CN105404687A - Personalized recommendation method and system for learning behavior - Google Patents

Abstract

The present invention discloses a personalized recommendation method and system for a learning behavior. The method comprises the following steps of: S1, acquiring information of a learner-learning resource matrix; S2, calculating the sparsity of the learner-learning resource matrix, and if the sparsity is less than alpha, mining the learning behavior of a learner, otherwise, performing the step S4; S3, according to the mined learning behavior of the learner, calculating a learning resource that is not actively evaluated to obtain a final weight value of the learning behavior of the learner, and replacing a corresponding position of the learner-learning resource matrix with the final weight value; S4, calculating the similarity to obtain a set of learners with most similar learning demands; and S5, pushing the most similar learning resource to the learner. According to the personalized recommendation method and the system for the learning behavior, provided by the present invention, the utilization rate of online learning resources is increased, the sparsity problem existent in an original collaborative filtering algorithm is avoided, and a factor that the learning of the learner needs to be transferred with time change is considered, so that the accuracy of learning resource recommendation is finally improved.

Description

The personalized recommendation method of learning behavior and system

Technical field

The present invention relates to a kind of data mining and work in coordination with field, particularly relate to a kind of personalized recommendation method and system of learning behavior.

Background technology

In learning process, oneself due principal status of public economy is experienced in order to make learner, under the restriction not being subject to time and space, learner not only can select to study in coordination with other learners, also independent study can be carried out according to the used mode of learning had of self, be necessary the learning behavior of learner is recorded and analyzes, thus for learner recommend quickly and accurately effectively resource and mode of learning to improve the learning efficiency of learner.

For personalized recommendation, be the significant challenge that current personalized recommendation shifts from ecommerce to learning platform to the tracking of learner's learning behavior and the basis of record carry out personalized recommendation to learner.Nowadays most popular personalized recommendation technology can be divided into the recommendation, information filtering recommendation, collaborative filtering recommending etc. three kinds based on correlation rule.Collaborative filtering recommending technology is then the recommend method be most widely used.But the cold start-up itself carried, the problem such as openness all can have a great impact the accurate rate of personalized recommendation.Therefore for variety of problems and the challenge of above-mentioned existence, learner is in the urgent need to more mature and stable personalized recommendation platform.

Summary of the invention

The present invention is intended at least solve the technical matters existed in prior art, especially innovatively proposes a kind of personalized recommendation method and system of learning behavior.

In order to realize above-mentioned purpose of the present invention, the invention provides a kind of personalized recommendation method of learning behavior, comprising the following steps:

S1, obtains learner-education resource matrix information: obtain M learner and mark to the active of N number of education resource, and the education resource scoring of not initiatively scoring is null value, forms M × N matrix E; Wherein element E _i,jrepresent i-th learner to the scoring of a jth education resource; Described M, N are positive integer, and i, j are followed successively by the positive integer being not more than M, N;

S2, calculates learner-education resource matrix openness, if openness Sparsity < α, excavates learner's learning behavior, and wherein, α is the openness threshold value of setting, otherwise performs step S4;

S3, according to the learner's learning behavior excavated, corresponding weights are given respectively to the learning behavior of the education resource not carrying out initiatively scoring, and the final weights of learner's learning behavior are calculated according to the learning behavior of learner, described final weights are the learning behavior weights sum of learner to this education resource; These final weights are substituted on the relevant position in learner-education resource matrix;

S4, draws by Similarity measures learner's set that learner's learning demand is the most similar;

S5, by the most similar pushing learning resource to learner.

Utilize the present invention, give correct recommendation to the learning demand of learner; And along with the change of time is passed, learner's learning interest and learning demand change, thus substantially increase the accuracy rate of recommendation.

In the preferred embodiment of the present invention, openness in step S2 computing method are:

$Sparsity=\frac{EvalTotal}{LearnerTotal\&times;\mathrm{Re}sTotal},$

Wherein, EvalTotal represents the evaluation quantity of learner to education resource, and LearnerTotal represents learner's quantity, and ResTotal represents education resource quantity.

By calculating Sparsity value, preventing learner-education resource matrix too sparse, improve the accuracy of recommendation.

