CN107368521B - Knowledge recommendation method and system based on big data and deep learning - Google Patents

Abstract

The invention discloses a knowledge recommendation method and a system based on big data and deep learning, wherein the method comprises the following steps: performing scene analysis according to the scene data information; processing the data information of the operation and maintenance personnel by adopting a deep learning method to generate an operation and maintenance personnel portrait; selecting knowledge according to a scene analysis result and the operation and maintenance personnel portrait to obtain a knowledge recommendation set; performing knowledge recommendation according to the knowledge recommendation set; the system comprises a scene analysis module, an operation and maintenance personnel portrait generation module, a knowledge selection module and a knowledge recommendation module. The invention integrates the scene analysis result and the operation and maintenance personnel image to select knowledge and recommend knowledge, thus improving the accuracy of knowledge recommendation from two dimensions of real-time scene use and user use; a deep learning method is adopted to learn data information such as operation behaviors of operation and maintenance personnel, so that the operation and maintenance personnel portray more and more accurately. The invention can be widely applied to the field of computer application.

Description

Knowledge recommendation method and system based on big data and deep learning

Technical Field

The invention relates to the field of computer application, in particular to a knowledge recommendation method and a knowledge recommendation system based on big data and deep learning.

Background

Today, knowledge is promoted to strategic resources by enterprises, and the enterprises are imperative to adopt knowledge management. With the increasingly large service systems, the increasingly complex service logics, the more frequent system changes and the higher working requirements, the difficulty of the operation management work of the service support network is also increased. By constructing a unified knowledge base with rich contents and human participation, the aims of establishing a channel for uploading and issuing knowledge for enterprises, creating a learning type business support team and assisting companies in long-term sustainable health development can be fulfilled.

Knowledge Management (KM, Knowledge Management) is an emerging Management thinking and method in the new economic era of networks, and Management scholars predicted by durac as early as one, nine, six, and five years: "knowledge will replace land, labor, capital and machinery equipment, becoming the most important production factor. The concept of knowledge management combined with the entrance website, database and application computer software system constructed by the internet is a new century benefit device for accumulating knowledge wealth and creating more competitiveness due to the vigorous development of informatization (informatization) in the 90 s of the 20 th century.

So-called knowledge management, which is defined as: a knowledge system with both humanity and technology is established in an organization, so that the final aim of continuous innovation of knowledge is achieved through the processes of obtaining, creating, sharing, integrating, recording, accessing, updating and the like of information and knowledge in the organization, the knowledge is fed back to the knowledge system, the knowledge of individuals and the organization can be accumulated forever, the wisdom capital of the organization is formed by thinking from the perspective of the system, and the enterprise can make correct decisions to adapt to market transition.

In knowledge management, how knowledge is applied is also an important topic. At present, the application of knowledge is mainly based on manual retrieval and directory tree display, and the application modes all require more manpower and time investment and have low accuracy. The advent of knowledge recommendation technology changes the acquisition mode of knowledge from "search" to "recommendation", solving the problem.

At present, part of knowledge application can analyze the flow or the place and recommend knowledge according to the analysis result. Compared with the manual retrieval and directory tree display mode, the mode not only greatly reduces the manpower and time investment, but also can improve the accuracy of knowledge recommendation. However, this method fails to consider the key information of the users (such as the knowledge operation and maintenance personnel) of the knowledge application, and the accuracy is not ideal. Meanwhile, the method cannot continuously learn and self-correct to adapt to the changing conditions, and the accuracy rate of the method is lower and lower as time goes on.

Disclosure of Invention

To solve the above technical problems, the present invention aims to: the knowledge recommendation method based on big data and deep learning is high in accuracy.

Another object of the present invention is to: the knowledge recommendation system based on big data and deep learning is high in accuracy.

The technical scheme adopted by the invention is as follows:

a knowledge recommendation method based on big data and deep learning comprises the following steps:

performing scene analysis according to the scene data information;

processing the data information of the operation and maintenance personnel by adopting a deep learning method to generate an operation and maintenance personnel portrait;

selecting knowledge according to a scene analysis result and the operation and maintenance personnel portrait to obtain a knowledge recommendation set;

and carrying out knowledge recommendation according to the knowledge recommendation set.

Further, the step of performing scene analysis according to the scene data information includes:

acquiring scene data information in real time, and acquiring scene data operated by current operation and maintenance personnel;

carrying out data cleaning on the acquired scene data;

and carrying out real-time scene analysis and labeling operation on the scene data after data cleaning to obtain a scene mark.

Further, the step of processing the data information of the operation and maintenance personnel by using the deep learning method to generate the operation and maintenance personnel portrait comprises the following steps:

acquiring operation and maintenance personnel data information, wherein the operation and maintenance personnel data information comprises data acquired from information registered by an operation and maintenance personnel account and operation behavior data of the operation and maintenance personnel acquired in real time;

carrying out data cleaning on the collected operation and maintenance personnel data information;

carrying out model training on the operation and maintenance personnel by adopting a method of fusing weak model training and Boosting on the data information of the operation and maintenance personnel after data cleaning to obtain an operation and maintenance personnel portrait;

and calculating the similarity between the operation and maintenance personnel and the knowledge label according to the operation and maintenance personnel portrait to obtain the operation and maintenance personnel-knowledge similarity.

Further, the operation and maintenance personnel model training is carried out on the operation and maintenance personnel data information after data cleaning by adopting a method of fusing weak model training and Boosting to obtain an operation and maintenance personnel portrait, and the method comprises the following steps:

performing text modeling, so as to divide the operation and maintenance personnel data information after data cleaning into static information data and dynamic information data;

performing weak model training on a given training sample according to the requirement of text modeling to obtain a plurality of weak models;

the accuracy of the weak models is improved by adopting a Boosting method, and a result classifier of the operation and maintenance personnel model is obtained;

performing model verification on a result classifier of the operation and maintenance personnel model by adopting a given test sample;

and storing the operation and maintenance personnel model after the model verification is passed, and acquiring new operation and maintenance personnel data information after data cleaning in real time to continuously correct the operation and maintenance personnel model and the corresponding operation and maintenance personnel portrait.

