CN115809376A - Intelligent recommendation method based on big teaching data - Google Patents
Intelligent recommendation method based on big teaching data Download PDFInfo
- Publication number
- CN115809376A CN115809376A CN202211472211.6A CN202211472211A CN115809376A CN 115809376 A CN115809376 A CN 115809376A CN 202211472211 A CN202211472211 A CN 202211472211A CN 115809376 A CN115809376 A CN 115809376A
- Authority
- CN
- China
- Prior art keywords
- user
- data
- users
- similarity
- representing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an intelligent recommendation method based on big teaching data, which comprises the steps of obtaining historical record data searched by a user and transmitting the historical record data to a candidate queue, setting historical target information searched by the user into filtered candidate resources, obtaining data information according to the candidate resources to carry out screening and prediction, converting and transmitting original network data to model input data required for constructing a network model to obtain user preference behavior data, calculating item scores corresponding to the user preference behavior data and calculating similarity between users or similarity between items, establishing a learner model according to a course category corresponding to an item interested by the user and carrying out intelligent recommendation on the user, extracting three types of data of browsing, commenting and downloading from a plurality of preference behaviors of the user to the resources and carrying out statistics on the data, simultaneously realizing the quantitative formation of a user preference behavior data model, realizing user resource recommendation service, obtaining the interested resources and taking all the user preference behavior data into account, so that the recommendation quality is more accurate.
Description
Technical Field
The invention belongs to the technical field of teaching recommendation, and particularly relates to an intelligent recommendation method based on teaching big data.
Background
At present, with the rapid development of network technology, the current society has advanced into the intelligent information era, and the opening of network platform resource information enables all industry fields to share information, so that the demand of people on network course resources is more urgent. The increase of online teaching resources causes the 'information overload' of the teaching resources, generally, recommendation algorithms include collaborative filtering recommendation algorithms, mixed recommendation algorithms, content-based recommendation algorithms and the like, and the traditional collaborative filtering algorithms have the problems of single calculation characteristics and large calculation amount. In order to enable a user to recommend resources with load individualized requirements, the problem of overload of information of the user and teaching resources is solved, but the individualized resource recommendation establishes the relation between the user and the resources, but the content of information data received by the user is seriously unbalanced, so that the matching degree of the user and the teaching resource information and the user experience degree are reduced.
Disclosure of Invention
In view of the above, the present invention provides an intelligent recommendation method based on big teaching data, which improves the precision and quality of recommending user teaching resources and better meets the requirement of user personalized recommendation service, so as to solve the above technical problems.
The invention provides an intelligent recommendation method based on big teaching data, which comprises the following steps:
acquiring historical record data searched by a user and transmitting the historical record data to a candidate queue, and setting historical target information searched by the user into filtered candidate resources;
screening and predicting according to data information obtained by the candidate resources, converting and transmitting original network data to model input data required by building a network model to obtain user preference behavior data, wherein the user preference behavior data comprises browsing times, comment times and download times of learning resources, and the preset browsing times areWhereinRepresenting the browsing times of the user m to the resource n, and presetting the comment times asIn whichThe number of comments of the user m on the resource n is represented, and the preset download number isWhereinRepresenting the number of times of downloading the resource n by the user m;
calculating item scores corresponding to the user preference behavior data, calculating inter-user similarity or inter-item similarity according to the item scores, and acquiring items interested by the users based on the inter-user similarity or the inter-item similarity;
and establishing a learner model according to the course categories corresponding to the items in which the user is interested and intelligently recommending the user.
As a further improvement of the above technical solution, establishing a learner model according to a course category corresponding to an item in which a user is interested and performing intelligent recommendation on the user includes:
the process of automatically extracting the user interest by using the neural network relieves cold start, the class of the course liked by the user is obtained according to the user interest, and the learning ability similarity between the users is calculated to obtain similar users and the similar users are recommended;
determining the user category through user historical information, introducing learner behavior characteristics and learning ability similarity in-class calculation to obtain a model similarity score, calculating in-class user score similarity, and fusing the calculated model similarity score and the score similarity score to obtain a final user set for recommendation.
As a further improvement of the above technical solution, a process of automatically extracting user interest using a neural network to alleviate cold start includes:
classifying new users by LSTM text classification method, segmenting sentences into words and vectorizing, inputting word vectors, and outputting vectors at n moments through LSTM layer processing,,...The LSTM model comprises a first layer, namely an embedding layer, which is used for completing word segmentation and vectorization operations, a second layer, namely a spatial _ drop 1d layer, which is used for improving independence between features, a third layer, namely an LSTM layer, a fourth layer, which is used for obtaining the most significant features by using Max mapping, and a fifth layer, which is used for classifying by using SoftMax.