In the preferred embodiment of the present invention, in step S3, learning behavior comprises the behavior of resource access state, learning state behavior and test mode behavior;

The behavior of resource access state comprises collection weights, shares weights and download weights, and their weights are assigned as: collection weights are 3, share weights be 3, download weights be 3, collect+share weights be 4.5, share+download weights be 4.5, collect+share+download weights are 5;

Learning state behavior to comprise in study weights and has learnt weights, and their weights are assigned as: in study, weights are 4.5 and to have learnt weights be 5;

Test mode behavior comprises qualified weights, good weights and outstanding weights, and their weights are assigned as: qualified weights are 3.5, good weights be 4 and outstanding weights be 5.

By carrying out weight assignment to Learner behavior, ensure, when initiatively rating matrix is too sparse for learner, also can accurately recommend.

The present invention is that learning behavior establishes the corresponding weights specific density table of learning behavior, and wherein highest weight value is 5, and weights are higher, shows that learner's learning interest is denseer.

In the preferred embodiment of the present invention, in step S4, the computing method of similarity are:

$sim(u,v)=\frac{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}(W_{u,i}-\stackrel{\&OverBar;}{ru})(W_{v,i}-\stackrel{\&OverBar;}{rv})}{\sqrt{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}{(W_{u,i}-\stackrel{\&OverBar;}{ru})}^2}\sqrt{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}{(W_{v,i}-\stackrel{\&OverBar;}{rv})}^2}},$

Wherein, sim (u, v) is the similarity between learner u and learner v;

W _u,ibe learner u to education resource i, the final weights when openness Sparsity < α or initiatively scoring; Or W during openness Sparsity>=α _u,ifor initiatively scoring or non-active scoring;

W _v,ibe learner v to education resource i, the final weights when openness Sparsity < α or initiatively scoring; W during openness Sparsity>=α _v,ifor initiatively scoring or non-active scoring;

the average score of learner u to all education resources;

the average score of learner v to all education resources;

Item _u,vit is the education resource set that learner u and learner v mark jointly.

Calculate and find out the learner resource collection the most similar to learner, be convenient to recommend learner's study.

In the preferred embodiment of the present invention, the method for pushing of step S5 is:

$P_{u,i}=\stackrel{\&OverBar;}{R_u}+\frac{\underset{X\&Element;N}{\&Sigma;}sim(u,X)(R_{X,i}-\stackrel{\&OverBar;}{R_X})f\left(T_{u,i}\right)}{\underset{X\&Element;N}{\&Sigma;}\left|sim(u,X)\right|f\left(T_{u,i}\right)},$

Wherein, P _u,ifor learner u is to the scoring of education resource i;

the mean value of all score values of having marked of learner u;

Sim (u, X) represents that learner u and learner neighbours concentrate the similarity of certain learner;

R _x,ithat learner neighbours concentrate certain learner to the scoring of education resource i;

that learner neighbours concentrate certain learner to the average score of education resource;

X is one of the most similar learner neighbours learner of collecting in N;

The function of time span is (0,1); T _u,ithat learner u produces the time of learning interest and demand to education resource i;

N show that the most similar learner neighbours concentrate the number of taking out the highest education resource of similarity.

Recommendation education resource of the present invention changes along with the change of time, and the learning demand of learner also can constantly change, and such learner can not only learn a kind of education resource, thus substantially increases the accuracy rate of recommendation.

In the preferred embodiment of the present invention, to the method that learning behavior excavates be:

Obtain the behavior record to learner in server log file, comprise the id of learner, and the learning behavior of learner, comprise and share, download, learn;

Set up two-dimensional matrix, deposit learner id and this learner to the learning behavior of certain resource;

According to the behavior weight table of setting in advance, assignment is carried out to education resource;

Corresponding value is added in corresponding learner-education resource rating matrix, improves and recommend accuracy.