Further, the step of calculating the similarity between the operation and maintenance personnel and the knowledge tag according to the operation and maintenance personnel portrait to obtain the operation and maintenance personnel-knowledge similarity includes:

and calculating the name similarity according to the operation and maintenance personnel image, wherein the calculation formula of the name similarity is as follows:

wherein S is_name(U, C) is the name similarity between the concept U in the operation and maintenance personnel figure and the concept C in the knowledge label, U_i(i is more than or equal to 1 and less than or equal to n) is a character string semantic word segmentation result of the name U in the concept U, c_j(j is more than or equal to 1 and less than or equal to m) is the result of semantic segmentation of the character string of the name C in the concept C, n is the total number of the character strings of the name U in the concept U, m is the total number of the character strings of the name C in the concept C, and Sim (U_i,c_j) Is u_iAnd c_jSimilarity between them;

belongs according to the operation and maintenance personnel figureAnd calculating the similarity, wherein the attribute similarity calculation formula is as follows:

wherein S is_attri(U, C) is the attribute similarity between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge tag, U_aAnd C_aRepresenting the set of attributes of U and C, respectively, f being a given non-negative metric function, U_a∩C_aRepresenting a set of two concepts, U and C, having the same attribute, U_a-C_aRepresenting an attribute set, C, that is only present in the representation of the operation and maintenance person but not in the knowledge tag_a-U_aRepresenting the attribute set which is only contained in the knowledge tag but not contained in the operation and maintenance personnel image, wherein lambda and mu are given weight coefficients;

and calculating example similarity according to the operation and maintenance personnel image, wherein the example similarity calculation formula is as follows:

wherein S is_inst(U, C) is the example similarity between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label, P (U, C) represents the probability that an example randomly extracted from the example space belongs to the concepts U and C at the same time,

representing the probability that an instance randomly drawn from the instance space belongs only to concept U and not to concept C,

represents the probability that an instance randomly drawn from the instance space belongs only to concept C and not to concept U;

calculating the relation similarity according to the operation and maintenance personnel portrait to obtain the relation similarity S between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge label_relat(U, C), wherein the relationship comprises a synonymy relationship, an inheritance relationship, and an inclusion relationship, the weight of the synonymy relationship is greater than the weight of the inheritance relationship, and the weight of the synonymy relationship is greater than the weight of the inclusion relationship;

according toName similarity S_name(U, C), Attribute similarity S_attri(U, C), example similarity S_inst(U, C) and relationship similarity S_relat(U, C) calculating the similarity of the operation and maintenance personnel and the knowledge, wherein the calculation formula of the similarity of the operation and maintenance personnel and the knowledge Sim (U, C) is that Sim (U, C) is α S_name(U,C)+βS_attri(U,C)+εS_inst(U,C)+δS_relat(U, C), wherein α, β, epsilon, and delta are given name similarity coefficient, attribute similarity coefficient, instance similarity coefficient, and relationship similarity coefficient, respectively.

Further, the step of selecting knowledge according to the result of the scene analysis and the portrait of the operation and maintenance personnel to obtain a knowledge recommendation set includes:

searching the characteristics of the operation and maintenance personnel according to the operation and maintenance personnel images, and calculating an operation and maintenance similar knowledge set by combining the operation and maintenance personnel images;

acquiring scene characteristics in real time according to the scene analysis result, and calculating a scene similarity knowledge set according to the scene characteristics;

and judging whether the scene similar knowledge set and the operation and maintenance similar knowledge set have intersection, if so, forming a recommendation knowledge set according to the intersection, otherwise, re-acquiring scene characteristics, and re-training the operation and maintenance personnel model to obtain a new operation and maintenance personnel portrait.

Further, the step of selecting knowledge according to the result of the scene analysis and the portrait of the operation and maintenance personnel to obtain a knowledge recommendation set includes:

carrying out user scene retrieval on operation and maintenance personnel, wherein the user scene U is_mContains a scene u, and

wherein, U_TIs a full user scenario;

calculating a user scene U by adopting a set similarity calculation method_mSimilar knowledge scenes are sequenced to obtain a similar scene set C_n：

Wherein, U_m≈C_uUser scene U representing operation and maintenance personnel_mAnd knowledge scene C_uA is an intersection symbol, C_TIs a full knowledge scene set;

retrieving a given scene knowledge relationship model O^RTo obtain a scene C_nKnowledge-scene relationship pair P of_i,j：

Wherein, P_TAs a full-knowledge-scene-relationship pair, C_iFor knowledge scenarios, R₁As scene C_iThe properties of (a) to (b) are,

represents P_i,jBy R₁And C_iConnecting;

retrieving a given set of knowledge domains O^DTo obtain a compound containing_i,jAssociated domain knowledge set D_k：

Wherein D is_TIn the field of full-scale knowledge, R₂For domain knowledge D_kThe properties of (a) to (b) are,

represents P_i,jBy R₂And D_kConnecting;

from a full-scale domain knowledge recommendation set

To obtain knowledge recommendation set K_S：

The other technical scheme adopted by the invention is as follows:

a knowledge recommendation system based on big data and deep learning, comprising:

the scene analysis module is used for carrying out scene analysis according to the scene data information;

the operation and maintenance personnel portrait generation module is used for processing the data information of the operation and maintenance personnel by adopting a deep learning method to generate an operation and maintenance personnel portrait;

the knowledge selection module is used for selecting knowledge according to the result of the scene analysis and the operation and maintenance personnel portrait to obtain a knowledge recommendation set;

and the knowledge recommendation module is used for recommending knowledge according to the knowledge recommendation set.

Further, the operation and maintenance personnel portrait generation module comprises:

the information acquisition unit is used for acquiring operation and maintenance personnel data information, and the operation and maintenance personnel data information comprises data acquired from information registered by an operation and maintenance personnel account and operation behavior data of the operation and maintenance personnel acquired in real time;

the data cleaning unit is used for cleaning the data of the collected operation and maintenance personnel data information;

the model training unit is used for carrying out model training on the operation and maintenance personnel by adopting a method of fusing weak model training and Boosting on the operation and maintenance personnel data information after data cleaning to obtain an operation and maintenance personnel portrait;

and the similarity calculation unit is used for calculating the similarity between the operation and maintenance personnel and the knowledge label according to the operation and maintenance personnel portrait to obtain the operation and maintenance personnel-knowledge similarity.

Further, the similarity calculation unit includes:

the name similarity calculation operator unit is used for calculating name similarity according to the operation and maintenance personnel image, and the calculation formula of the name similarity is as follows:

wherein S is_name(U, C) is the name similarity between the concept U in the operation and maintenance personnel figure and the concept C in the knowledge label, U_i(i is more than or equal to 1 and less than or equal to n) is a character string semantic word segmentation result of the name U in the concept U, c_j(j is more than or equal to 1 and less than or equal to m) is the result of semantic segmentation of the character string of the name C in the concept C, n is the total number of the character strings of the name U in the concept U, m is the total number of the character strings of the name C in the concept C, and Sim (U_i,c_j) Is u_iAnd c_jSimilarity between them;

the attribute similarity calculation operator unit is used for calculating the attribute similarity according to the operation and maintenance personnel image, and the attribute similarity calculation formula is as follows:

wherein S is_attri(U, C) is the attribute similarity between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge tag, U_aAnd C_aRepresenting the set of attributes of U and C, respectively, f being a given non-negative metric function, U_a∩C_aRepresenting a set of two concepts, U and C, having the same attribute, U_a-C_aRepresenting an attribute set, C, that is only present in the representation of the operation and maintenance person but not in the knowledge tag_a-U_aRepresenting the attribute set which is only contained in the knowledge tag but not contained in the operation and maintenance personnel image, wherein lambda and mu are given weight coefficients;

the example similarity operator unit is used for calculating the example similarity according to the operation and maintenance personnel image, and the example similarity calculation formula is as follows:

