CN110851734B - Content recommendation method and device - Google Patents
Content recommendation method and device Download PDFInfo
- Publication number
- CN110851734B CN110851734B CN201911061257.7A CN201911061257A CN110851734B CN 110851734 B CN110851734 B CN 110851734B CN 201911061257 A CN201911061257 A CN 201911061257A CN 110851734 B CN110851734 B CN 110851734B
- Authority
- CN
- China
- Prior art keywords
- content
- feedback information
- parameter
- submatrix
- basic data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The application discloses a content recommendation method and device. Wherein the method comprises the following steps: obtaining basic data of an object, wherein the basic data comprises: object identification, content identification and feedback information of the object to the content; updating the basic data through the correction feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data; performing matrix decomposition on the original matrix based on the objective function to obtain a first submatrix and a second submatrix; and determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object. The method and the device solve the technical problem that in the prior art, when content recommendation is performed based on feedback behaviors of the user, the accuracy of a recommendation result is low due to certain errors of the feedback behaviors of the user.
Description
Technical Field
The application relates to the field of data processing, in particular to a content recommendation method and device.
Background
A knowledge management system (Knowledge Management System, KMS) is an information system that gathers, processes, shares all knowledge of an organization, including knowledge acquisition, storage, distribution, and application processes, supporting learning and decision-making for enterprise employees. Customer service representatives (Customer Service Representative, CSR) such as call centers can learn knowledge points such as product introduction or service standards offline on the one hand and rely on knowledge bases to solve customer problems during online conversation on the other hand based on KMS in the customer service field.
At present, KMS provides functions such as pattern matching and search reasoning to feed back knowledge points required by users, but the method for solving information overload by relying on a search engine has two problems: the retrieval operation is time-consuming and not personalized enough. In order to improve satisfaction of KMS users, knowledge recommendation methods are emerging in various fields.
Content-based knowledge recommendation requires collection of attributes of users and items, which are generally difficult to collect, and there is also insufficient attribute information of users and items in KMSs. Therefore, knowledge recommendation selects collaborative filtering method, collaborative filtering uses collective collaborative information to recommend, in short, uses the preference of a community having a common experience to recommend information of interest to a user, as shown in fig. 1, user C likes article a, and other users a and B like article a also like article C, thereby recommending article a to user C.
The data collected by collaborative filtering is a record of the behavior of the user on the article, i.e., user feedback. User feedback is the user's behavior on the item, such as explicit scoring, implicit number of clicks, etc. Collaborative filtering recommendation is performed based on information fed back by a user, and noise or deviation exists in the information fed back by the user, so that accuracy of recommendation can be affected. User feedback presence noise reflects two aspects: 1) Misleading behavior, such as scoring too high a movie that is played in its lead by particularly enjoying a lead angle; 2) Abnormal behavior, such as an ultrahigh number of searches for a certain knowledge point. The information fed back by the user is directly used for collaborative filtering, and the final recommendation result can be affected to a certain extent.
Aiming at the problem of low accuracy of a recommendation result caused by a certain error of feedback behavior of a user when content recommendation is performed based on the feedback behavior of the user in the prior art, no effective solution is proposed at present.
Disclosure of Invention
The embodiment of the application provides a content recommendation method and device, which at least solve the technical problem of low recommendation result accuracy caused by a certain error of feedback behavior of a user when content recommendation is performed based on the feedback behavior of the user in the prior art.
According to an aspect of an embodiment of the present application, there is provided a recommendation method for content, including: obtaining basic data of an object, wherein the basic data comprises: object identification, content identification and feedback information of the object to the content; updating the basic data through the correction feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data; performing matrix decomposition on the original matrix based on the objective function to obtain a first submatrix and a second submatrix; and determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object.
Further, updating the basic data by correcting the feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data, including: extracting characteristic parameters in the feedback information; replacing feedback information in the basic data by using the characteristic parameters so as to correct the feedback information; and updating the basic data by using the corrected feedback information, and constructing an original matrix and an objective function for decomposing the original matrix by using the updated basic data.
Further, extracting the characteristic parameters in the feedback information includes: extracting difficulty parameters of the content and capability parameters of the object; determining a probability density function according to the difficulty parameter of the content and the capability parameter of the object, wherein the probability density function is used for representing the probability of the object grasping the content; and determining the probability density function as a characteristic parameter.
Further, extracting a difficulty parameter of the content and a capability parameter of the object, including: determining the average value of the times of accessing the content by a plurality of objects as a difficulty coefficient corresponding to the content; the average of the number of times the object accesses the plurality of contents is determined as the capability coefficient of the object.
Further, determining a probability density function based on the difficulty parameter of the content and the capability parameter of the object, comprising: and determining a probability density function based on the Gaussian probability density function, wherein a scale function of the Gaussian probability density function is a capability parameter of the object, a random variable of the Gaussian probability density function is feedback information of the object on the content, and a position parameter of the Gaussian probability density function is a difficulty coefficient of the content.
Further, constructing an objective function for decomposing the original matrix using the updated base data, including: obtaining an error function of feedback information of any object to any content, wherein the error function is the difference between actual feedback information of the object to the content and predictive feedback information of the object to the content, and the predictive feedback information is obtained through prediction of a first submatrix and a second submatrix; acquiring a first sum of error functions of each object for each content; acquiring a regular term of an objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix and a norm square of a second submatrix; the first sum value and the second sum value of the regularization term are determined as an objective function for decomposing the original matrix.
Further, updating the basic data by correcting the feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data, including: extracting characteristic parameters in the feedback information; constructing an original matrix by using the basic data; and constructing an objective function for decomposing the original matrix based on the characteristic parameters and the updated basic data.