As a further improvement of the technical scheme, the method for calculating the learning capacity similarity among the users to obtain similar users and recommending the similar users comprises the following steps:
presetting c as user preference and mapping to resource categoryMapping the interest categories of the users according to the class categories,represents the learning ability level of the learner, whereinCorresponding to the primary level, the middle level and the high level respectively, the similarity calculation in the new user class uses standard Gaussian distribution, and the expression isWhere U represents the total number of users,indicating a user of the jth category c,which is indicative of a target user of the mobile communication system,、respectively represent the learning levels of the two users,the larger the value of (A) is, the greater the similarity between two users is, and conversely, the smaller the similarity is.
As a further improvement of the technical scheme, the historical learning information of the user is counted to obtain the interest label of the user, and the expression isWhereinRepresenting a userThe number of times the historical information learned class i lessons,a process for representing a total number of learning of a user, for performing a calculation in a user interest category based on extracted user interest tags, comprising:
calculating the similarity of learning characteristics according to the learning characteristics of old users, calculating the behavior characteristics of the same course by using two users, expressing the behaviors of the courses by using vectors, wherein the expression isWhere U represents the total number of users,which is indicative of a target user of the mobile communication system,indicating other users in the same category of the user,,,representing a userAs to the behavioral characteristics of the lesson K,representing a userFor the behavior characteristics of course K, K represents the course that two users have learned together,respectively representing the time of the learner in learning the course, the acquired experience, the acquired score, the learning progress and the job submitting times;
the expression of similarity calculation of common courses of two users isWhere k represents a course of co-learning between two users,n represents the total number of courses that two users have learned together;
combining the learning level similarity expression and the similarity expression of the common courses of the two users to obtain the similarity of the learner model, wherein the expression is。
As a further improvement of the above technical solution, the method for calculating a category of interest of a target user in history information of the target user according to an expression of an interest tag of the user, and finding other users having the same interest as the target user to obtain a user score includes:
according to the expression asCalculating the score similarity between users, wherein the user set isCollection of course resources,Representing a userAndi represents a common scored course,andrespectively represent、The mean value among the common score items,、respectively represent usersAndthe step of course scoring, the learner model similarity and the scoring similarity are added to obtain an expression asSorting the calculation results from high to low, and taking the first N as a neighbor set;
calculating to obtain direct neighbors through the scores, predicting the scores of the courses which do not generate scores for the target user in the final neighbor users, selecting the first N courses with the highest predicted scores for recommendation, and calculating the expression asIn whichRepresenting a userThe predictive score for the non-scored item i,representing a userThe average score of (a) is calculated,representing a userThe average score of (a) is calculated,representThe score for the item i is given to,representing a userIs selected.
The technical scheme is further improved, the user preference behavior data resource score is calculated, the flow times, the comment times and the download times have availability, a normalization method is used for carrying out data standardization processing, and the expression isWhereinIndicates the score, X indicates the number of times,the minimum value of the number of times is represented,maximum value of the representation times, uniformly mapped to;
Further obtaining an expression asWhereinRepresenting a userFor resourcesThe number of times of browsing of (a) is scored,representing a userTo resourcesThe number of reviews scored in (a) is,representing a userTo resourcesScoring the download times to obtain the userFor resourcesPreliminary user preference behavior data scoring of,According to the number of comments, the number of comments and the downloadDifferent weights are given to different degrees of contribution of the times to the scores, and the final user preference behavior data score is obtained by combining the user preference behavior data weights, wherein the expression isWherein,、Andrespectively a browsing frequency weight, a comment frequency weight and a download frequency weight,representing a userFor resourcesThe end user preferred behavior data score.
As a further improvement of the above technical solution, the screening and predicting according to the data information obtained from the candidate resource includes:
screening and predicting data information obtained by a network structure, converting and transmitting original network data to obtain model input data required by constructing a network model, wherein the expression isWhereinRepresenting a user item matrix;
if it isThen, the user's interest in the project is obtained, the information obtained through calculation is higher than the resource information score ignored by the user, and a relation matrix of the user and the project is obtained according to the difference between the high degree of interest and the lack of interest;
when the user pays attention to the resource information, the user pays attention to the history of the user in searching the page, when the user pays attention to the network page, the ID information of the network user is recorded, and if the user ID information isDetermining user search behavior characteristics by browsing records by userRepresenting, obtaining a user information interaction record according to the relation matrix of the user and the item and recording the record as,Obtaining behavior height by correlating with user search behavior characteristicsThe expression isWhereinIndicates the number of platform users,and representing the quantity of the historical records, and obtaining the length of the queue by obtaining the action height through the column splitting and the set operation and coinciding with the basic information of the user.
AsThe technical scheme is further improved, the history records searched by the user are taken as the characteristic points of the user, classification is carried out by adopting SoftMax, and if the unknown user is u, a function is constructedWherein P represents the probability of the ith class, v represents the feature quantification, w represents the dimension coefficient, and x represents the edge node distribution according to a functionAnd classifying the item queue to obtain a candidate queue.