wherein S is_inst(U, C) is the example similarity between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label, P (U, C) represents the probability that an example randomly extracted from the example space belongs to the concepts U and C at the same time,

representing the probability that an instance randomly drawn from the instance space belongs only to concept U and not to concept C,

represents the probability that an instance randomly drawn from the instance space belongs only to concept C and not to concept U;

the relational similarity calculation operator unit is used for calculating the relational similarity according to the operation and maintenance personnel image to obtain the relational similarity S between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label_relat(U,C)，The relationships comprise a synonymy relationship, an inheritance relationship and an inclusion relationship, wherein the weight of the synonymy relationship is greater than that of the inheritance relationship, and the weight of the synonymy relationship is greater than that of the inclusion relationship;

an operation and maintenance personnel-knowledge similarity calculation operator unit for calculating the similarity S according to the name_name(U, C), Attribute similarity S_attri(U, C), example similarity S_inst(U, C) and relationship similarity S_relat(U, C) calculating the similarity of the operation and maintenance personnel and the knowledge, wherein the calculation formula of the similarity of the operation and maintenance personnel and the knowledge Sim (U, C) is that Sim (U, C) is α S_name(U,C)+βS_attri(U,C)+εS_inst(U,C)+δS_relat(U, C), wherein α, β, epsilon, and delta are given name similarity coefficient, attribute similarity coefficient, instance similarity coefficient, and relationship similarity coefficient, respectively.

The method of the invention has the beneficial effects that: the method comprises the steps of selecting knowledge according to a scene analysis result and an operation and maintenance person portrait to obtain a knowledge recommendation set and recommending the knowledge according to the knowledge recommendation set, wherein the scene analysis result and the operation and maintenance person portrait are integrated to select and recommend the knowledge, so that not only can knowledge recommendation be performed according to a real-time use scene of the knowledge application, but also user information of the knowledge application, namely the operation and maintenance person portrait, is considered during the knowledge recommendation, and the accuracy of the knowledge recommendation is improved from the two dimensions of the real-time use scene and the use user; the method comprises the steps of processing data information of the operation and maintenance personnel by adopting a deep learning method and generating the portrait of the operation and maintenance personnel, wherein the operation and maintenance personnel are learned by adopting the deep learning method, so that the portrait of the operation and maintenance personnel can be trained, and the portrait model of the operation and maintenance personnel can be continuously corrected by combining with subsequent operation behavior data of the operation and maintenance personnel, so that the portrait of the operation and maintenance personnel is more and more accurate.

The system of the invention has the advantages that: the knowledge recommendation system comprises a knowledge selection module for selecting knowledge according to a scene analysis result and an operation and maintenance person portrait to obtain a knowledge recommendation set and a knowledge recommendation module for recommending knowledge according to the knowledge recommendation set, wherein the scene analysis result and the operation and maintenance person portrait are integrated to select and recommend knowledge, so that not only can knowledge be recommended according to a real-time use scene of knowledge application, but also user information of the knowledge application, namely the operation and maintenance person portrait, is considered during the knowledge recommendation, and the accuracy of the knowledge recommendation is improved from the real-time use scene and the use user; the operation and maintenance person portrait generation module learns data information such as operation behaviors of the operation and maintenance person by adopting a deep learning method, can train the operation and maintenance person portrait, and can continuously correct an operation and maintenance person portrait model by combining with subsequent operation behavior data of the operation and maintenance person, so that the operation and maintenance person portrait is more and more accurate.

Drawings

FIG. 1 is an overall flow chart of a knowledge recommendation method based on big data and deep learning according to the present invention;

FIG. 2 is a flowchart illustrating a knowledge recommendation method according to an embodiment of the present invention;

FIG. 3 is a flowchart of an operation and maintenance personnel image generation process according to an embodiment of the present invention;

FIG. 4 is a flow chart of a knowledge selection process according to an embodiment of the invention.

Detailed Description

Referring to fig. 1, a knowledge recommendation method based on big data and deep learning includes the following steps:

performing scene analysis according to the scene data information;

processing the data information of the operation and maintenance personnel by adopting a deep learning method to generate an operation and maintenance personnel portrait;

selecting knowledge according to a scene analysis result and the operation and maintenance personnel portrait to obtain a knowledge recommendation set;

and carrying out knowledge recommendation according to the knowledge recommendation set.

Further as a preferred embodiment, the step of performing scene analysis according to scene data information includes:

acquiring scene data information in real time, and acquiring scene data operated by current operation and maintenance personnel;

carrying out data cleaning on the acquired scene data;

and carrying out real-time scene analysis and labeling operation on the scene data after data cleaning to obtain a scene mark.

Further preferably, the step of processing the data information of the operation and maintenance personnel by using the deep learning method to generate the operation and maintenance personnel portrait includes:

acquiring operation and maintenance personnel data information, wherein the operation and maintenance personnel data information comprises data acquired from information registered by an operation and maintenance personnel account and operation behavior data of the operation and maintenance personnel acquired in real time;

carrying out data cleaning on the collected operation and maintenance personnel data information;

carrying out model training on the operation and maintenance personnel by adopting a method of fusing weak model training and Boosting on the data information of the operation and maintenance personnel after data cleaning to obtain an operation and maintenance personnel portrait;

and calculating the similarity between the operation and maintenance personnel and the knowledge label according to the operation and maintenance personnel portrait to obtain the operation and maintenance personnel-knowledge similarity.

Further, as a preferred embodiment, the step of performing model training on the operation and maintenance personnel data information after data cleaning by using a method of fusing weak model training and Boosting to obtain an operation and maintenance personnel portrait includes:

performing text modeling, so as to divide the operation and maintenance personnel data information after data cleaning into static information data and dynamic information data;

performing weak model training on a given training sample according to the requirement of text modeling to obtain a plurality of weak models;

the accuracy of the weak models is improved by adopting a Boosting method, and a result classifier of the operation and maintenance personnel model is obtained;

performing model verification on a result classifier of the operation and maintenance personnel model by adopting a given test sample;

and storing the operation and maintenance personnel model after the model verification is passed, and acquiring new operation and maintenance personnel data information after data cleaning in real time to continuously correct the operation and maintenance personnel model and the corresponding operation and maintenance personnel portrait.