Further, constructing an objective function for decomposing the original matrix based on the feature parameters and the updated base data, including: obtaining an error function of feedback information of any object to any content, wherein the error function is the difference between characteristic parameters of the object to the content and predictive feedback information of the object to the content, and the predictive feedback information is obtained through prediction of a first submatrix and a second submatrix; obtaining a first sum of squares of error functions of each object for each content; acquiring a regular term of an objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix, a norm square of a second submatrix, a square of a difficulty coefficient of content and a square of a capability coefficient of an object; the first sum value and the second sum value of the regularization term are determined as an objective function for decomposing the original matrix.
Further, extracting the characteristic parameters in the feedback information includes: extracting difficulty parameters of the content and capability parameters of the object; determining a probability density function according to the difficulty parameter of the content and the capability parameter of the object, wherein the probability density function is used for representing the probability of the object grasping the object; and determining the probability density function as a characteristic parameter.
Further, constructing an objective function for decomposing the original matrix based on the feature parameters and the updated base data, including: obtaining an error function of feedback information of any object to any content, wherein a first difference value of the feedback information of the object to the content and a characteristic parameter is obtained, a second difference value of the first difference value and a predicted feedback information is obtained, the second difference value is determined to be the error function, and the predicted feedback information is obtained through prediction of a first submatrix and a second submatrix; obtaining a first sum of squares of error functions of each object for each content; acquiring a regular term of an objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix, a norm square of a second submatrix, a square of a difficulty coefficient of content and a square of a capability coefficient of an object; the first sum value and the second sum value of the regularization term are determined as an objective function for decomposing the original matrix.
Further, extracting the characteristic parameters in the feedback information includes: extracting difficulty parameters of the content and capability parameters of the object; and determining characteristic parameters according to the difficulty parameters of the content and the capability parameters of the object based on the item reaction theoretical model.
According to an aspect of an embodiment of the present application, there is provided a recommendation apparatus for content, including: the system comprises an acquisition module, a storage module and a storage module, wherein the acquisition module is used for acquiring basic data of an object, and the basic data comprise: object identification, content identification and feedback information of the object to the content; the correction module is used for updating the basic data through correction feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data; the decomposition module is used for carrying out matrix decomposition on the original matrix based on the objective function to obtain a first submatrix and a second submatrix; and the determining module is used for determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object.
According to an aspect of an embodiment of the present application, there is provided a computer storage medium storing a plurality of instructions adapted to be loaded by a processor and to perform the method steps of any one of the above.
According to an aspect of an embodiment of the present application, there is provided an intelligent interactive tablet, including: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to perform the method steps of any of the above.
In the embodiment of the application, basic data of an object is acquired, wherein the basic data comprises: object identification, content identification and feedback information of the object to the content; updating the basic data through the correction feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data; performing matrix decomposition on the original matrix based on the objective function to obtain a first submatrix and a second submatrix; and determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object. The technical problems that the accuracy of the recommendation result is low due to certain errors of the feedback behavior of the user when content recommendation is performed based on the feedback behavior of the user in the prior art are solved by correcting the feedback information in the basic data to provide noise in the feedback information and recommending knowledge points by adopting matrix decomposition (Matrix Factorization, MF) technology.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute a limitation on the application. In the drawings:
FIG. 1 is a schematic diagram of knowledge recommendation using collaborative filtering methods according to the prior art;
FIG. 2 is a project characteristic of IPT;
FIG. 3 is a flow chart of a method of recommending content according to an embodiment of the present application;
FIG. 4a is a diagram of R in accordance with an embodiment of the present application m×n Schematic of (2);
FIG. 4b is a diagram of R in accordance with an embodiment of the present application m×k Schematic of (2);
FIG. 4c is a Q in accordance with an embodiment of the present application k×m Schematic of (2);
FIG. 5 is a flowchart of another content recommendation method according to an embodiment of the present application;
FIG. 6 is a schematic diagram of probabilistic modeling of potential features in an identified scene;
FIG. 7 is a flowchart of a recommendation method for yet another content according to an embodiment of the present application; and
fig. 8 is a schematic diagram of a content recommendation apparatus according to an embodiment of the present application.
The terms appearing in the embodiments of the present application are explained below:
IRT: item Response Theory based on the project response theory, IRT belongs to one of the common theories of cognitive diagnosis in psychology, i.e. the cognitive condition of a tested person is estimated according to the answer of the tested person to a certain question. The essence of the item is the test question, and the response is the answer of the tested person. IRT can also be said to be a category in machine learning, i.e., a category (capability level) to which a subject belongs is determined from the response of the subject to a test question.
Rasch and Lord respectively propose IRT models, and in view of knowledge recommendation scenes without considering item distinction, the method adopts the models proposed by Rasch to introduce IRT theory fed back by users. The Rasch model is also known as a parametric Logistic model, as follows:
where b represents the project difficulty coefficient and θ represents the user ability coefficient. Fig. 2 is an item characteristic curve (Item Characteristic Curve, ICC) of IPT, which can be used to describe the relationship of item (problem) difficulty and user capability, where the abscissa represents the user capability value and the ordinate represents the probability that the user has made the problem. Obviously, the lower the capability value, the smaller the probability of making the question, and the higher the capability value, the greater the probability of making the question. Taking a student answer as an example, the project is a problem with different difficulties, the user is a student with different abilities, the reaction is the result of the student answer (such as 0 represents an answer mistake and 1 represents an answer pair), and the probability of the student answer pair with higher abilities can be seen in ICC is higher.
Detailed Description
In order that those skilled in the art will better understand the present application, a technical solution in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present application without making any inventive effort, shall fall within the scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Example 1
According to an embodiment of the present application, there is provided an embodiment of a recommended method of content, it being noted that the steps shown in the flowcharts of the drawings may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in an order different from that herein.
Fig. 3 is a flowchart of a content recommendation method according to an embodiment of the present application, as shown in fig. 3, the method including the steps of:
step S102, basic data of the object is acquired, wherein the basic data comprises: object identification, content identification and feedback information of the object to the content.