As a further improvement of the above technical solution, acquiring history data searched by a user and transmitting the history data to a candidate queue includes:
assuming that N is the number of files, different files contain key information, and if the number of key information in a file indicatesThe number of times that the key information appears in the file isWhereinRepresenting the most times of key information, obtaining the key information by adopting an exhaustive algorithm, and selecting the maximum frequency of occurrence of words;
By usingA parameter representing key information, expressed asWherein N represents a number of the groups,the key words are highlighted to obtain the resource information which is most interesting to the user.
The invention provides an intelligent recommendation method based on big teaching data, which comprises the steps of obtaining historical record data searched by a user and transmitting the historical record data to a candidate queue, setting historical target information searched by the user into filtered candidate resources, screening and predicting according to data information obtained by the candidate resources, converting and transmitting original network data to model input data required by constructing a network model to obtain user preference behavior data, calculating item scores corresponding to the user preference behavior data, calculating similarity between users or similarity between items according to the item scores, obtaining items interested by the user based on the similarity between the users or the similarity between the items, establishing a learner model according to course categories corresponding to the items interested by the user and intelligently recommending the user, pre-classifying the user through a neural network, calculating and recommending in the categories, relieving the problem of cold start, calculating multiple characteristics in old users and solving the defect of calculation of single characteristics of collaborative filtering scores. Three types of data of browsing, commenting and downloading are extracted from a plurality of preference behaviors of a user on resources, the frequency data of the three types of data are counted, quantification is realized at the same time, a user preference behavior data model is formed, user resource recommendation service is realized through the model, interested resources are obtained, all user preference behavior data are taken into consideration, and therefore a more mature and complete personalized recommendation model is built, and recommendation quality is more accurate.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of an intelligent recommendation method based on big teaching data provided by the invention;
fig. 2 is a diagram of an intelligent recommendation process of a learning model provided by the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.
Referring to fig. 1, the invention provides an intelligent recommendation method based on big teaching data, which comprises the following steps:
s10: acquiring historical record data searched by a user and transmitting the historical record data to a candidate queue, and setting historical target information searched by the user into filtered candidate resources;
s11: screening and predicting according to data information obtained by the candidate resources, converting and transmitting original network data to model input data required by building a network model to obtain user preference behavior data, wherein the user preference behavior data comprises browsing times, comment times and download times of learning resources, and the preset browsing times areIn whichRepresenting the browsing times of the user m to the resource n, and presetting the comment times asIn whichThe number of comments of the user m on the resource n is represented, and the preset download number isIn whichRepresenting the number of times of downloading the resource n by the user m;
s12: calculating item scores corresponding to the user preference behavior data, calculating inter-user similarity or inter-item similarity according to the item scores, and acquiring items interested by the users based on the inter-user similarity or the inter-item similarity;
s13: and establishing a learner model according to the course categories corresponding to the items in which the user is interested and intelligently recommending the user.
In the embodiment, the explicit preference behavior refers to some personal basic information such as name, gender, profession, occupation, hobby, interest and the like which are filled by a user when the user registers on the network platform, the implicit preference behavior refers to implicit behavior which is left by the user on the network platform for required resources such as user browsing, browsing time, user comment, downloading and the like, the explicit preference behavior is usually changed by the user deliberately and is not real-time, the user's preference behavior cannot be reflected comprehensively and objectively, the implicit preference behavior is usually not very stable and is changed along with the change of the user's hobby, and the explicit preference behavior has the characteristics of real-time performance, objectivity and the like, and can comprehensively present the user's interest. When the temperature is higher than the set temperatureAt this time, although the user of the network platform has the search record, the user does not necessarily favor the network resource, and conversely,in the process, the user and the platform resource do not intersect, the record may not be known to exist, the search is not carried out, the interest is not lost, a lot of network information is ignored by most users, the interest heat is not formed, even if some items are poor in quality and unstable in signal, the users can dig out, and the rest data disappear due to the fact that the search quantity is lacked, and the difficulty in developing the network resource is increased due to the fact that the situation causes difficulty in developing the network resource.