Further, as a preferred embodiment, the step of calculating the similarity between the operation and maintenance person and the knowledge tag according to the operation and maintenance person portrait to obtain the operation and maintenance person-knowledge similarity includes:

and calculating the name similarity according to the operation and maintenance personnel image, wherein the calculation formula of the name similarity is as follows:

wherein S is_name(U, C) is the name similarity between the concept U in the operation and maintenance personnel figure and the concept C in the knowledge label, U_i(i is more than or equal to 1 and less than or equal to n) is a character string semantic word segmentation result of the name U in the concept U, c_j(j is more than or equal to 1 and less than or equal to m) is the result of semantic segmentation of the character string of the name C in the concept C, n is the total number of the character strings of the name U in the concept U, m is the total number of the character strings of the name C in the concept C, and Sim (U_i,c_j) Is u_iAnd c_jSimilarity between them;

and performing attribute similarity calculation according to the operation and maintenance personnel image, wherein the attribute similarity calculation formula is as follows:

wherein S is_attri(U, C) is the attribute similarity between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge tag, U_aAnd C_aRepresenting the set of attributes of U and C, respectively, f being a given non-negative metric function, U_a∩C_aRepresenting a set of two concepts, U and C, having the same attribute, U_a-C_aRepresenting an attribute set, C, that is only present in the representation of the operation and maintenance person but not in the knowledge tag_a-U_aRepresenting the attribute set which is only contained in the knowledge tag but not contained in the operation and maintenance personnel image, wherein lambda and mu are given weight coefficients;

and calculating example similarity according to the operation and maintenance personnel image, wherein the example similarity calculation formula is as follows:

wherein S is_inst(U, C) are drawn by operation and maintenance personnelThe similarity of the instances between the concept U in the image and the concept C in the knowledge tag, P (U, C) represents the probability that an instance randomly drawn from the instance space belongs to both concepts U and C,

representing the probability that an instance randomly drawn from the instance space belongs only to concept U and not to concept C,

represents the probability that an instance randomly drawn from the instance space belongs only to concept C and not to concept U;

calculating the relation similarity according to the operation and maintenance personnel portrait to obtain the relation similarity S between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge label_relat(U, C), wherein the relationship comprises a synonymy relationship, an inheritance relationship, and an inclusion relationship, the weight of the synonymy relationship is greater than the weight of the inheritance relationship, and the weight of the synonymy relationship is greater than the weight of the inclusion relationship;

according to the name similarity S_name(U, C), Attribute similarity S_attri(U, C), example similarity S_inst(U, C) and relationship similarity S_relat(U, C) calculating the similarity of the operation and maintenance personnel and the knowledge, wherein the calculation formula of the similarity of the operation and maintenance personnel and the knowledge Sim (U, C) is that Sim (U, C) is α S_name(U,C)+βS_attri(U,C)+εS_inst(U,C)+δS_relat(U, C), wherein α, β, epsilon, and delta are given name similarity coefficient, attribute similarity coefficient, instance similarity coefficient, and relationship similarity coefficient, respectively.

Further, as a preferred embodiment, the step of obtaining a knowledge recommendation set by selecting knowledge according to a result of the scene analysis and the operation and maintenance person portrait includes:

searching the characteristics of the operation and maintenance personnel according to the operation and maintenance personnel images, and calculating an operation and maintenance similar knowledge set by combining the operation and maintenance personnel images;

acquiring scene characteristics in real time according to the scene analysis result, and calculating a scene similarity knowledge set according to the scene characteristics;

and judging whether the scene similar knowledge set and the operation and maintenance similar knowledge set have intersection, if so, forming a recommendation knowledge set according to the intersection, otherwise, re-acquiring scene characteristics, and re-training the operation and maintenance personnel model to obtain a new operation and maintenance personnel portrait.

Further, as a preferred embodiment, the step of obtaining a knowledge recommendation set by selecting knowledge according to a result of the scene analysis and the operation and maintenance person portrait includes:

carrying out user scene retrieval on operation and maintenance personnel, wherein the user scene U is_mContains a scene u, and

wherein, U_TIs a full user scenario;

calculating a user scene U by adopting a set similarity calculation method_mSimilar knowledge scenes are sequenced to obtain a similar scene set C_n：

Wherein, U_m≈C_uUser scene U representing operation and maintenance personnel_mAnd knowledge scene C_uA is an intersection symbol, C_TIs a full knowledge scene set;

retrieving a given scene knowledge relationship model O^RTo obtain a scene C_nKnowledge-scene relationship pair P of_i,j：

Wherein, P_TAs a full-knowledge-scene-relationship pair, C_iFor knowledge scenarios, R₁As scene C_iThe properties of (a) to (b) are,

represents P_i,jBy R₁And C_iConnecting;

retrieving a given set of knowledge domains O^DTo obtain a compound containing_i,jAssociated domain knowledge set D_k：

Wherein D is_TIn the field of full-scale knowledge, R₂For domain knowledge D_kThe properties of (a) to (b) are,

represents P_i,jBy R₂And D_kConnecting;

from a full-scale domain knowledge recommendation set

To obtain knowledge recommendation set K_S：

Wherein, the scene knowledge relation model O^RAnd field of knowledge set O^DCalculated or pre-given before knowledge selection. The knowledge recommendation method and the knowledge recommendation system sequentially search the user scene, match the similarity of the user scene, match the knowledge-scene relationship and match the knowledge field when selecting the knowledge, and finally find out the knowledge recommendation set.

The invention relates to a knowledge recommendation system based on big data and deep learning, which comprises:

the scene analysis module is used for carrying out scene analysis according to the scene data information;

the operation and maintenance personnel portrait generation module is used for processing the data information of the operation and maintenance personnel by adopting a deep learning method to generate an operation and maintenance personnel portrait;

the knowledge selection module is used for selecting knowledge according to the result of the scene analysis and the operation and maintenance personnel portrait to obtain a knowledge recommendation set;

and the knowledge recommendation module is used for recommending knowledge according to the knowledge recommendation set.

Further preferably, the operation and maintenance person representation generating module includes:

the information acquisition unit is used for acquiring operation and maintenance personnel data information, and the operation and maintenance personnel data information comprises data acquired from information registered by an operation and maintenance personnel account and operation behavior data of the operation and maintenance personnel acquired in real time;

the data cleaning unit is used for cleaning the data of the collected operation and maintenance personnel data information;

the model training unit is used for carrying out model training on the operation and maintenance personnel by adopting a method of fusing weak model training and Boosting on the operation and maintenance personnel data information after data cleaning to obtain an operation and maintenance personnel portrait;

and the similarity calculation unit is used for calculating the similarity between the operation and maintenance personnel and the knowledge label according to the operation and maintenance personnel portrait to obtain the operation and maintenance personnel-knowledge similarity.