Specifically, the object may be a user who needs to recommend content to the object, and the content may be knowledge to be recommended. In an alternative embodiment, the device recommends problems to the user, which are defined above, when the user uses interactive smart tablet learning. The object identifier may be account information of the user in the recommendation application, the content identifier is information such as a serial number of the content in the recommendation application or recommendation equipment, and the feedback information of the object to the content may be the number of times the user reads the content, the clicking behavior, and the like.
In an alternative embodiment, the above-described base data may be described using triplet information, which may be collected from KMSs, where the data may be obtained from a user's history of using a recommended application. The triplet data may be represented as < u, i, r >, where u represents the user, i.e., the object identification; i represents knowledge, i.e. content identification; r represents the number of times the user u reads the knowledge i, i.e. feedback information of the object on the content.
Step S104, updating the basic data through the correction feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data.
Specifically, the original matrix is formed based on the updated basic data. Taking basic data as triplet information < u, i, R > as an example, if the correction feedback information is to replace 'R' in the triplet information with 'f (R; mu, sigma)', the original matrix R is < u, i, f (R; mu, sigma) >.
The objective function may be a minimum objective function, which is a function of two sub-matrices obtained by decomposing the original matrix, and when the minimum objective function reaches its minimum value, the decomposition result of the original matrix is the best result.
In the above scheme, the correction of the feedback information may be performed by performing feature extraction on the feedback information to perform standardization processing on the feedback information, so as to reduce the influence of abnormal behavior on the recommendation result. After the feedback information is corrected, the corrected feedback information is used for updating the basic data, and then the updated basic data is used for constructing an original matrix and/or an objective function for decomposing the original matrix.
In an alternative embodiment, the feedback information is modified, which may be that the feedback information is replaced by the feature information after the feature extraction is performed on the feedback information. In another alternative embodiment, the feedback information is modified, or after the feedback information is extracted by the features, the feedback information may be adjusted by using the feature information.
And S106, performing matrix decomposition on the original matrix based on the objective function to obtain a first submatrix and a second submatrix.
The objective function may be a minimum objective function, that is, after the first matrix and the second submatrix obtained by decomposition are brought into the objective function, the objective function may be minimized.
In an alternative embodiment, the original matrix is set as a large sparse matrix R, the first submatrix and the second submatrix are respectively a small matrix P and a small matrix Q, the small matrix P and the small matrix Q have k hidden factors, the R is decomposed to obtain the small matrix P and the small matrix Q, and the P and the Q are used for fitting and approaching the R to obtain a matrix after fittingThe expression can be expressed by the following formula: />Where m represents the number of objects, n represents the number of contents, k is a preset hidden factor, FIG. 4a is an R according to an embodiment of the present application m×n Schematic of (2); FIG. 4b is a diagram of a P in accordance with an embodiment of the present application m×k Schematic of (2); FIG. 4c is a Q in accordance with an embodiment of the present application k×m Is a schematic diagram of (a). As shown in FIG. 4a, FIG. 4b and FIG. 4c, the best decomposition result is the fitted matrix +.>An infinite approximation R.
Step S108, determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object.
The objective function is trained as a matrix factorization model to form two sub-matrices that can be used for content recommendation. In an alternative embodiment, the score of each content allowed to be recommended for the user may be calculated according to the first sub-matrix and the second sub-matrix, the content is ranked from high to low according to the order of the scores, and finally, the first target content in the ranking is selected to be recommended to the user u, or the specified number of content in the ranking before the ranking is selected to be recommended to the user u.
The score for each content allowed to be recommended to the user can be obtained as follows: the two sub-matrices obtained by decomposition are P and Q respectively, wherein P u Represents the ith row, q of matrix P i Representing the ith row of matrix Q, the matrix p is calculated when determining the target content for recommendation to user u based on the first and second sub-matrices u And each q i Inner product of (i), i.eAnd obtaining scores corresponding to the n contents.
As can be seen from the above, the above embodiment of the present application obtains the basic data of the object, where the basic data includes: object identification, content identification and feedback information of the object to the content; updating the basic data through the correction feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data; performing matrix decomposition on the original matrix based on the objective function to obtain a first submatrix and a second submatrix; and determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object. The technical problems that the accuracy of the recommendation result is low due to certain errors of the feedback behavior of the user when content recommendation is performed based on the feedback behavior of the user in the prior art are solved by correcting the feedback information in the basic data to provide noise in the feedback information and recommending knowledge points by adopting matrix decomposition (Matrix Factorization, MF) technology.
Example 2
The present application proposes three embodiments of step S102 in example 1, and one of them will be described in example 2. According to an embodiment of the present application, an embodiment of a method for recommending content is further provided, and it should be noted that fig. 5 is a flowchart of another method for recommending content according to an embodiment of the present application, as shown in fig. 5, the method includes the following steps:
step S102, basic data of an object is acquired, wherein the basic data comprises: object identification, content identification and feedback information of the object to the content.
Step S502, extracting characteristic parameters in the feedback information.
And step S504, replacing feedback information in the basic data by using the characteristic parameters so as to correct the feedback information.
And step S506, updating the basic data by using the corrected feedback information, and constructing an original matrix and an objective function for decomposing the original matrix by using the updated basic data.
And S106, performing matrix decomposition on the original matrix based on an objective function to obtain a first submatrix and a second submatrix.
And step S108, determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object.
Steps S102, S106 and S108 in this embodiment are the same as steps S102, S106 and S108 in embodiment 1, respectively, and are not described here again.
As an optional embodiment, step S502, extracting the characteristic parameters in the feedback information includes:
in step S5021, the difficulty parameter of the content and the capability parameter of the object are extracted.