It should be noted that, in order to solve the problems of excessive data loss and sparse network caused by the concern of the user, a candidate network needs to be established, the accuracy of information analysis is improved, and the browsing rate of a network platform is rapidly increased. And taking SoftMax as a classifier, taking the ID information of the user as label, proposing a user behavior characteristic as SoftMax, and according to the collaborative filtering result, different network users may have similar characteristics. Filtering and screening the adjacent information queues, wherein the items in the adjacent spaces are extremely similar although users are different, so that if a required queue is obtained, only the queues on the two adjacent sides of the required queue need to be analyzed and filtered to obtain the optimal selection, the same historical records and item queues need to be searched, the items corresponding to the queues are selected in a map, finally, the adjacent item queues are found, the items are processed, and the items with higher user concentration degree are selected. The method comprises the steps of constructing a network resource information acquisition model to obtain course resource information, combining all the obtained information to form a project information queue with high attention, and finally classifying the frequently browsed information by using a SoftMax classifier to obtain an individualized algorithm design, wherein the design is simple and accurate in structure, high in safety and practicability and beyond the expected effect. The network course resources are running at the temperature and show an upward trend, personalized development is achieved, webpage browsing frequency is increased, and precision and user experience of intelligent recommendation of big teaching data are promoted.
Optionally, establishing a learner model according to the course categories corresponding to the items in which the user is interested, and performing intelligent recommendation on the user, includes:
s20: the process of automatically extracting the user interest by using the neural network relieves cold start, the class of the course liked by the user is obtained according to the user interest, and the learning ability similarity between the users is calculated to obtain similar users and the similar users are recommended;
s21: determining the user category through user historical information, introducing learner behavior characteristics and learning ability similarity in-class calculation to obtain a model similarity score, calculating in-class user score similarity, and fusing the calculated model similarity score and the score similarity score to obtain a final user set for recommendation.
In this embodiment, the screening and predicting according to the data information obtained from the candidate resource includes: screening and predicting data information obtained by a network structure, converting and transmitting original network data to obtain model input data required by constructing a network model, and expressing the model input data asWhereinRepresenting a user item matrix; if it isWhen the user is interested in the project, the calculated information is higher than the resource information score ignored by the user, and a relation matrix of the user and the project is obtained according to the difference between the high attention degree and the non-attention degree; the user pays attention to the history of the user searching the page when paying attention to the resource information, when the user pays attention to the network page, the ID information of the network user is recorded, if the user ID information isDetermining user search behavior characteristics by browsing records by userTo express, according to the relation matrix of the user and the project, obtaining the information interaction record of the user and recording the information interaction record as,Obtaining behaviors by correlating with user search behavior characteristicsHeightThe expression isWhereinIndicates the number of platform users,and representing the quantity of the historical records, and obtaining the length of the queue by obtaining the action height through column separation and set operation and coinciding with the basic information of the user.
It should be noted that, the network platform transmits the history data searched by a specific user to the candidate queue, sets the history target information searched by the user as the filtered candidate resource, analyzes and displays the obtained data, the praise and the badness of many users to the resource queue reflect the behavior of the user in the network, the expressed information can describe the personal habit, the quality and the hobby of the user, and the like, and the resource information given by the user can obtain the characteristic expression of one person from the behavior.
Optionally, the process of automatically extracting user interest using a neural network mitigates cold starts, comprising:
classifying new users by LSTM text classification method, segmenting sentences into words and vectorizing, inputting word vectors, and outputting vectors at n moments through LSTM layer processing,,...Inputting the vector output by the upper layer to carry out maximum pooling operation to obtain a characteristic vector h, and processing the characteristic vector h of the upper layer by a SoftMax layerAnd finishing information classification, wherein the LSTM model comprises a first layer, namely an embedding layer, finishing word segmentation and vectorization operations, a second layer, namely a spatial _ drop 1d layer, improving the independence between the features, a third layer, namely an LSTM layer, a fourth layer for acquiring the most significant features by using MaxPholing, and a fifth layer for classifying by using SoftMax.
In this embodiment, calculating the learning ability similarity between users to obtain similar users and recommending the similar users includes: presetting c as user preference and mapping to resource categoryMapping the interest categories of the users according to the class categories,represents a learning ability level of a learner, whereinCorresponding to the primary level, the middle level and the high level respectively, the similarity calculation in the new user class uses standard Gaussian distribution, and the expression isWhere U represents the total number of users,indicating a user of the jth category c,a representation of the target user is provided,、respectively representing the learning levels of the two users,the larger the value of (A) is, the more two are usedThe greater the similarity of the users, and conversely the smaller the similarity.
It should be noted that the LSTM model is a new Network model for improving a Recurrent Neural Network (RNN), and the LSTM plays an expanding role in the RNN. The LSTM realizes information screening through a gate control structure, an input gate determines how many effective input information are reserved, an output gate determines how many effective unit states are output at the current moment, and a forgetting gate determines whether information output in the last state is reserved. The collaborative filtering is based on the user or project, the similarity between users or projects is calculated, the top N are removed from the top N as neighbors according to the similarity from high to low, the collaborative filtering algorithm only considers the scoring characteristics of the users to the projects in application, does not consider the inherent characteristics and behaviors of the users, ignores the objective information of the users, and correspondingly increases the calculation complexity along with the increase of the number of the users. The teaching resource recommendation algorithm uses a collaborative filtering algorithm, and needs to calculate characteristics such as user score similarity and learning level of a target user so as to improve intelligent recommendation efficiency.