Further preferably, the similarity calculation unit includes:

the name similarity calculation operator unit is used for calculating name similarity according to the operation and maintenance personnel image, and the calculation formula of the name similarity is as follows:

wherein S is_name(U, C) is the name similarity between the concept U in the operation and maintenance personnel figure and the concept C in the knowledge label, U_i(i is more than or equal to 1 and less than or equal to n) is a character string semantic word segmentation result of the name U in the concept U, c_j(j is more than or equal to 1 and less than or equal to m) is the result of semantic segmentation of the character string of the name C in the concept C, n is the total number of the character strings of the name U in the concept U, m is the total number of the character strings of the name C in the concept C, and Sim (U_i,c_j) Is u_iAnd c_jSimilarity between them;

the attribute similarity calculation operator unit is used for calculating the attribute similarity according to the operation and maintenance personnel image, and the attribute similarity calculation formula is as follows:

wherein S is_attri(U, C) is the attribute similarity between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge tag, U_aAnd C_aRepresenting the set of attributes of U and C, respectively, f being a given non-negative metric function, U_a∩C_aRepresenting a set of two concepts, U and C, having the same attribute, U_a-C_aRepresentation is operation and maintenance onlyAttribute sets, C, present in person representations but absent in knowledge tags_a-U_aRepresenting the attribute set which is only contained in the knowledge tag but not contained in the operation and maintenance personnel image, wherein lambda and mu are given weight coefficients;

the example similarity operator unit is used for calculating the example similarity according to the operation and maintenance personnel image, and the example similarity calculation formula is as follows:

wherein S is_inst(U, C) is the example similarity between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label, P (U, C) represents the probability that an example randomly extracted from the example space belongs to the concepts U and C at the same time,

representing the probability that an instance randomly drawn from the instance space belongs only to concept U and not to concept C,

represents the probability that an instance randomly drawn from the instance space belongs only to concept C and not to concept U;

the relational similarity calculation operator unit is used for calculating the relational similarity according to the operation and maintenance personnel image to obtain the relational similarity S between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label_relat(U, C), wherein the relationship comprises a synonymy relationship, an inheritance relationship, and an inclusion relationship, the weight of the synonymy relationship is greater than the weight of the inheritance relationship, and the weight of the synonymy relationship is greater than the weight of the inclusion relationship;

an operation and maintenance personnel-knowledge similarity calculation operator unit for calculating the similarity S according to the name_name(U, C), Attribute similarity S_attri(U, C), example similarity S_inst(U, C) and relationship similarity S_relat(U, C) calculating the similarity of the operation and maintenance personnel and the knowledge, wherein the calculation formula of the similarity of the operation and maintenance personnel and the knowledge Sim (U, C) is that Sim (U, C) is α S_name(U,C)+βS_attri(U,C)+εS_inst(U,C)+δS_relat(U, C) wherein α, β,. epsilon.and.delta.are eachGiven name similarity coefficient, attribute similarity coefficient, instance similarity coefficient, and relationship similarity coefficient.

The invention will be further explained and explained with reference to the drawings and the embodiments in the description.

Example one

Aiming at the problem of low accuracy of knowledge recommendation in the prior art, the invention provides an efficient knowledge recommendation technology applied to an operation and maintenance environment. The knowledge recommendation technology models the use scene of the knowledge base and the use user (namely, operation and maintenance personnel) through deep learning algorithms such as reinforcement learning, and further carries out accurate knowledge popularization and application based on the established model. The recommendation technology has the following characteristics:

(1) the technology is used for training the portrait of the operation and maintenance personnel by off-line learning of historical operation behaviors of the operation and maintenance personnel and continuously correcting the portrait model of the operation and maintenance personnel by combining subsequent operation behavior data of the operation and maintenance personnel, so that the portrait of the operation and maintenance personnel is more and more accurate.

(2) The technology analyzes the system operation scene of the operation and maintenance work in real time, evaluates (namely selects knowledge) by combining the trained operation and maintenance personnel figure, and can accurately recommend the knowledge.

The detailed implementation of the knowledge recommendation of the present invention will be described in detail below in terms of both noun explanation and implementation.

Interpretation of noun

The invention relates to the following terms:

knowledge recommendation: recommendations for knowledge data. The knowledge recommendation pushes and introduces related knowledge to the user through the processing content and the user model of the operation and maintenance platform system user, on one hand, the knowledge recommendation can help the user to accelerate the processing of work, and on the other hand, more deep related information is provided to enhance the user capability.

Deep learning: research from artificial neural networks has led to the discovery of distributed feature representations of data by combining lower-level features to form more abstract higher-level representation attribute classes or features.

A knowledge base: structured, easy-to-operate, easy-to-use and fully organized knowledge clusters in knowledge engineering are interconnected knowledge slice sets which are stored, organized, managed and used in a computer memory by adopting a certain knowledge representation mode (or a plurality of knowledge representation modes) according to the needs of solving problems in a certain (or certain) field. These knowledge pieces include theoretical knowledge related to the domain, factual data, heuristic knowledge derived from expert experience (such as definition, theorem, algorithm, etc. related to a domain), and common knowledge. The difference between general applications and knowledge-based systems is: while general applications implicitly encode problem-solving knowledge in the program, knowledge-based systems explicitly express problem-solving knowledge in the application domain and separately compose a relatively independent program entity.

Operation and maintenance informatization system: the method takes the daily operation maintenance management process of an IT department as a core, takes event tracking as a main line, aims to solve eight management problems (process management, event management, problem management, change management, release management, operation management, knowledge management and comprehensive analysis management) in the IT operation management, and provides an efficient and standard IT operation management platform for the IT department. The system not only realizes the interface with the service system used in the enterprise at present, but also integrates the system functions of customer service, operation and maintenance, service management and the like, and can carry out stage prompt on the responsible person in the forms of mails, mobile phone short messages and the like, thereby improving the service response efficiency of system maintenance; by integrating information, the comprehensive management of various resources is realized, including the effective management of various static resources, basic data and spare part resources, so that the quick response capability of the operation and maintenance of the IT department is comprehensively improved, and meanwhile, a perfect data model is established for the business knowledge accumulation and business assessment of the IT department.

The Boosting method comprises the following steps: the method is used for improving the accuracy of the weak classification algorithm, and can integrate a plurality of classifiers into one classifier.

(II) concrete implementation process of knowledge recommendation method

As shown in fig. 2, taking the application of the knowledge base of the operation and maintenance information system as an example, the knowledge recommendation method of the present invention specifically includes the following steps:

the method comprises the following steps: and carrying out scene analysis to generate a scene mode.

In order to acquire scene information operated by operation and maintenance personnel in real time and carry out scene-knowledge correlation analysis to improve the accuracy of knowledge recommendation, the invention provides a new scene analysis algorithm. As shown in fig. 1, the detailed steps of the new scene analysis algorithm are as follows:

step 1: acquiring scene data information: and acquiring scene data P operated by current operation and maintenance personnel.

Step 2: data cleaning: and (4) carrying out data cleaning on the scene data P, wherein the aim is to carry out examination and verification on the scene data P, delete repeated information, correct existing errors and ensure data consistency.

Step 3: scene mode generation: the process is simply a process of analyzing and labeling the collected and cleaned scene data, and finally the scene mark O can be obtained^C。

Step two: and generating an operation and maintenance personnel portrait.

The invention provides a new operation and maintenance personnel analysis algorithm, the operation and maintenance personnel portrait is constructed by acquiring static data information and dynamic data information of the operation and maintenance personnel and applying machine learning algorithms such as reinforcement learning, the operation and maintenance personnel portrait model can be corrected and adjusted by using the new operation and maintenance personnel data acquired during operation, and the accuracy of the operation and maintenance personnel portrait is improved.