Specifically, in the knowledge recommendation scene, the users have different capacities, the knowledge has different difficulties, the same user needs different reading times when grasping knowledge points with different difficulties, and the users with different capacities need different reading times when grasping knowledge points with the same difficulty. Based on IRT theory, mapping to knowledge scene, the reading times of the user to the knowledge can be collected as user feedback, and the user feedback is used as the reaction of the user to the knowledge. Based on the historical user response to knowledge, the probability of how much the user needs to master knowledge points under the given user capacity level and given knowledge difficulty coefficient can be obtained, and the probability can truly reflect the preference of the user to knowledge.
Step S5023, determining a probability density function according to the difficulty parameter of the content and the capability parameter of the object, wherein the probability density function is used for representing the probability that the object grasps the content.
In step S5025, the probability density function is determined as the feature parameter.
Based on IRT theory, probability modeling of potential features in a knowledge scene can be illustrated by using FIG. 6, and the horizontal axis is user feedback and the vertical axis is potential features in combination with FIG. 6. If the number of readings of a knowledge by a user is closer to μ (the average number of knowledge readings), it is relatively effective to say that the grasping probability is higher within σ variance (the average number of user readings). If μ±σ is exceeded, the user feedback that noise is present can be excluded because the user is unaware and cannot be used as effective feedback.
The scheme combines the feedback of the object, the capability parameter of the object and the difficulty parameter of the content, estimates the probability of grasping knowledge by the user according to the IRT model, and takes the probability as the potential characteristic of feedback behavior, so that the preference (grasping degree) of the object to the content is reflected; taking the recommended questions for students as examples, the feature is the probability of correct answer.
As an optional embodiment, step S5021, extracting a difficulty parameter of the content and a capability parameter of the object, includes:
In step S50211, the average value of the times of accessing one content by the plurality of objects is determined as the difficulty coefficient corresponding to the content.
In an alternative embodiment, assuming that U users read the ith knowledge point, each user reads Nu, the difficulty coefficient of the content may be determined using the following formula:
wherein μ represents the content difficulty Di, and is defined as the average reading times of the ith content, and the average reading times are within μ times, so that the user can grasp the ith knowledge; if the average reading times exceeds mu times, the user ability is weak; and if the average reading times are lower than mu times, the user is more powerful.
In step S50213, the average value of the number of times the object accesses the plurality of contents is determined as the capability coefficient of the object.
In an alternative embodiment, assuming that I knowledge is read by the u-th user, each of which is Ni, the capability coefficients of the object may be calculated using the following formula:
where σ represents user capability Au, defined as the average number of readings for the u-th user. The average reading times are within sigma times, and the knowledge point can be mastered by a u-th user; if the average reading times exceeds sigma times, the knowledge difficulty is higher; the average reading times are lower than sigma times, which indicates that the knowledge difficulty is lower.
As an alternative embodiment, determining the probability density function based on the difficulty parameter of the content and the capability parameter of the object includes: and determining a probability density function based on the Gaussian probability density function, wherein a scale function of the Gaussian probability density function is a capability parameter of the object, a random variable of the Gaussian probability density function is feedback information of the object on the content, and a position parameter of the Gaussian probability density function is a difficulty coefficient corresponding to the content.
It is known from IRT theory that the number of reads r in triples < u, i, r > representing user base data is a reaction, subject to gaussian distribution. A gaussian probability model can therefore be used in combination with the user's ability Au and knowledge difficulty Di to extract potential features. Even though a gaussian probability model is used as the IRT model for the knowledge scene, in an alternative embodiment, the probability density function described above may be determined by the following formula:
wherein r represents user feedback, i.e. the number of reads in the triplet; μ represents knowledge difficulty Di; sigma represents user capability Au; f (r; mu, sigma) is a potential feature, is a probability density, and reflects the probability of the user grasping knowledge under the comprehensive user capability, knowledge difficulty and user feedback.
In step S504, the information under feedback in the base data is replaced with the extracted feature parameters, i.e., the triplet < u, i, r > is corrected to the triplet < u, i, f (r; μ, σ >).
The above scheme builds a gaussian probability model based on the collected triples < u, i, r > to extract the feature parameters, i.e., probability density functions f (r; μ, σ). And replacing feedback information by using the characteristic parameters to construct new triples, namely < u, i, f (r; mu, sigma) >, so that noise possibly existing in user feedback is eliminated through the characteristic parameters, and the elimination process uniformly considers the user capability and content difficulty.
As an alternative embodiment, step S506, using the updated base data to construct an objective function for decomposing the original matrix, includes:
step S5061, obtaining an error function of feedback information of any one object to any one content, where the error function is a difference between actual feedback information of the object to the content and predicted feedback information of the object to the content, and the predicted feedback information is obtained by predicting the first sub-matrix and the second sub-matrix.
Specifically, the error function refers to an error between feedback information recorded in the basic data (corrected basic data) and feedback information predicted from the first and second submatrices, and the smaller the error, the more accurate the result of decomposition is described, and the process of decomposition is also a process of training with the objective function as a model.
In step S5063, a first sum of error functions for each object for each content is acquired.
After obtaining the error function of each object for each content, the error functions are summed to obtain the first sum. For a simple example, if there are three users U1, U2, and U3, and there are three contents A1, A2, and A3, the error function < U1, A1> of the feedback information when U1 reads A1, and the error function < U1, A2> … … of the feedback information when U1 reads A2 are obtained, respectively, until the error function < U3, A3> of the feedback information when U3 reads A3 is obtained. And adding the nine error functions to obtain the first sum.
In step S5065, a regularization term of the objective function is obtained, where the regularization term is formed by a product of a preset first regularization term parameter and a second regularization term parameter, and the second regularization term parameter is formed by a sum of a norm square of the first submatrix and a norm square of the second submatrix.
The first regularization term parameter may be a preset constant, and the second regularization term parameter is determined according to a norm square of the first submatrix and a norm square of the second submatrix.
In step S5067, the second sum of the first sum and the regularization term is determined as an objective function for decomposing the original matrix.