The invention also discloses a kind of personalized recommendation system of learning behavior, comprising: learner's module, for preserving user profile, comprising personal information and learning interest; Learning behavior log pattern, is divided into explicit behavior and implicit expression behavior by learner's learning behavior; Education resource module, for preserving education resource, comprises text, Voice & Video; Data acquisition module, for by learner after login learning platform, recording and tracking is carried out in the behavior operation carried out education resource; Recommending module, for carrying out the recommendation of personalized education resource to learner; The information that in data collecting module collected learner module, each learner records in learning behavior log pattern, this information is obtained recommending education resource to be kept in education resource module according to the personalized recommendation method of learning behavior of the present invention in recommending module, finally the most similar education resource in education resource module is recommended learner.

In the preferred embodiment of the present invention, after user enters learning system, if new user, then need to learner carry out personal information, learning interest acquisition and be saved to learner's module; Non-new user supplements up-to-date learning demand and learning interest and is saved to learner's module.

In the preferred embodiment of the present invention, explicit behavior is the information that learner's active is submitted to system, comprises the personal information that new user submits to when registering;

Implicit expression behavior is the learning behavior of the implicit expression of learner, in comprising the download to education resource, collect, share, learning, learn, qualified, good and outstanding.

In the preferred embodiment of the present invention, described data acquisition module is the learning behavior gathering learner, and described learning behavior comprises the behavior of resource access state, learning state behavior and test mode behavior;

The behavior of resource access state comprises collection, shares, downloads, collects+share, shares+download, collects+share+download;

Learning state behavior, comprises study neutralization and learns;

Test mode behavior, comprises qualified, good and outstanding.

The learning demand of personalized recommendation system to learner of learning behavior of the present invention gives correct recommendation; Substantially increase the accuracy rate of recommendation.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the personalized recommendation method of learning behavior of the present invention;

Fig. 2 is the module diagram of the personalized recommendation system of learning behavior of the present invention.

Embodiment

Be described below in detail embodiments of the invention, the example of described embodiment is shown in the drawings, and wherein same or similar label represents same or similar element or has element that is identical or similar functions from start to finish.Being exemplary below by the embodiment be described with reference to the drawings, only for explaining the present invention, and can not limitation of the present invention being interpreted as.

The invention discloses a kind of personalized recommendation method of learning behavior, as shown in Figure 1, comprise the following steps:

The first step, obtains learner-education resource matrix information, is specially acquisition M learner and marks to the active of N number of education resource, and the education resource scoring of initiatively scoring is null value, forms M × N matrix E; Wherein element E _i,jrepresent i-th learner to the scoring of a jth education resource; M, N are positive integer, and i, j are followed successively by the positive integer being not more than M, N.After learner enters learning platform, all learning behaviors of learner all can be recorded in journal file by detailed, system obtains useful data by the journal file produced from web server, such as learner enters or leaves the time of learning platform, to the download of education resource, collects or shares.

In the present embodiment, marking system (the initiatively scoring employing prior art to education resource is provided in learning platform, be similar to user in Amazon or Taobao and buy certain commodity, then system requirements you to carry out scoring the same), after learner learnt certain education resource, can select to mark to resource, so scoring has just been stored into learner-education resource rating matrix, this is initiatively scoring (explicit), but after also having learner to learn certain resource, do not go scoring, scoring is 0 point, be null value, rating matrix so will be caused sparse, affect resource recommendation effect.

Second step, calculates learner-education resource matrix openness, if openness Sparsity < α, excavates learner's learning behavior, performs the 3rd step, and wherein, α is the openness threshold value of setting, otherwise performs the 4th step.In the present embodiment, openness computing method are:

$Sparsity=\frac{EvalTotal}{LearnerTotal\&times;\mathrm{Re}sTotal},$

Wherein, EvalTotal represents the evaluation quantity of learner to education resource, and LearnerTotal represents learner's quantity, and ResTotal represents education resource quantity.

3rd step, excavates learning behavior, and concrete method is:

Learning behaviors such as obtaining the behavior record to learner in server log file, comprise the id of learner, and the learning behavior of learner, can include but not limited to share, download, learnt.Set up two-dimensional matrix, deposit learner id and this learner to the learning behavior of certain resource.According to the behavior weight table of setting in advance, assignment is carried out to education resource; In the present embodiment, weight table is the corresponding weights of each behavior.Corresponding value is added in corresponding learner-education resource rating matrix.