As shown in fig. 2 and fig. 3, the detailed steps of the operation and maintenance personnel analysis algorithm are as follows:

step 1: acquiring operation and maintenance personnel data information: on one hand, data are obtained from the account registration information of the operation and maintenance personnel, and on the other hand, the operation behavior data of the operation and maintenance personnel are collected in real time.

Step 2: data cleaning: and the collected operation and maintenance personnel data information is subjected to data cleaning, so that the operation and maintenance personnel data information is examined and checked, repeated information is deleted, existing errors are corrected, and the data consistency is ensured.

Step 3: text modeling: the data after data cleaning is mainly divided into static information data and dynamic information data. The static information data refers to the relatively stable information of the operation and maintenance personnel, such as name, birth date, gender and the like. The dynamic information data refers to the behavior information of the operation and maintenance personnel, which changes constantly, and comprises daily operation behavior, browsing behavior and the like.

Step 4: training a plurality of weak models: and training a plurality of weak models by using machine learning algorithms such as SVM and the like. When performing text classification, the computer may be made to look at the training samples provided to it (i.e., given training samples): each training sample is composed of a vector (i.e., a vector composed of text features) and a class label (for indicating which class this training sample belongs to), such as D_i＝(x_i,y_i) In x_iThat is, a text vector (which is higher in dimensionality), y_iAre the class labels. In a binary linear classification, this class label has only two values, 1 and-1 (used to indicate whether it belongs to or does not belong to this class, respectively). With this representation, the spacing of a sample point from a hyperplane can be defined as: delta_i＝y_i(wx_i+b)。

Step 5: boosting accuracy of a plurality of weak models: obtaining a training sample subset through the operation of a training sample set, then training the training sample subset by using a weak classification algorithm to generate a series of base classifiers, and finally obtaining a result classifier of the operation and maintenance personnel model through a Boosting method.

The method comprises the steps of putting other weak classification algorithms except Step4 into a Boosting framework as a basis classification algorithm, obtaining different training sample subsets through the operation of the Boosting framework on a training sample set, and then using the training sample subsets to train to generate a basis classifier (each time a training sample subset is obtained, one basis classifier is generated on the training sample subset by using one basis classification algorithm, so that the number n of training rounds is given₀Then n can be generated₀Individual basis classifier), and then the n is processed by Boosting framework algorithm₀Weighted fusion by individual basis classifierAnd a final result classifier is generated.

Step 6: and (3) model verification: model verification is carried out on the trained portrait model of the operation and maintenance personnel (the model verification can be completed by sampling or given test samples) so as to ensure the accuracy of the model.

Step 7: and (4) saving the model: and storing the preliminary operation and maintenance personnel image model after model verification, and then acquiring new operation and maintenance personnel data information in real time in the operation process to continuously correct the operation and maintenance personnel image model and the operation and maintenance personnel image result.

Step 8: and (3) calculating the similarity of the operation and maintenance personnel and the knowledge: and calculating the similarity Sim (U, C) of the operation and maintenance personnel-knowledge according to the concept name, the attribute, the example and the relationship in the operation and maintenance personnel portrait.

The process of calculating the operation and maintenance personnel-knowledge similarity can be further subdivided into:

1) and (3) performing name similarity calculation:

wherein S is_name(U, C) is the name similarity between the concept U in the operation and maintenance personnel figure and the concept C in the knowledge label, U_i(i is more than or equal to 1 and less than or equal to n) is a character string semantic word segmentation result of the name U in the concept U, c_j(j is more than or equal to 1 and less than or equal to m) is the result of semantic segmentation of the character string of the name C in the concept C, n is the total number of the character strings of the name U in the concept U, m is the total number of the character strings of the name C in the concept C, and Sim (U_i,c_j) Is u_iAnd c_jSimilarity between them;

2) and performing attribute similarity calculation, wherein the attribute similarity calculation formula is as follows:

wherein S is_attri(U, C) is the attribute similarity between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge tag, U_aAnd C_aRepresenting the set of attributes of U and C, respectively, f being a given non-negative metric function, U_a∩C_aRepresenting a set of two concepts, U and C, having the same attribute, U_a-C_aIs shown onlyIs an attribute set that is present in the operation and maintenance person representation but not in the knowledge tag, C_a-U_aRepresenting the attribute set which is only contained in the knowledge tag but not contained in the operation and maintenance personnel image, wherein lambda and mu are given weight coefficients;

3) example similarity calculations were performed:

wherein S is_inst(U, C) is the example similarity between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label, P (U, C) represents the probability that an example randomly extracted from the example space belongs to the concepts U and C at the same time,

representing the probability that an instance randomly drawn from the instance space belongs only to concept U and not to concept C,

represents the probability that an instance randomly drawn from the instance space belongs only to concept C and not to concept U;

4) calculating the relation similarity according to the operation and maintenance personnel portrait to obtain the relation similarity S between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge label_relat(U, C). The relationships comprise synonymy relationships, inheritance relationships and containment relationships. When the relationship similarity is calculated, the weight of the synonymy relationship is greater than that of the inheritance relationship, and the weight of the synonymy relationship is greater than that of the inclusion relationship;

5) according to the name similarity S_name(U, C), Attribute similarity S_attri(U, C), example similarity S_inst(U, C) and relationship similarity S_relat(U, C) calculating the similarity of the operation and maintenance personnel and the knowledge, wherein the calculation formula of the similarity of the operation and maintenance personnel and the knowledge Sim (U, C) is that Sim (U, C) is α S_name(U,C)+βS_attri(U,C)+εS_inst(U,C)+δS_relat(U, C), wherein α, β, epsilon, and delta are given name similarity coefficient, attribute similarity coefficient, instance similarity coefficient, and relationship similarity coefficient, respectively.

Step three: knowledge selection: and selecting knowledge according to the result of the scene analysis and the portrait of the operation and maintenance personnel to obtain a knowledge recommendation set.

As shown in fig. 4, knowledge selection can be further refined into the following process:

1) searching the characteristics of the operation and maintenance personnel according to the operation and maintenance personnel images, and calculating an operation and maintenance similar knowledge set by combining the operation and maintenance personnel images;

2) acquiring scene characteristics in real time according to the scene analysis result, and calculating a scene similarity knowledge set according to the scene characteristics;

3) and judging whether the scene similar knowledge set and the operation and maintenance similar knowledge set have intersection, if so, forming a recommendation knowledge set according to the intersection, otherwise, re-acquiring scene characteristics, and re-training the operation and maintenance person portrait model to obtain a new operation and maintenance person portrait.