Optionally, the historical learning information of the user is counted to obtain an interest tag of the user, and the expression isWhereinRepresenting a userThe historical information learns the number of times class i lessons,a process for representing a total number of learning of a user, for performing a calculation in a user interest category based on extracted user interest tags, comprising:
calculating the similarity of the learning characteristics according to the learning characteristics of the old users, and calculating the behavior characteristics of the same course by using two usersThe behavior is expressed by using vector, and the expression isWhere U represents the total number of users,a representation of the target user is provided,indicating other users in the same category of the user,,,representing a userFor the behavioral characteristics of the lesson K,representing a userFor the behavior characteristics of course K, K represents the course that two users have learned together,respectively representing the time of the learner in learning the course, the acquired experience, the acquired score, the learning progress and the job submitting times;
the expression of similarity calculation of common courses of two users isWhere k represents a course of co-learning between two users,n represents the total number of courses that two users have learned together;
combining the learning level similarity expression and the similarity expression of the common courses of the two users to obtain the similarity of the learner model, wherein the expression is。
In this embodiment, calculating a category of interest of a target user in the history information of the target user according to an expression of the user interest tag, and finding other users having the same interest as the target user to obtain a user score includes: according to the expression asCalculating the score similarity between users, wherein the user set isCollection of course resources,Representing a userAndi represents a common scored course,andrespectively represent、The average value among the common score items,、respectively represent usersAndthe step of course scoring, the learner model similarity and the scoring similarity are added to obtain an expression asSorting the calculation results from high to low, and taking the first N as a neighbor set; obtaining direct neighbors through the scoring calculation, predicting the scores of the courses which do not generate scores for the target user in the final neighbor users, selecting the first N courses with the highest predicted scores for recommendation, and obtaining the calculation expression ofWhereinRepresenting a userThe predictive score for the non-scored item i,representing a userThe average score of (a) is calculated,representing a userThe average score of (a) is calculated,representThe score for the item i is given to,representing a userIs selected.
It should be noted that the basic information of the user reflects little user interest, and if the user is briefly described for data mining and analysis, the user interest can be acquired more accurately, so that the problem of classifying learners manually can be solved, and the problem of cold start of new users can be alleviated. The new user can relieve cold start by using a method for automatically extracting the user interest by using a neural network, obtains the class of the course liked by the user through the user interest, and then calculates the learning ability similarity between the users in the class to obtain the similar user for recommendation. The old user can determine the user category through user history information, the learner learning behavior characteristics and learning ability similarity are introduced into the class, the model monarch score is obtained through calculation, on the basis, the in-class user score similarity is calculated, and the model monarch score are calculated in a fusion mode to obtain a final user set for recommendation. Classifying new users by adopting an LSTM text classification method, collecting a teaching resource data set of a certain teaching website, including personal registration information of 1000 users and course resources of 10 categories, preprocessing the data, removing stop words, segmenting words, completing data vectorization, vectorizing category labels, and dividing training and testing data sets to obtain LSTM model performance analysis results.
Optionally, calculating a user preferred behavior numberAccording to resource grading, the flow times, the comment times and the download times have usability, a normalization method is used for carrying out data standardization processing, and the expression isWhereinIndicates the score, X indicates the number of times,the minimum value of the number of times is represented,represents the maximum value of the times and is uniformly mapped to;
Further obtaining an expression asWhereinRepresenting a userFor resourcesThe number of times of browsing of (a) is scored,representing a userTo resourcesThe number of reviews scored in (a) is,representing a userFor resourcesScoring the download times to obtain the userFor resourcesPreliminary user preferred behavior data scoring,According to different contribution degrees of the comment times, the comment times and the download times to the scores, different weights are given, and the user preference behavior data weight is combined to obtain the final user preference behavior data score, wherein the expression isWherein,、Andrespectively a browsing frequency weight, a comment frequency weight and a downloading frequency weight,representing a userFor resourcesThe end user preferred behavior data score.