In an alternative embodiment, the original matrix R can be constructed based on the triples < u, i, f (R; μ, σ) >, assuming p u Is the ith row, q of matrix P i Is the ith row of matrix Q, the objective function for matrix decomposition is shown as follows:
where λ is a regularized term parameter. Continuous fitting by random gradient descentAnd f (R; mu, sigma) and minimizing the square difference of the two, so as to obtain the values of the two matrixes P and Q approaching R.
According to the scheme, the Gaussian probability model of the knowledge scene is built based on the IRT, the potential features are extracted from the user feedback information by combining the user capability and the project difficulty, the user feedback is replaced by the potential features based on the Gaussian probability model, the learning training of the matrix decomposition model is participated, and the purpose of eliminating noise in the user feedback is achieved. The knowledge recommendation scheme based on the IRT can be used for recommendation scenes with user capacity and project difficulty, such as k12 problem recommendation, medical examination knowledge recommendation, call center knowledge recommendation and the like, and has wider applicability.
Example 3
Two other ways of step S102 are described below in example 3. According to an embodiment of the present application, an embodiment of a content recommendation method is provided, and it should be noted that fig. 7 is a flowchart of another content recommendation method according to an embodiment of the present application, and as shown in fig. 7, the method includes the following steps:
Step S102, basic data of an object is acquired, wherein the basic data comprises: object identification, content identification and feedback information of the object to the content.
Step S702, extracting the characteristic parameters in the feedback information.
Step S704, constructing an original matrix by using the basic data.
Step S706, constructing an objective function for decomposing the original matrix based on the feature parameters and the updated basic data.
And S106, performing matrix decomposition on the original matrix based on an objective function to obtain a first submatrix and a second submatrix.
And step S108, determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object.
Steps S102, S106 and S108 in this embodiment are the same as steps S102, S106 and S108 in embodiment 1, respectively, and are not described here again.
A second way of modifying the feedback information is described first, as an optional embodiment, step S706, constructing an objective function for decomposing the original matrix based on the feature parameters and the updated basic data, including: obtaining an error function of feedback information of any object to any content, wherein the error function is the difference between characteristic parameters of the object to the content and predictive feedback information of the object to the content, and the predictive feedback information is obtained through prediction of a first submatrix and a second submatrix; obtaining a first sum of squares of error functions of each object for each content; acquiring a regular term of an objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix, a norm square of a second submatrix, a square of a difficulty coefficient of content and a square of a capability coefficient of an object; the first sum value and the second sum value of the regularization term are determined as an objective function for decomposing the original matrix.
In an alternative embodiment, triples based<u，i，r>An original matrix R can be constructed assuming p u Is the ith row, q of matrix P i Is the ith row of matrix Q, r ui Feedback information of the ith knowledge point of the ith user is represented; the u-th user capability coefficient a u The method comprises the steps of carrying out a first treatment on the surface of the Ith knowledge point difficulty coefficient d i The objective function of the matrix decomposition is as follows:
where λ is the first regularization term parameter described above. By random gradient descent, fit continuouslyAnd f (r) ui ；d i ,a u ) The square difference between the two is minimized, and the values of the two matrices P and Q approaching R can be obtained. The update formula of the parameters in random gradient descent is as follows:
q i ←q i +γ(e·p u -λq i )
p u ←p u +γ(e·q i -λp u )
wherein, the liquid crystal display device comprises a liquid crystal display device,indicating bias. Based on triplets<u，i，r>Training a model by using matrix decomposition technology to learn predictive feedback +.>Capability factor a for user u u And difficulty coefficient d of knowledge point i i 。
As an optional embodiment, step S702, extracting the characteristic parameters in the feedback information includes: extracting difficulty parameters of the content and capability parameters of the object; determining a probability density function according to the difficulty parameter of the content and the capability parameter of the object, wherein the probability density function is used for representing the probability of the object grasping the object; and determining the probability density function as a characteristic parameter.
The above steps are similar to steps S5021, S5023, S5025, and will not be repeated here.
A third embodiment of correcting the feedback information will be described. In this method, a method for decomposing an original matrix based on only corrected basic data is constructed, and the characteristic parameters and the objective functions in the following embodiments are different from those in the second method.
As an optional embodiment, step S706, constructing an objective function for decomposing the original matrix based on the feature parameters and the updated basic data, includes: obtaining an error function of feedback information of any object to any content, wherein a first difference value of the feedback information of the object to the content and the characteristic parameter is obtained, a second difference value of the first difference value and the prediction feedback information of the object to the content is obtained, the second difference value is determined to be the error function, and the prediction feedback information is obtained through prediction of a first submatrix and a second submatrix; obtaining a first sum of squares of error functions of each object for each content; acquiring a regular term of an objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix, a norm square of a second submatrix, a square of a difficulty coefficient of content and a square of a capability coefficient of an object; the first sum value and the second sum value of the regularization term are determined as an objective function for decomposing the original matrix.
In an alternative embodiment, triples based<u，i，r>An original matrix R can be constructed assuming p u Is the ith row, q of matrix P i Is the ith row of matrix Q, r ui Feedback information of the ith knowledge point of the ith user is represented; the u-th user capability coefficient a u The method comprises the steps of carrying out a first treatment on the surface of the Ith knowledge point difficulty coefficient d i The objective function of the matrix decomposition is as follows:
where λ is a regularized term parameter. Continuous fitting by random gradient descentAnd f (d) i ,a u ) The square difference between the two is minimized, and the values of the two matrices P and Q approaching R can be obtained. The update formula of the parameters in random gradient descent is as follows:
q i ←q i +γ(e·p u -λq i )
p u ←p u +γ(e·q i -λp u )
wherein, the liquid crystal display device comprises a liquid crystal display device,indicating bias. Based on triplets<u，i，r>Training a model by using matrix decomposition technology to learn predictive feedback +.>Capability factor a for user u u And difficulty coefficient d of knowledge point i i 。
As an alternative embodiment, extracting the characteristic parameters in the feedback information includes: extracting difficulty parameters of the content and capability parameters of the object; and determining characteristic parameters according to the difficulty parameters of the content and the capability parameters of the object based on the item reaction theoretical model.