If the knowledge label is U_kThen the set of recommended knowledge services can be represented as K_S. At this time, the detailed refining steps of knowledge selection are as follows:

step 1: carrying out user scene retrieval on operation and maintenance personnel, wherein the user scene U is_mContains a scene u, and

wherein, U_TIs a full user scenario;

step 2: calculating a user scene U by adopting a set similarity calculation method_mSimilar knowledge scenes are sequenced to obtain a similar scene set C_n：

Wherein, U_m≈C_uUser scene U representing operation and maintenance personnel_mAnd knowledge scene C_uA is an intersection symbol, C_TIs a full knowledge scene set;

step 3: retrieving a given scene knowledge relationship model O^RTo obtain a scene C_nKnowledge-scene relationship pair P of_i,j：

Wherein, P_TAs a full-knowledge-scene-relationship pair, C_iFor knowledge scenarios, R₁As scene C_iThe properties of (a) to (b) are,

represents P_i,jBy R₁And C_iConnecting;

step 4: retrieving a given set of knowledge domains O^DTo obtain a compound containing_i,jAssociated domain knowledge set D_k：

Wherein D is_TIn the field of full-scale knowledge, R₂For domain knowledge D_kThe properties of (a) to (b) are,

represents P_i,jBy R₂And D_kConnecting;

step 5: from a full-scale domain knowledge recommendation set

To obtain knowledge recommendation set K_S：

Compared with the prior art, the invention has the following advantages:

(1) high accuracy and high efficiency: the knowledge recommendation technology is adopted, so that the problem that the hit rate of knowledge retrieval is greatly different due to different understanding expressions of individuals on knowledge is avoided, the time consumption of manual knowledge retrieval is reduced, the knowledge application can be popularized in real time, and the processing efficiency of users is improved.

(2) Providing an operation and maintenance personnel portrait: the operation behavior of the user is introduced as input, the user model of the operation and maintenance personnel portrait model is created, knowledge pushing is carried out according to the user model, and the accuracy of knowledge recommendation is improved.

(3) The self-learning capability is provided: through a deep learning algorithm based on machine learning, the operation and maintenance personnel portrait model continuously acquires the operation behaviors of the operation and maintenance personnel to correct the model, so that the operation and maintenance personnel portrait model is advanced along with time, the operation times of the operation and maintenance personnel are increased, and the accuracy is higher and higher.

While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (6)

1. A knowledge recommendation method based on big data and deep learning is characterized in that: the method comprises the following steps:

performing scene analysis according to the scene data information;

processing the data information of the operation and maintenance personnel by adopting a deep learning method to generate an operation and maintenance personnel portrait;

selecting knowledge according to a scene analysis result and the operation and maintenance personnel portrait to obtain a knowledge recommendation set;

performing knowledge recommendation according to the knowledge recommendation set;

the operation and maintenance personnel data information is processed by adopting a deep learning method to generate an operation and maintenance personnel portrait, and the operation and maintenance personnel portrait generation method comprises the following steps:

acquiring operation and maintenance personnel data information, wherein the operation and maintenance personnel data information comprises data acquired from information registered by an operation and maintenance personnel account and operation behavior data of the operation and maintenance personnel acquired in real time;

carrying out data cleaning on the collected operation and maintenance personnel data information;

carrying out model training on the operation and maintenance personnel by adopting a method of fusing weak model training and Boosting on the data information of the operation and maintenance personnel after data cleaning to obtain an operation and maintenance personnel portrait;

calculating the similarity between the operation and maintenance personnel and the knowledge label according to the operation and maintenance personnel figure to obtain the operation and maintenance personnel-knowledge similarity;

the step of calculating the similarity between the operation and maintenance personnel and the knowledge label according to the operation and maintenance personnel portrait to obtain the operation and maintenance personnel-knowledge similarity comprises the following steps:

and calculating the name similarity according to the operation and maintenance personnel image, wherein the calculation formula of the name similarity is as follows:

wherein S is_name(U, C) is the name similarity between the concept U in the operation and maintenance personnel figure and the concept C in the knowledge label, U_i(i is more than or equal to 1 and less than or equal to n) is a character string semantic word segmentation result of the name U in the concept U, c_j(j is more than or equal to 1 and less than or equal to m) is the result of semantic segmentation of the character string of the name C in the concept C, n is the total number of the character strings of the name U in the concept U, m is the total number of the character strings of the name C in the concept C, and Sim (U_i,c_j) Is u_iAnd c_jSimilarity between them;

and performing attribute similarity calculation according to the operation and maintenance personnel image, wherein the attribute similarity calculation formula is as follows:

wherein S is_attri(U, C) is the attribute similarity between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge tag, U_aAnd C_aRepresenting the set of attributes of U and C, respectively, f being a given non-negative metric function, U_a∩C_aRepresenting a set of two concepts, U and C, having the same attribute, U_a-C_aRepresenting an attribute set, C, that is only present in the representation of the operation and maintenance person but not in the knowledge tag_a-U_aRepresenting the attribute set which is only contained in the knowledge tag but not contained in the operation and maintenance personnel image, wherein lambda and mu are given weight coefficients;

and calculating example similarity according to the operation and maintenance personnel image, wherein the example similarity calculation formula is as follows:

wherein S is_inst(U, C) is the example similarity between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label, and P (U, C) represents the random extraction from the example spaceThe probability that an instance is subordinate to both concepts U and C,

representing the probability that an instance randomly drawn from the instance space belongs only to concept U and not to concept C,

represents the probability that an instance randomly drawn from the instance space belongs only to concept C and not to concept U;

calculating the relation similarity according to the operation and maintenance personnel portrait to obtain the relation similarity S between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge label_relat(U, C), wherein the relationship comprises a synonymy relationship, an inheritance relationship, and an inclusion relationship, the weight of the synonymy relationship is greater than the weight of the inheritance relationship, and the weight of the synonymy relationship is greater than the weight of the inclusion relationship;

according to the name similarity S_name(U, C), Attribute similarity S_attri(U, C), example similarity S_inst(U, C) and relationship similarity S_relat(U, C) calculating the similarity of the operation and maintenance personnel and the knowledge, wherein the calculation formula of the similarity of the operation and maintenance personnel and the knowledge Sim (U, C) is that Sim (U, C) is α S_name(U,C)+βS_attri(U,C)+εS_inst(U,C)+δS_relat(U, C), wherein α, β, epsilon, and delta are given name similarity coefficient, attribute similarity coefficient, instance similarity coefficient, and relationship similarity coefficient, respectively.

2. The knowledge recommendation method based on big data and deep learning of claim 1, wherein: the step of performing scene analysis according to the scene data information includes:

acquiring scene data information in real time, and acquiring scene data operated by current operation and maintenance personnel;

carrying out data cleaning on the acquired scene data;

and carrying out real-time scene analysis and labeling operation on the scene data after data cleaning to obtain a scene mark.

3. The knowledge recommendation method based on big data and deep learning of claim 1, wherein: the operation and maintenance personnel model training is carried out on the operation and maintenance personnel data information after data cleaning by adopting a method of fusing weak model training and Boosting to obtain the operation and maintenance personnel portrait, and the method comprises the following steps:

performing text modeling, so as to divide the operation and maintenance personnel data information after data cleaning into static information data and dynamic information data;

performing weak model training on a given training sample according to the requirement of text modeling to obtain a plurality of weak models;

the accuracy of the weak models is improved by adopting a Boosting method, and a result classifier of the operation and maintenance personnel model is obtained;

performing model verification on a result classifier of the operation and maintenance personnel model by adopting a given test sample;

and storing the operation and maintenance personnel model after the model verification is passed, and acquiring new operation and maintenance personnel data information after data cleaning in real time to continuously correct the operation and maintenance personnel model and the corresponding operation and maintenance personnel portrait.