It should be noted that the method for evaluating the performance of the recommendation system mainly comprises an average absolute deviation MAE, a root mean square error RMSE, an accuracy Precision and a Recall rate Recall, which have different emphasis points and are applied to different scenes, wherein the MAE and the RMSE are mainly applied to a rating prediction recommendation scene, the Precision and the Recall are mainly applied to a TOP-N recommendation scene, and parametersSelecting average absolute deviation MAE as an evaluation index, feeding a final recommendation result back to a user by using a TOP-N list, selecting accuracy and recall as evaluation preparation, wherein the calculation expression isWherein N represents a set of resources,representing the absolute error between the prediction score and the true score,the presentation recommendation algorithm predicts the resource set of user u,representing the set of resources that user u is interested in the test set. Parameter(s)For adjusting the specific gravity of the result of the calculation of the degree of similarity, e.g.Indicating that no consideration is given to user preference behavior data scoring, i.e. also traditional collaborative filtering methods,indicating that the user's rating of the resource is not considered.And for marketing of recommendation precision under different value-taking conditions, the average absolute deviation MAE is used as an evaluation standard, and the smaller the value of the MAE algorithm is, the higher the recommendation quality is. Precision following for MAE algorithmThe change of the value is changed, the effect is not optimal when the value is too large or too small, the value change amplitude of the MAE algorithm is small along with the continuous increase of the data set and gradually tends to be stable, and when the value change is too large or too small, the effect is not optimal, and the value change amplitude of the MAE algorithm is small and gradually tends to be stableThe MAE value is lowest, and the model method recommendation precision is highest.
Optionally, the screening and predicting according to the data information obtained from the candidate resource includes:
screening and predicting data information obtained by a network structure, converting and transmitting original network data to obtain model input data required by constructing a network model, and expressing the model input data asWhereinRepresenting a user item matrix;
if it isThen, the user's interest in the item is obtained, the information obtained through calculation is higher than the resource information score ignored by the user, and the user is interested in the item according to the difference between high interest degree and no interest degreeObtaining a relation matrix of the user and the project;
the user pays attention to the history of the user searching the page when paying attention to the resource information, when the user pays attention to the network page, the ID information of the network user is recorded, if the user ID information isDetermining user search behavior characteristics by browsing records by userRepresenting, obtaining a user information interaction record according to the relation matrix of the user and the item and recording the record as,Obtaining behavior height by correlating with user search behavior characteristicsThe expression isWhereinIndicates the number of platform users,and representing the quantity of the historical records, and obtaining the length of the queue by obtaining the action height through the column splitting and the set operation and coinciding with the basic information of the user.
In the embodiment, the history records searched by the user are used as the characteristic points of the user and classified by adopting SoftMax, and if the unknown user is u, a function is constructedWhere P denotes the probability of class i and v denotes the featureQuantification, w representing the dimensional coefficient, x representing the edge node distribution, according to a functionAnd classifying the item queue to obtain a candidate queue. Acquiring historical record data searched by a user and transmitting the historical record data to a candidate queue, wherein the method comprises the following steps: assuming that N is the number of files, different files contain key information, and if the number of key information in a file indicatesThe number of times that the key information appears in the file isIn whichRepresenting the most times of key information, obtaining the key information by adopting an exhaustive algorithm, and selecting the maximum frequency of occurrence of words(ii) a By usingA parameter representing key information, expressed asWherein N represents a number of,and representing key words with emphasis so as to obtain the resource information which is most interested in by the user.
It should be noted that after how to express key word emphasis is obtained, resource information in which a user is most interested can be easily obtained to recommend the user to obtain a desired product, if the user does not obtain the same product, the user can be advised to select a special product similar to the product, and product information with different characteristics but the same model structure can be recommended to the user. The neural network is used for classifying users in advance, and then calculation and recommendation are performed in the class, so that the cold start problem is relieved, the defects of traditional collaborative filtering single scoring feature calculation are overcome for multi-feature calculation in the old user class, the method can be better suitable for scenes with large data volume, and the learner model has certain practical significance in teaching resource recommendation in a big data environment.
In all examples shown and described herein, any particular value should be construed as merely exemplary, and not as a limitation, and thus other examples of example embodiments may have different values.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined or explained in subsequent figures.
The above examples are merely illustrative of several embodiments of the present invention, and the description thereof is more specific and detailed, but not to be construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.
Claims (10)
1. An intelligent recommendation method based on big teaching data is characterized by comprising the following steps:
acquiring historical record data searched by a user and transmitting the historical record data to a candidate queue, and setting historical target information searched by the user into filtered candidate resources;
screening and predicting according to data information obtained by the candidate resources, converting and transmitting original network data to model input data required by building a network model to obtain user preference behavior data, wherein the user preference behavior data comprises browsing times, comment times and download times of learning resources, and the preset browsing times areIn whichRepresenting the browsing times of the user m to the resource n, and presetting the comment times asIn whichThe number of times of commenting on the resource n by the user m is represented, and the preset downloading number of times isWhereinRepresenting the number of times of downloading the resource n by the user m;
calculating item scores corresponding to the user preference behavior data, calculating inter-user similarity or inter-item similarity according to the item scores, and acquiring items interested by the users based on the inter-user similarity or the inter-item similarity;
and establishing a learner model according to the course categories corresponding to the items in which the user is interested and intelligently recommending the user.