As known from IRT theory, user feedback is a triplet<u，i，r>The reading times r in the system is a reaction which is determined by the user capacity coefficient and the knowledge point difficulty system. In an alternative embodiment, suppose the number of reads of the i knowledge points by the ith user is r u i The capability factor of the u-th user is a u The difficulty coefficient of the ith knowledge point is d i Can correct r ui IRT model of (c):
this correction value will be used in constructing the objective function, i.e. in decomposing the original matrix by training the model. According to the scheme, through constructing the IRT model, noise possibly existing in user feedback is removed, and the user capacity coefficient and the project difficulty coefficient are uniformly considered in the removing process.
Example 4
According to an embodiment of the present application, there is provided an embodiment of a content recommendation apparatus, and fig. 8 is a schematic diagram of a content recommendation apparatus according to an embodiment of the present application, as shown in fig. 8, including:
an obtaining module 80, configured to obtain basic data of an object, where the basic data includes: object identification, content identification and feedback information of the object to the content.
The correction module 82 is configured to update the base data by correcting the feedback information, and construct an original matrix and/or an objective function for decomposing the original matrix based on the updated base data.
The decomposition module 84 is configured to perform matrix decomposition on the original matrix based on the objective function, so as to obtain a first sub-matrix and a second sub-matrix.
The determining module 86 is configured to determine target content according to the first sub-matrix and the second sub-matrix, where the target content is a content to be recommended to the object.
As an alternative embodiment, the correction module includes: the first extraction submodule is used for extracting characteristic parameters in the feedback information; the first correction submodule is used for replacing feedback information in the basic data by using the characteristic parameters so as to correct the feedback information; and the construction submodule is used for updating the basic data by using the corrected feedback information, and constructing an original matrix and an objective function for decomposing the original matrix by using the updated basic data.
As an alternative embodiment, the first extraction submodule includes: the first extraction unit is used for extracting difficulty parameters of the content and capability parameters of the object; a first determining unit configured to determine a probability density function according to a difficulty parameter of the content and a capability parameter of the object, where the probability density function is used to represent a probability that the object grasps the content; and the second determining unit is used for determining the probability density function as a characteristic parameter.
As an alternative embodiment, the first extraction unit comprises: the first determining subunit is used for determining the average value of the times of the plurality of objects accessing the content as the difficulty coefficient corresponding to the content; and the second determination subunit is used for determining the average value of the times of the object accessing the plurality of contents as the capability coefficient of the object.
As an alternative embodiment, the first determining unit comprises: and the third determining subunit is used for determining a probability density function based on a Gaussian probability density function, wherein the scale function of the Gaussian probability density function is a capability parameter of the object, the random variable of the Gaussian probability density function is feedback information of the object on the content, and the position parameter of the Gaussian probability density function is a difficulty coefficient of the content.
As an alternative embodiment, the building sub-module comprises: the first acquisition unit is used for acquiring an error function of feedback information of any object to any content, wherein the error function is the difference between actual feedback information of the object to the content and predicted feedback information of the object to the content, and the predicted feedback information is obtained through prediction of a first submatrix and a second submatrix; a second acquisition unit configured to acquire a first sum value of error functions of each object for each content; the third acquisition unit is used for acquiring a regular term of the objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of the first submatrix and a norm square of the second submatrix; and the third determining unit is used for determining the second sum value of the first sum value and the regular term as an objective function for decomposing the original matrix.
As an alternative embodiment, the correction module includes: the second extraction submodule is used for extracting characteristic parameters in the feedback information; a first construction sub-module for constructing an original matrix using the base data; and the second construction submodule is used for constructing an objective function for decomposing the original matrix based on the characteristic parameters and the updated basic data.
As an alternative embodiment, the second building sub-module comprises: a fourth obtaining unit, configured to obtain an error function of feedback information of any one object to any one content, where the error function is a difference between a characteristic parameter of the object to the content and prediction feedback information of the object to the content, and the prediction feedback information is obtained by predicting the first submatrix and the second submatrix; a fifth acquisition unit configured to acquire a first sum of squares of error functions of each object for each content; a sixth obtaining unit, configured to obtain a regularization term of the objective function, where the regularization term is formed by a product of a preset first regularization term parameter and a second regularization term parameter, and the second regularization term parameter is formed by a sum of a norm square of the first submatrix, a norm square of the second submatrix, a square of a difficulty coefficient of the content, and a square of a capability coefficient of the object; and the fourth determining unit is used for determining the second sum value of the first sum value and the regular term as an objective function for decomposing the original matrix.
As an alternative embodiment, the second extraction sub-module comprises: the second extraction unit is used for extracting difficulty parameters of the content and capability parameters of the object; a fifth determining unit for determining a probability density function according to the difficulty parameter of the content and the ability parameter of the object, wherein the probability density function is used for representing the probability of the object grasping the object; and a sixth determining unit for determining the probability density function as the characteristic parameter.
As an alternative embodiment, the second building sub-module comprises: the third extraction unit is used for extracting characteristic parameters in the feedback information; a seventh obtaining unit, configured to obtain an error function of feedback information of any one object on any one content, where a first difference value between the feedback information of the object on the content and the characteristic parameter is obtained, a second difference value between the first difference value and the predicted feedback information is obtained, and the predicted feedback information is obtained through prediction of the first submatrix and the second submatrix; an eighth obtaining unit that obtains a first sum of squares of error functions of each object for each content; a seventh obtaining unit, configured to obtain a regularization term of the objective function, where the regularization term is formed by a product of a preset first regularization term parameter and a second regularization term parameter, and the second regularization term parameter is formed by a sum of a norm square of the first submatrix, a norm square of the second submatrix, a square of a difficulty coefficient of the content, and a square of a capability coefficient of the object; and a seventh determining unit, configured to determine the second sum value of the first sum value and the regularization term as an objective function for decomposing the original matrix.