According to the learner's learning behavior excavated, corresponding weights are given respectively to the learning behavior of the education resource not carrying out initiatively scoring, and the final weights of learner's learning behavior are calculated according to the learning behavior of learner, final weights are the learning behavior weights sum of learner to this education resource; These final weights are substituted on the relevant position in learner-education resource matrix.In the present embodiment, learning behavior comprises the behavior of resource access state, learning state behavior and test mode behavior.

The behavior of resource access state comprises collection weights, shares weights and download weights, and their weights are assigned as: collection weights are 3, share weights be 3, download weights be 3, collect+share weights be 4.5, share+download weights be 4.5, collect+share+download weights are 5.

Learning state behavior to comprise in study weights and has learnt weights, and their weights are assigned as: in study, weights are 4.5 and to have learnt weights be 5.

Test mode behavior comprises qualified weights, good weights and outstanding weights, and their weights are assigned as: qualified weights are 3.5, good weights be 4 and outstanding weights be 5.

By arranging weights like this, reasonably calculating the score value of education resource, when the matrix of learner to education resource active evaluation is very sparse, also can ensure the accuracy of recommending.

4th step, draws by Similarity measures learner's set that learner's learning demand is the most similar.In the present embodiment, the computing method of similarity are:

$sim(u,v)=\frac{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}(W_{u,i}-\stackrel{\&OverBar;}{ru})(W_{v,i}-\stackrel{\&OverBar;}{rv})}{\sqrt{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}{(W_{u,i}-\stackrel{\&OverBar;}{ru})}^2}\sqrt{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}{(W_{v,i}-\stackrel{\&OverBar;}{rv})}^2}},$

Wherein, sim (u, v) is the similarity between learner u and learner v;

W _u,ibe learner u to education resource i, the final weights when openness Sparsity < α or initiatively scoring; Or W during openness Sparsity>=α _u,ifor initiatively scoring or non-active scoring;

W _v,ibe learner v to education resource i, the final weights when openness Sparsity < α or initiatively scoring; W during openness Sparsity>=α _v,ifor initiatively scoring or non-active scoring;

the average score of learner u to all education resources;

the average score of learner v to all education resources;

Item _u,vit is the education resource set that learner u and learner v mark jointly.

5th step, by the most similar pushing learning resource to learner.In the present embodiment, method for pushing is:

$P_{u,i}=\stackrel{\&OverBar;}{R_u}+\frac{\underset{X\&Element;N}{\&Sigma;}sim(u,X)(R_{X,i}-\stackrel{\&OverBar;}{R_X})f\left(T_{u,i}\right)}{\underset{X\&Element;N}{\&Sigma;}\left|sim(u,X)\right|f\left(T_{u,i}\right)},$

Wherein, P _u,ifor learner u is to the scoring of education resource i;

the mean value of all score values of having marked of learner u;

Sim (u, X) represents that learner u and learner neighbours concentrate the similarity of certain learner;

R _x,ithat learner neighbours concentrate certain learner to the scoring of education resource i;

that learner neighbours concentrate certain learner to the average score of education resource;

X is one of the most similar learner neighbours learner of collecting in N;

The function of time span is (0,1); T _u,ithat learner u produces the time of learning interest and demand to education resource i;

N show that the most similar learner neighbours concentrate the number of taking out the highest education resource of similarity.

The invention also discloses a kind of personalized recommendation system of learning behavior, as shown in Figure 2, it comprises: learner's module, for preserving user profile, comprises personal information and learning interest; In this embodiment, after user enters learning system, if new user, then need to learner carry out personal information, learning interest acquisition and be saved to learner's module; Non-new user supplements up-to-date learning demand and learning interest and is saved to learner's module.Learning behavior log pattern, is divided into explicit behavior and implicit expression behavior by learner's learning behavior.In the present embodiment, explicit behavior is the information that learner's active is submitted to system, comprises the personal information that new user submits to when registering; Implicit expression behavior is the learning behavior of the implicit expression of learner, in comprising the download to education resource, collect, share, learning, learn, qualified, good and outstanding.Education resource module, for preserving education resource, comprises text, Voice & Video; Data acquisition module, for by learner after login learning platform, recording and tracking is carried out in the behavior operation carried out education resource.In the present embodiment, data acquisition module is the learning behavior gathering learner, and learning behavior comprises the behavior of resource access state, learning state behavior and test mode behavior; The behavior of resource access state comprises collection, shares, downloads, collects+share, shares+download, collects+share+download; Learning state behavior, comprises study neutralization and learns; Test mode behavior, comprises qualified, good and outstanding.Recommending module, for carrying out the recommendation of personalized education resource to learner.The information that in data collecting module collected learner module, each learner records in learning behavior log pattern, this information utilized the personalized recommendation method of learning behavior of the present invention to obtain recommending education resource to be kept in education resource module in recommending module, finally the most similar education resource in education resource module is recommended learner.