2. The intelligent recommendation method based on big teaching data as claimed in claim 1, wherein building a learner model according to the course category corresponding to the item of interest of the user and making intelligent recommendation to the user comprises:
the process of automatically extracting the user interest by using the neural network relieves cold start, the class of the course liked by the user is obtained according to the user interest, and the learning ability similarity between the users is calculated to obtain similar users and the similar users are recommended;
determining the user category through user historical information, introducing learner behavior characteristics and learning ability similarity in the category to obtain a model similarity score and calculate user score similarity in the category, and combining the model similarity score and the score similarity score to obtain a final user set for recommendation.
3. The intelligent recommendation method based on big teaching data as claimed in claim 2, wherein the process of automatically extracting user interest using neural network alleviates cold start, comprising:
classifying new users by LSTM text classification method, segmenting sentences into words and vectorizing, inputting word vectors, and outputting vectors at n moments through LSTM layer processing,,...The LSTM model comprises a first layer, namely an embedding layer, which is used for completing word segmentation and vectorization operations, a second layer, namely a spatial _ drop 1d layer, which is used for improving independence between features, a third layer, namely an LSTM layer, a fourth layer, which is used for obtaining the most significant features by using Max mapping, and a fifth layer, which is used for classifying by using SoftMax.
4. The intelligent recommendation method based on big teaching data as claimed in claim 2, wherein calculating learning ability similarity among users to obtain similar users and recommending the users comprises:
presetting c as user preference and mapping as resource categoryMapping the interest categories of the users according to the course categories,represents a learning ability level of a learner, whereinCorresponding to the primary level, the middle level and the high level respectively, the similarity calculation in the new user class uses standard Gaussian distribution, and the expression isWhere U represents the total number of users,indicating a user of the jth category c,which is indicative of a target user of the mobile communication system,、respectively represent the learning levels of the two users,the larger the value of (A) is, the greater the similarity between two users is, and conversely, the smaller the similarity is.
5. The intelligent recommendation method based on big teaching data as claimed in claim 4, wherein the historical learning information of the user is counted to obtain the interest label of the user, and the expression isWhereinRepresenting a userLearning of ith history informationThe number of times a class course is to be classified,a process for calculating in user interest categories from extracted user interest tags, representing a total number of learning of a user, comprising:
calculating the similarity of learning characteristics according to the learning characteristics of old users, calculating the behavior characteristics of the same course by using two users, expressing the behaviors of the courses by using vectors, wherein the expression isWhere U represents the total number of users,a representation of the target user is provided,indicating other users in the same category and,,,representing a userAs to the behavioral characteristics of the lesson K,representing a userFor the behavior characteristics of course K, K represents the course that two users have learned together,respectively representing the time of the learner in learning the course, the acquired experience, the acquired score, the learning progress and the job submitting times;
the expression of similarity calculation of common courses of two users isWhere k represents a course of co-learning between two users,n represents the total number of courses that two users have learned together;
6. The intelligent recommendation method based on the education big data as claimed in claim 5, wherein the category of interest of the target user in the history information of the target user is calculated according to the expression of the user interest tag, and other users with the same interest as the target user are found to obtain the user score, including:
according to the expression asCalculating the scoring similarity among users, wherein the user set isCollection of course resources,Representing a userAndi represents a common scored course,andrespectively represent、The average value among the common score items,、respectively represent usersAndthe step of course scoring, which is to add the learner model similarity and the scoring similarity to obtain an expression asSorting the calculation results from high to low, and taking the first N as a neighbor set;
direct calculation by the above-mentioned scoreNeighbor, predicting the grade of the course which does not generate grade of the target user in the final neighbor users, selecting the first N courses with the highest predicted grade for recommendation, and the calculation expression isWhereinRepresenting a userThe predictive score for the non-scored item i,representing a userThe average score of (a) is calculated,representing a userThe average score of (a) is calculated,representThe score for the item i is given to,representing a userThe neighbor set of (2).
7. The base of claim 1The intelligent recommendation method for the big teaching data is characterized in that the resource score of the user preference behavior data is calculated, the flow times, the comment times and the download times have usability, a normalization method is used for carrying out data standardization processing, and the expression isWhereinIndicates the score, X indicates the number of times,the minimum value of the number of times is represented,represents the maximum value of the times and is uniformly mapped to;
Further obtaining an expression asIn whichRepresenting a userTo resourcesThe number of times of browsing of (a) is scored,representing a userFor resourcesThe number of reviews scored in (a) is,representing a userFor resourcesScoring the download times to obtain the userFor resourcesPreliminary user preference behavior data scoring of,According to different contribution degrees of the comment times, the comment times and the download times to the scores, different weights are given, and the user preference behavior data weight is combined to obtain the final user preference behavior data score, wherein the expression isWherein,、Andrespectively a browsing frequency weight, a comment frequency weight and a downloading frequency weight,representing a userFor resourcesThe end user preferred behavior data score.
