplete knowledge graph, then matching the incomplete knowledge graph with knowledge points, finding out an optimal matching subgraph, and finally finding out the knowledge graph of the incomplete part of the question, namely, obtaining an answer of the question. The invention is easy to realize and can generate answers in batches. The invention improves the efficiency of students to self learn the classes and reduces the burden of teachers in teaching the classes.

Claims

1. A method for solving problems by using computer knowledge base map and NTN model based on deep learning is characterized in that an inference data set containing all basic knowledge points is generated by using computer basic knowledge map and NTN model, then incomplete knowledge map corresponding to problems is generated by using PCANET and GRU network, then approximate subgraphs closest to the incomplete knowledge map in the knowledge map are matched by using PSQUERY algorithm, and missing parts in the incomplete knowledge map are obtained by inference by using a TransE method, so that correct results of problems are obtained;

the method specifically comprises the following steps:

step 1, data preprocessing is carried out on a basic knowledge graph of a computer, so that the basic knowledge graph is changed into a triplet form from a neo4j database visualization form; meanwhile, screening out knowledge points outside the examination range;

step 4, training by using the data set processed in the step 1 and using a GRU network model, and generating a data set capable of identifying knowledge point keywords;

step 5, data preprocessing is carried out on the basic knowledge question base of the computer to obtain a data set which can be trained by the PCANET network model;

step 6, training the data set which can identify the key words of the knowledge points and is formed in the step 4 and the computer basic knowledge graph data set which is formed in the step 5 by using the PCANET network model generated in the step 3, and generating a incomplete knowledge graph for describing the problem;

step 7, matching the optimal approximate subgraphs of the incomplete knowledge graphs of the description problem in the step 6 in the basic knowledge graph of the university computer by using a PSQUERY method;

step 8, obtaining a missing part in a knowledge graph for describing the problem by using a TransE method;

2. The method for solving the problems of university computer knowledge based on deep learning of claim 1, wherein in the step 2, the advantages of using NTN and triplet are: the main training mode of the NTN is realized by representing the entity in the database as a vector, and a triplet can also be regarded as a vector, namely, each data in the triplet is regarded as a numerical value in the vector; therefore, the two can be combined, thereby achieving the effect of training the triples by using NTN; specifically, NTN is used for training the triples; NTN represents each object or individual in the dataset as a vector, directly operating on the vector; the main steps of NTN are: writing a custom layer, initializing tensor shape, activating function and tensor parameters, defining and comparing maximum edge loss function, and finally summarizing data training to obtain a result; these vector vectors can capture the fact about the entity and whether it is part of a relationship, each defined by parameters of a new neural tensor network.

3. The method for solving the problems of the university computer knowledge based on deep learning of claim 1, wherein in the step 4, a data set of knowledge point keywords is generated by training using a GRU network model; the GRU can transmit the current data to be used at the next moment; the GRU is composed of innumerable structures called memory blocks, each of which includes an input gate, an output gate, a forget gate and a memory unit; the input gate judges the data passing through the selection by activating the output data of the data input to the input gate by the function; a commonly used activation function is a sigmoid function sigmiod (x) =1/(1+e ∈ (-x)), and a number having a value (- ≡infinity) can be mapped to (0, 1); if the number of mappings is greater than the threshold set in advance, this data can be output, otherwise it is not.

4. A university computer knowledge solution method based on deep learning as claimed in claim 3, characterized in that the activation functions are sigmoid activation function, tanh activation function, relu activation function.

5. The method for solving the problems of the university computer knowledge based on the deep learning according to claim 1, wherein in the step 7, the PSQUERY algorithm is used to match the optimal approximate subgraph of the generated result, firstly, the PSQUERT algorithm extracts the characteristics of the query graph, and then codes according to the extracted characteristics respectively to form the node and the graph code; constructing an index tree based on graph coding for filtering; and finally, generating a candidate graph set, and then carrying out sub-graph isomorphism verification to obtain a final result set.

6. The method for solving the problems of the university computer knowledge based on deep learning of claim 1, wherein in the step 9, the incomplete part of the knowledge graph is obtained by using the transition; the TransE treats the relationships in each triplet instance as a transformation from a head entity to a tail entity based on a distributed vector representation of the entities and relationships.