This recommend method and system can be combined with E-learning platform, do not enter and can provide more personalized learning experience for E-learning platform learner, and the development of E-learning platform will be promoted further.

In the description of this instructions, specific features, structure, material or feature that the description of reference term " embodiment ", " some embodiments ", " example ", " concrete example " or " some examples " etc. means to describe in conjunction with this embodiment or example are contained at least one embodiment of the present invention or example.In this manual, identical embodiment or example are not necessarily referred to the schematic representation of above-mentioned term.And the specific features of description, structure, material or feature can combine in an appropriate manner in any one or more embodiment or example.

Although illustrate and describe embodiments of the invention, those having ordinary skill in the art will appreciate that: can carry out multiple change, amendment, replacement and modification to these embodiments when not departing from principle of the present invention and aim, scope of the present invention is by claim and equivalents thereof.

Claims (10)

1. a personalized recommendation method for learning behavior, is characterized in that, comprises the following steps:

S1, obtains learner-education resource matrix information: obtain M learner and mark to the active of N number of education resource, and the education resource scoring of not initiatively scoring is null value, forms M × N matrix E; Wherein element E _i,jrepresent i-th learner to the scoring of a jth education resource; Described M, N are positive integer, and i, j are followed successively by the positive integer being not more than M, N;

S2, calculates learner-education resource matrix openness, if openness Sparsity < α, excavates learner's learning behavior, performs step S3, otherwise performs step S4, and wherein, α is the openness threshold value of setting;

S3, according to the learner's learning behavior excavated, corresponding weights are given respectively to the learning behavior of the education resource not carrying out initiatively scoring, and the final weights of learner's learning behavior are calculated according to the learning behavior of learner, described final weights are the learning behavior weights sum of learner to this education resource; These final weights are substituted on the relevant position in learner-education resource matrix;

S4, draws by Similarity measures learner's set that learner's learning demand is the most similar;

S5, by the most similar pushing learning resource to learner.

2. the personalized recommendation method of learning behavior according to claim 1, is characterized in that, computing method openness in step S2 are:

$Sparsity=\frac{EvalTotal}{LearnerTotal\&times;\mathrm{Re}sTotal},$

Wherein, EvalTotal represents the evaluation quantity of learner to education resource, and LearnerTotal represents learner's quantity, and ResTotal represents education resource quantity.

3. the personalized recommendation method of learning behavior according to claim 1, is characterized in that, in step S3, learning behavior comprises the behavior of resource access state, learning state behavior and test mode behavior;

The behavior of resource access state comprises collection weights, shares weights and download weights, and weights are assigned as: collection weights are 3, share weights be 3, download weights be 3, collect+share weights be 4.5, share+download weights be 4.5, collect+share+download weights are 5;

Learning state behavior to comprise in study weights and has learnt weights, and weights are assigned as: in study, weights are 4.5 and to have learnt weights be 5;

Test mode behavior comprises qualified weights, good weights and outstanding weights, and weights are assigned as: qualified weights are 3.5, good weights be 4 and outstanding weights be 5.