8. The intelligent recommendation method based on big teaching data as claimed in claim 1, wherein the screening and prediction based on the data information derived from the candidate resources comprises:
screening and predicting data information obtained by a network structure, converting and transmitting original network data to obtain model input data required by constructing a network model, and expressing the model input data asWhereinRepresenting a user item matrix;
if it isWhen the user is interested in the project, the calculated information is higher than the resource information score ignored by the user, and a relation matrix of the user and the project is obtained according to the difference between the high attention degree and the non-attention degree;
the user is also interested in the user's history when searching for pages when focusing on resource information,when the user pays attention to the network page, recording the ID information of the network user, if the user ID information isDetermining user search behavior characteristics by browsing records by userTo express, according to the relation matrix of the user and the project, obtaining the information interaction record of the user and recording the information interaction record as,Obtaining behavior height by correlating with user search behavior characteristicsThe expression isWhereinIndicates the number of platform users,and representing the quantity of the historical records, and obtaining the length of the queue by obtaining the action height through the column splitting and the set operation and coinciding with the basic information of the user.
9. The intelligent recommendation method based on big teaching data as claimed in claim 8, wherein the history records searched by the user are classified as the feature points of the user by SoftMax, and if the unknown user is u, a function is constructedWherein P represents the probability of the ith class, v represents the feature quantification, w represents the dimensional coefficient, and x represents the edge node distribution according to a functionAnd classifying the item queue to obtain a candidate queue.
10. The intelligent recommendation method based on big teaching data as claimed in claim 1, wherein the step of obtaining historical record data searched by the user and transmitting the historical record data to the candidate queue comprises:
assuming that N is the number of files, different files contain key information, and if the number of the key information in the files representsThe number of times that the key information appears in the file isIn whichRepresenting the most times of key information, obtaining the key information by adopting an exhaustive algorithm, and selecting the maximum frequency of occurrence of words;
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211472211.6A CN115809376A (en) | 2022-11-23 | 2022-11-23 | Intelligent recommendation method based on big teaching data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211472211.6A CN115809376A (en) | 2022-11-23 | 2022-11-23 | Intelligent recommendation method based on big teaching data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115809376A true CN115809376A (en) | 2023-03-17 |
Family
ID=85483872
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211472211.6A Pending CN115809376A (en) | 2022-11-23 | 2022-11-23 | Intelligent recommendation method based on big teaching data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115809376A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117290398A (en) * | 2023-09-27 | 2023-12-26 | 广东科学技术职业学院 | Course recommendation method and device based on big data |
-
2022
- 2022-11-23 CN CN202211472211.6A patent/CN115809376A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117290398A (en) * | 2023-09-27 | 2023-12-26 | 广东科学技术职业学院 | Course recommendation method and device based on big data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021203819A1 (en) | Content recommendation method and apparatus, electronic device, and storage medium | |
CN109299372B (en) | Intelligent course selection recommendation method based on collaborative filtering | |
CN106951471B (en) | SVM-based label development trend prediction model construction method | |
TW201706884A (en) | Data analysis system, data analysis method, data analysis program, and storage medium | |
CN105069470A (en) | Classification model training method and device | |
CN107016122B (en) | Knowledge recommendation method based on time migration | |
CN111949887A (en) | Item recommendation method and device and computer-readable storage medium | |
CN111966888B (en) | Aspect class-based interpretability recommendation method and system for fusing external data | |
CN111914162B (en) | Method for guiding personalized learning scheme based on knowledge graph | |
CN111177538A (en) | Unsupervised weight calculation-based user interest tag construction method | |
CN111241992B (en) | Face recognition model construction method, recognition method, device, equipment and storage medium | |
CN111523055A (en) | Collaborative recommendation method and system based on agricultural product characteristic attribute comment tendency | |
CN113946754A (en) | User portrait based rights and interests recommendation method, device, equipment and storage medium | |
CN113656687B (en) | Teacher portrait construction method based on teaching and research data | |
CN111949885A (en) | Personalized recommendation method for scenic spots | |
CN115809376A (en) | Intelligent recommendation method based on big teaching data | |
CN112598405B (en) | Business project data management method and system based on big data | |
CN108509588B (en) | Lawyer evaluation method and recommendation method based on big data | |
CN112464106B (en) | Object recommendation method and device | |
CN113255843A (en) | Speech manuscript evaluation method and device | |
CN116663909A (en) | Provider risk identification data processing method and device | |
JP2005267095A (en) | Information display method and device, and information display program | |
CN111583363A (en) | Visual automatic generation method and system for image-text news | |
CN115510326A (en) | Internet forum user interest recommendation algorithm based on text features and emotional tendency | |
CN115829683A (en) | Power integration commodity recommendation method and system based on inverse reward learning optimization |
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 |