As an alternative embodiment, the third extraction unit comprises: an extraction subunit, configured to extract a difficulty parameter of the content and an ability parameter of the object; and a fourth determining subunit, configured to determine the feature parameter according to the difficulty parameter of the content and the capability parameter of the object by using the item reaction theoretical model.
Example 5
According to an embodiment of the present application, there is provided a computer storage medium storing a plurality of instructions adapted to be loaded by a processor and to perform the method steps of any one of embodiments 1, 2 and 3.
Example 6
According to an embodiment of the present application, there is provided an intelligent interactive tablet, including: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to perform the method steps of any of embodiments 1, 2 and 3.
The foregoing embodiment numbers of the present application are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
In the foregoing embodiments of the present application, the descriptions of the embodiments are emphasized, and for a portion of this disclosure that is not described in detail in this embodiment, reference is made to the related descriptions of other embodiments.
In the several embodiments provided in the present application, it should be understood that the disclosed technology may be implemented in other manners. The above-described embodiments of the apparatus are merely exemplary, and the division of the units, for example, may be a logic function division, and may be implemented in another manner, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some interfaces, units or modules, or may be in electrical or other forms.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The foregoing is merely a preferred embodiment of the present application and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present application, which are intended to be comprehended within the scope of the present application.
Claims (11)
1. A content recommendation method, comprising:
obtaining basic data of an object, wherein the basic data comprises: object identification, content identification and feedback information of the object to the content;
updating the basic data by correcting the feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data;
performing matrix decomposition on the original matrix based on an objective function to obtain a first submatrix and a second submatrix;
determining target content according to the first submatrix and the second submatrix, wherein the target content is content to be recommended to the object;
the updating of the basic data through the correction feedback information, and the construction of an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data comprises the following steps: extracting a difficulty parameter of the content and a capability parameter of the object, wherein the average value of the times of the plurality of objects accessing the content is determined as a difficulty coefficient corresponding to the content, and the average value of the times of the plurality of objects accessing the content is determined as a capability coefficient of the object; determining a probability density function according to the difficulty parameter of the content and the capability parameter of the object, wherein the probability density function is used for representing the probability that the object grasps the content; determining the probability density function as a characteristic parameter; replacing feedback information in the basic data by using the characteristic parameters so as to correct the feedback information; and updating the basic data by using the corrected feedback information, and constructing an original matrix and an objective function for decomposing the original matrix by using the updated basic data.
2. The method of claim 1, wherein determining a probability density function based on the difficulty parameter of the content and the ability parameter of the object comprises:
and determining the probability density function based on a Gaussian probability density function, wherein a scale function of the Gaussian probability density function is a capability parameter of the object, a random variable of the Gaussian probability density function is feedback information of the object to the content, and a position parameter of the Gaussian probability density function is a difficulty coefficient of the content.
3. The method of claim 1, wherein constructing an objective function for decomposing the original matrix using the updated base data comprises:
obtaining an error function of feedback information of any object to any content, wherein the error function is the difference between actual feedback information of the object to the content and predictive feedback information of the object to the content, and the predictive feedback information is obtained through prediction of a first submatrix and a second submatrix;
acquiring a first sum of error functions of each object for each content;
acquiring a regular term of the objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix and a norm square of a second submatrix;
And determining the second sum value of the first sum value and the regular term as an objective function for decomposing the original matrix.
4. Method according to claim 1, characterized in that updating the base data by modifying the feedback information and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated base data comprises:
extracting characteristic parameters in the feedback information;
constructing an original matrix by using the basic data;
and constructing an objective function for decomposing the original matrix based on the characteristic parameters and the updated basic data.
5. The method of claim 4, wherein constructing an objective function for decomposing the original matrix based on the feature parameters and the updated base data comprises:
obtaining an error function of feedback information of any object to any content, wherein the error function is the difference between characteristic parameters of the object to the content and predictive feedback information of the object to the content, and the predictive feedback information is obtained through prediction of a first submatrix and a second submatrix;
Obtaining a first sum of squares of error functions of each object for each content;
acquiring a regular term of the objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix, a norm square of a second submatrix, a square of a difficulty coefficient of the content and a square of an ability coefficient of the object;
and determining the second sum value of the first sum value and the regular term as an objective function for decomposing the original matrix.
6. The method of claim 4, wherein extracting the characteristic parameters in the feedback information comprises:
extracting difficulty parameters of the content and capability parameters of the object;
determining a probability density function according to the difficulty parameter of the content and the capability parameter of the object, wherein the probability density function is used for representing the probability that the object grasps the object;
and determining the probability density function as the characteristic parameter.
7. The method of claim 4, wherein constructing an objective function for decomposing the original matrix based on the feature parameters and the updated base data comprises:
Obtaining an error function of feedback information of any object to any content, wherein a first difference value of the feedback information of the object to the content and the characteristic parameter is obtained, a second difference value of the first difference value and prediction feedback information is obtained, the second difference value is determined to be the error function, and the prediction feedback information is obtained through prediction of a first submatrix and a second submatrix;
obtaining a first sum of squares of error functions of each object for each content;
acquiring a regular term of the objective function, wherein the regular term consists of a product of a preset first regular term parameter and a second regular term parameter, and the second regular term parameter consists of a sum of a norm square of a first submatrix, a norm square of a second submatrix, a square of a difficulty coefficient of the content and a square of an ability coefficient of the object;
and determining the second sum value of the first sum value and the regular term as an objective function for decomposing the original matrix.
8. The method of claim 7, wherein extracting the characteristic parameters in the feedback information comprises:
extracting a difficulty parameter of the content and a capability parameter of the object;
And determining the characteristic parameters according to the difficulty parameters of the content and the capability parameters of the object based on a project reaction theoretical model.