4. the personalized recommendation method of learning behavior according to claim 1, is characterized in that, in step S4, the computing method of similarity are:

$sim(u,v)=\frac{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}(W_{u,i}-\stackrel{\&OverBar;}{ru})(W_{v,i}-\stackrel{\&OverBar;}{rv})}{\sqrt{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}{(W_{u,i}-\stackrel{\&OverBar;}{ru})}^2}\sqrt{\underset{i\&Element;{\mathrm{Item}}_{u,v}}{\&Sigma;}{(W_{v,i}-\stackrel{\&OverBar;}{rv})}^2}},$

Wherein, sim (u, v) is the similarity between learner u and learner v;

W _u,ibe learner u to education resource i, the final weights when openness Sparsity < α or initiatively scoring; Or W during openness Sparsity>=α _u,ifor initiatively scoring or non-active scoring;

W _v,ibe learner v to education resource i, the final weights when openness Sparsity < α or initiatively scoring; W during openness Sparsity>=α _v,ifor initiatively scoring or non-active scoring;

the average score of learner u to all education resources;

the average score of learner v to all education resources;

Item _u,vit is the education resource set that learner u and learner v mark jointly.

5. the personalized recommendation method of learning behavior according to claim 1, is characterized in that, the method for pushing of step S5 is:

$P_{u,i}=\stackrel{\&OverBar;}{R_u}+\frac{\underset{X\&Element;N}{\&Sigma;}sim(u,X)(R_{X,i}-\stackrel{\&OverBar;}{R_X})f\left(T_{u,i}\right)}{\underset{X\&Element;N}{\&Sigma;}\left|sim(u,X)\right|f\left(T_{u,i}\right)},$

Wherein, P _u,ifor learner u is to the scoring of education resource i;

the mean value of all score values of having marked of learner u;

Sim (u, X) represents that learner u and learner neighbours concentrate the similarity of certain learner;

R _x,ithat learner neighbours concentrate certain learner to the scoring of education resource i;

that learner neighbours concentrate certain learner to the average score of education resource;

X is one of the most similar learner neighbours learner of collecting in N;

The function of time span is (0,1); T _u,ithat learner u produces the time of learning interest and demand to education resource i;

N show that the most similar learner neighbours concentrate the number of taking out the highest education resource of similarity.

6. the personalized recommendation method of learning behavior according to claim 1, is characterized in that, to the method that learning behavior excavates is:

Obtain the behavior record to learner in server log file, comprise the id of learner, and the learning behavior of learner, comprise and share, download, learn;

Set up two-dimensional matrix, deposit learner id and this learner to the learning behavior of certain resource;

According to the behavior weight table of setting in advance, assignment is carried out to education resource;

Corresponding value is added in corresponding learner-education resource rating matrix.

7. a personalized recommendation system for learning behavior, is characterized in that, comprising:

Learner's module, for preserving user profile, comprises personal information and learning interest;

Learning behavior log pattern, is divided into explicit behavior and implicit expression behavior by learner's learning behavior;

Education resource module, for preserving education resource, comprises text, Voice & Video;

Data acquisition module, for by learner after login learning platform, recording and tracking is carried out in the behavior operation carried out education resource;

Recommending module, for carrying out the recommendation of personalized education resource to learner;

The information that in data collecting module collected learner module, each learner records in learning behavior log pattern, the personalized recommendation method of this information described learning behavior according to claim 1 in recommending module is obtained recommending education resource to be kept in education resource module, finally the most similar education resource in education resource module is recommended learner.

8. the personalized recommendation system of learning behavior according to claim 7, is characterized in that, after user enters learning system, if new user, then need to learner carry out personal information, learning interest acquisition and be saved to learner's module; Non-new user supplements up-to-date learning demand and learning interest and is saved to learner's module.

9. the personalized recommendation system of learning behavior according to claim 7, is characterized in that, explicit behavior is the information that learner's active is submitted to system, comprises the personal information that new user submits to when registering;

Implicit expression behavior is the learning behavior of the implicit expression of learner, in comprising the download to education resource, collect, share, learning, learn, qualified, good and outstanding.

10. the personalized recommendation system of learning behavior according to claim 7, is characterized in that, described data acquisition module is the learning behavior gathering learner, and described learning behavior comprises the behavior of resource access state, learning state behavior and test mode behavior;

The behavior of resource access state comprises collection, shares, downloads, collects+share, shares+download, collects+share+download;

Learning state behavior, comprises study neutralization and learns;

Test mode behavior, comprises qualified, good and outstanding.