4. The knowledge recommendation method based on big data and deep learning of claim 1, wherein: the step of selecting knowledge according to the result of the scene analysis and the portrait of the operation and maintenance personnel to obtain a knowledge recommendation set comprises the following steps:

searching the characteristics of the operation and maintenance personnel according to the operation and maintenance personnel images, and calculating an operation and maintenance similar knowledge set by combining the operation and maintenance personnel images;

acquiring scene characteristics in real time according to the scene analysis result, and calculating a scene similarity knowledge set according to the scene characteristics;

and judging whether the scene similar knowledge set and the operation and maintenance similar knowledge set have intersection, if so, forming a recommendation knowledge set according to the intersection, otherwise, re-acquiring scene characteristics, and re-training the operation and maintenance personnel model to obtain a new operation and maintenance personnel portrait.

5. The knowledge recommendation method based on big data and deep learning of claim 1, wherein: the step of selecting knowledge according to the result of the scene analysis and the portrait of the operation and maintenance personnel to obtain a knowledge recommendation set comprises the following steps:

carrying out user scene retrieval on operation and maintenance personnel, wherein the user scene U is_mContains a scene u, and

wherein, U_TIs a full user scenario;

calculating a user scene U by adopting a set similarity calculation method_mSimilar knowledge scenes are sequenced to obtain a similar scene set C_n：

Wherein, U_m≈C_uUser scene U representing operation and maintenance personnel_mAnd knowledge scene C_uA is an intersection symbol, C_TIs a full knowledge scene set;

retrieving a given scene knowledge relationship model O^RTo obtain a scene C_nKnowledge-scene relationship pair P of_i,j：

Wherein, P_TAs a full-knowledge-scene-relationship pair, C_iFor knowledge scenarios, R₁As scene C_iThe properties of (a) to (b) are,

represents P_i,jBy R₁And C_iConnecting;

retrieving a given set of knowledge domains O^DTo obtain a compound containing_i,jAssociated domain knowledge set D_k：

Wherein D is_TIn the field of full-scale knowledge, R₂For domain knowledge D_kThe properties of (a) to (b) are,

represents P_i,jBy R₂And D_kConnecting;

from a full-scale domain knowledge recommendation set

To obtain knowledge recommendation set K_S：

6. A knowledge recommendation system based on big data and deep learning, characterized by: the method comprises the following steps:

the scene analysis module is used for carrying out scene analysis according to the scene data information;

the operation and maintenance personnel portrait generation module is used for processing the data information of the operation and maintenance personnel by adopting a deep learning method to generate an operation and maintenance personnel portrait;

the knowledge selection module is used for selecting knowledge according to the result of the scene analysis and the operation and maintenance personnel portrait to obtain a knowledge recommendation set;

the knowledge recommendation module is used for recommending knowledge according to the knowledge recommendation set;

the operation and maintenance personnel portrait generation module comprises:

the information acquisition unit is used for acquiring operation and maintenance personnel data information, and the operation and maintenance personnel data information comprises data acquired from information registered by an operation and maintenance personnel account and operation behavior data of the operation and maintenance personnel acquired in real time;

the data cleaning unit is used for cleaning the data of the collected operation and maintenance personnel data information;

the model training unit is used for carrying out model training on the operation and maintenance personnel by adopting a method of fusing weak model training and Boosting on the operation and maintenance personnel data information after data cleaning to obtain an operation and maintenance personnel portrait;

the similarity calculation unit is used for calculating the similarity between the operation and maintenance personnel and the knowledge label according to the operation and maintenance personnel portrait to obtain the operation and maintenance personnel-knowledge similarity;

the similarity calculation unit includes:

the name similarity calculation operator unit is used for calculating name similarity according to the operation and maintenance personnel image, and the calculation formula of the name similarity is as follows:

wherein S is_name(U, C) is the name similarity between the concept U in the operation and maintenance personnel figure and the concept C in the knowledge label, U_i(i is more than or equal to 1 and less than or equal to n) is a character string semantic word segmentation result of the name U in the concept U, c_j(j is more than or equal to 1 and less than or equal to m) is the result of semantic segmentation of the character string of the name C in the concept C, n is the total number of the character strings of the name U in the concept U, m is the total number of the character strings of the name C in the concept C, and Sim (U_i,c_j) Is u_iAnd c_jSimilarity between them;

the attribute similarity calculation operator unit is used for calculating the attribute similarity according to the operation and maintenance personnel image, and the attribute similarity calculation formula is as follows:

wherein S is_attri(U, C) is the attribute similarity between the concept U in the operation and maintenance personnel portrait and the concept C in the knowledge tag, U_aAnd C_aRepresenting the set of attributes of U and C, respectively, f being a given non-negative metric function, U_a∩C_aRepresenting a set of two concepts, U and C, having the same attribute, U_a-C_aRepresenting an attribute set, C, that is only present in the representation of the operation and maintenance person but not in the knowledge tag_a-U_aRepresenting the attribute set which is only contained in the knowledge tag but not contained in the operation and maintenance personnel image, wherein lambda and mu are given weight coefficients;

the example similarity operator unit is used for calculating the example similarity according to the operation and maintenance personnel image, and the example similarity calculation formula is as follows:

wherein S is_inst(U, C) is the example similarity between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label, P (U, C) represents the probability that an example randomly extracted from the example space belongs to the concepts U and C at the same time,

representing the probability that an instance randomly drawn from the instance space belongs only to concept U and not to concept C,

represents the probability that an instance randomly drawn from the instance space belongs only to concept C and not to concept U;

the relational similarity calculation operator unit is used for calculating the relational similarity according to the operation and maintenance personnel image to obtain the relational similarity S between the concept U in the operation and maintenance personnel image and the concept C in the knowledge label_relat(U, C), wherein the relationship comprises a synonymy relationship, an inheritance relationship, and an inclusion relationship, the weight of the synonymy relationship is greater than the weight of the inheritance relationship, and the weight of the synonymy relationship is greater than the weight of the inclusion relationship;

an operation and maintenance personnel-knowledge similarity calculation operator unit for calculating the similarity S according to the name_name(U, C), Attribute similarity S_attri(U, C), example similarity S_inst(U, C) and relationship similarity S_relat(U, C) calculating the similarity of the operation and maintenance personnel and the knowledge, wherein the calculation formula of the similarity of the operation and maintenance personnel and the knowledge Sim (U, C) is that Sim (U, C) is α S_name(U,C)+βS_attri(U,C)+εS_inst(U,C)+δS_relat(U, C), wherein α, β, epsilon, and delta are given name similarity coefficient, attribute similarity coefficient, instance similarity coefficient, and relationship similarity coefficient, respectively.

Images