9. A content recommendation apparatus, comprising:
the device comprises an acquisition module, a storage module and a storage module, wherein the acquisition module is used for acquiring basic data of an object, and the basic data comprises: object identification, content identification and feedback information of the object to the content;
the correction module is used for updating the basic data by correcting the feedback information, and constructing an original matrix and/or an objective function for decomposing the original matrix based on the updated basic data;
the decomposition module is used for carrying out matrix decomposition on the original matrix based on an objective function to obtain a first submatrix and a second submatrix;
the determining module is used for determining target content according to the first submatrix and the second submatrix, wherein the target content is the content to be recommended to the object;
the correction module is further configured to: extracting a difficulty parameter of the content and a capability parameter of the object, wherein the average value of the times of the plurality of objects accessing the content is determined as a difficulty coefficient corresponding to the content, and the average value of the times of the plurality of objects accessing the content is determined as a capability coefficient of the object; determining a probability density function according to the difficulty parameter of the content and the capability parameter of the object, wherein the probability density function is used for representing the probability that the object grasps the content; determining the probability density function as a characteristic parameter; replacing feedback information in the basic data by using the characteristic parameters so as to correct the feedback information; and updating the basic data by using the corrected feedback information, and constructing an original matrix and an objective function for decomposing the original matrix by using the updated basic data.
10. A computer storage medium storing a plurality of instructions adapted to be loaded by a processor and to perform the method steps of any of claims 1 to 8.
11. An intelligent interactive tablet, comprising: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to perform the method steps of any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911061257.7A CN110851734B (en) | 2019-11-01 | 2019-11-01 | Content recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911061257.7A CN110851734B (en) | 2019-11-01 | 2019-11-01 | Content recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110851734A CN110851734A (en) | 2020-02-28 |
| CN110851734B true CN110851734B (en) | 2023-09-05 |
Family
ID=69599490
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911061257.7A Active CN110851734B (en) | 2019-11-01 | 2019-11-01 | Content recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110851734B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102135989A (en) * | 2011-03-09 | 2011-07-27 | 北京航空航天大学 | Normalized matrix-factorization-based incremental collaborative filtering recommending method |
| CN107562875A (en) * | 2017-08-31 | 2018-01-09 | 北京麒麟合盛网络技术有限公司 | A kind of update method of model, apparatus and system |
| WO2018079968A1 (en) * | 2016-10-26 | 2018-05-03 | 주식회사 뤼이드 | Method and apparatus for providing personalized educational content, and computer program |
| CN108509573A (en) * | 2018-03-27 | 2018-09-07 | 陕西科技大学 | Book recommendation method based on matrix decomposition collaborative filtering and system |
| CN110378818A (en) * | 2019-07-22 | 2019-10-25 | 广西大学 | Personalized exercise recommended method, system and medium based on difficulty |
-
2019
- 2019-11-01 CN CN201911061257.7A patent/CN110851734B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102135989A (en) * | 2011-03-09 | 2011-07-27 | 北京航空航天大学 | Normalized matrix-factorization-based incremental collaborative filtering recommending method |
| WO2018079968A1 (en) * | 2016-10-26 | 2018-05-03 | 주식회사 뤼이드 | Method and apparatus for providing personalized educational content, and computer program |
| CN107562875A (en) * | 2017-08-31 | 2018-01-09 | 北京麒麟合盛网络技术有限公司 | A kind of update method of model, apparatus and system |
| CN108509573A (en) * | 2018-03-27 | 2018-09-07 | 陕西科技大学 | Book recommendation method based on matrix decomposition collaborative filtering and system |
| CN110378818A (en) * | 2019-07-22 | 2019-10-25 | 广西大学 | Personalized exercise recommended method, system and medium based on difficulty |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110851734A (en) | 2020-02-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20150278350A1 (en) | Recommendation System With Dual Collaborative Filter Usage Matrix | |
| CN110222233B (en) | Video recommendation method and device, server and storage medium | |
| US20130035912A1 (en) | Systems and methods of processing personality information | |
| US10210627B1 (en) | Image processing system for determining metrics of objects represented therein | |
| US20190065550A1 (en) | Query optimizer for combined structured and unstructured data records | |
| Sachse et al. | When nonresponse mechanisms change: Effects on trends and group comparisons in international large-scale assessments | |
| Lian et al. | Mutual reinforcement of academic performance prediction and library book recommendation | |
| CN111931043B (en) | Recommending method and system for science and technology resources | |
| Paek et al. | A comparison of three IRT approaches to examinee ability change modeling in a single-group anchor test design | |
| Ma et al. | Category-level model selection for the sequential G-DINA model | |
| CN115525838A (en) | Health knowledge recommendation method and system based on big data user portrait | |
| Kang et al. | Online calibration of a joint model of item responses and response times in computerized adaptive testing | |
| JP5356286B2 (en) | Question recommendation device, method and program | |
| Nguyen et al. | Understanding the dynamics of inter-provincial migration in the Mekong Delta, Vietnam: an agent-based modeling study | |
| CN107423740A (en) | The acquisition methods and device of salient region of image | |
| US20040193560A1 (en) | System and method for case-based reasoning | |
| CN110851734B (en) | Content recommendation method and device | |
| Zeller et al. | Do adaptive representations of the item-position effect in APM improve model fit? a simulation study | |
| JP5373670B2 (en) | Question recommendation device, method and program | |
| CN108960954B (en) | Content recommendation method and system based on user group behavior feedback | |
| CN110399279A (en) | A kind of intelligent measure for inhuman intelligent body | |
| CN107038169B (en) | Object recommendation method and object recommendation device | |
| KR20200036168A (en) | System for matching crops to individual by customizing individual body rhythm | |
| CN112365302B (en) | Product recommendation network training method, device, equipment and medium | |
| CN114820121A (en) | Insurance business pushing system and method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |