CN107256494A - A kind of item recommendation method and device - Google Patents
A kind of item recommendation method and device Download PDFInfo
- Publication number
- CN107256494A CN107256494A CN201710352695.3A CN201710352695A CN107256494A CN 107256494 A CN107256494 A CN 107256494A CN 201710352695 A CN201710352695 A CN 201710352695A CN 107256494 A CN107256494 A CN 107256494A
- Authority
- CN
- China
- Prior art keywords
- article
- scored
- scoring
- user
- recommended
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0255—Targeted advertisements based on user history
-
- 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
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Databases & Information Systems (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The applicable field of computer technology of the present invention is there is provided a kind of item recommendation method and device, and this method includes:According to the similarity of scored in history score data article and article to be scored, calculate the factor of influence that history score data treats the prediction scoring of scoring article, the limited Boltzmann machine model that the factor of influence input that calculating is obtained is pre-established, prediction scoring of the user to article to be recommended is calculated by being limited Boltzmann machine model, generate recommendation list, according to the recommendation list of generation, recommend article to user's output, the problem of so as to solve cold start-up during recommendation new article, improve the accuracy rate of recommendation.
Description
Technical field
The invention belongs to field of computer technology, more particularly to a kind of item recommendation method and device.
Background technology
With making rapid progress for Internet technology, the life style of user there occurs great change.It is full in information beautiful jade
The Internet era of mesh, competitive excitation, how to help user fast and accurately to pick out its article interested, one is interconnected
Net enterprise most important.Based on above mentioned problem, commending system technology is arisen at the historic moment.Collaborative filtering is used in commending system
A most wide, most popular technology, conventional collaborative filtering has the method based on closest method and based on model.Base
Clustering Model, Bayesian Classification Model, hidden factor model, graph model are subdivided into the method for model, wherein for the hidden factor
The research effect of model is best.The most typical representative of the research of proposed algorithm model based on the hidden factor be 2007 by
Limited Boltzmann machine (the Restricted of 2 layers of utilization that RuslanSalakhutdinov et al. is proposed
BoltzmannMachines, abbreviation RBM) learn the hidden factor of user or article, set up model and recommended.
But, although deep learning has successfully been applied to commending system field, and obtain by the collaborative filtering based on RBM
Good recommendation effect, but rating matrix of the user to article is used only in it, and under normal circumstances, rating matrix is very dilute
Dredge, so as to cause the recommendation accuracy rate degradation of commending system.In addition, the existing collaborative filtering based on RBM for
The problem of also there is cold start-up in new article.
The content of the invention
It is an object of the invention to provide a kind of item recommendation method and device, it is intended to solves existing commending system for new
The problem of there is cold start-up in article and recommend accuracy rate it is not high, cause the problem of recommendation effect is bad.
On the one hand, the invention provides a kind of item recommendation method, methods described comprises the steps:
According to the similarity of scored in history score data article and article to be scored, the history score data is calculated
To the factor of influence of the prediction scoring of the article to be scored;
The limited Boltzmann machine model that the factor of influence input that the calculating is obtained is pre-established, by described limited
Boltzmann machine model calculates prediction of the user to the article to be recommended and scored, and generates recommendation list;
According to the recommendation list of the generation, article is recommended to user output.
On the other hand, the invention provides a kind of article recommendation apparatus, described device includes:
Factor calculating unit, for the similarity according to scored in history score data article and article to be scored, meter
Calculate factor of influence of the history score data to the prediction scoring of the article to be scored;
List generation unit, for calculating the limited Boltzmann machine that the input of obtained factor of influence is pre-established by described
Model, calculates prediction of the user to the article to be recommended by the limited Boltzmann machine model and scores, generation is pushed away
Recommend list;And
List recommendation unit, for the recommendation list according to the generation, article is recommended to user output.
The present invention calculates history scoring number according to the similarity of scored in history score data article and article to be scored
The limited bohr pre-established according to the factor of influence for the prediction scoring for treating scoring article, the factor of influence input that calculating is obtained
Hereby graceful machine model, calculates prediction scoring of the user to article to be recommended by being limited Boltzmann machine model, generates recommendation list,
According to the recommendation list of generation, article is recommended to user's output, so that the problem of solving cold start-up when recommending new article, carries
The high accuracy rate recommended.
Brief description of the drawings
Fig. 1 is the implementation process figure for the item recommendation method that the embodiment of the present invention one is provided;
Fig. 2 is the structural representation for the article recommendation apparatus that the embodiment of the present invention two is provided;And
Fig. 3 is the preferred structure schematic diagram for the article recommendation apparatus that the embodiment of the present invention two is provided.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Implementing for the present invention is described in detail below in conjunction with specific embodiment:
Embodiment one:
Fig. 1 shows the implementation process for the item recommendation method that the embodiment of the present invention one is provided, and for convenience of description, only shows
The part related to the embodiment of the present invention is gone out, details are as follows:
In step S101, according to the similarity of scored in history score data article and article to be scored, calculating is gone through
Commentary on historical events or historical records divided data treats the factor of influence of the prediction scoring of scoring article.
In embodiments of the present invention, scored the similarity of article and article score according to the user being previously obtained, and counts
Calculate the factor of influence that the article that scored in history score data treats the prediction scoring of scoring article.
Preferably, the article that scored in history score data is calculated treats the factor of influence of the prediction scoring of scoring article
Before, scored the content of text of article in the history score data that article to be scored and user are received first, to these texts
Content carries out participle, gone after stop words pretreatment operation, as the input content of default Word2vec models, to Word2vec
Model is trained, after Word2vec model trainings are finished, and obtains the term vector of the word of each in content of text, using training
Word2vec models obtain after scoring article and the article that scored content in each word term vector represent after, by right
The term vector of all words in item contents, which is overlapped, to be obtained the term vectors of each item contents and represents, so that convenient pushing away
Article is represented during recommending.
Preferably, after item contents are represented using term vector, formula is used
The similarity between the content of each article in the history score data of article to be scored and user is calculated, wherein, Ti、Tj
The term vector of two item contents of expression of progress Similarity Measure is represented, n represents the size of term vector, Ti,i、Tj,iRepresent Ti
And TjComponent.
Preferably, between the content of each article in obtaining the history score data of article to be scored and user
After similarity, according to similar between article to be scored and the content of each article in the history score data of user
Degree, is selected and the maximum scoring article of article similarity to be scored, to improve the accuracy of factor of influence.
Preferably, select and the maximum article of scoring of article similarity to be scored after, according to user couple with it is to be evaluated
Divide the scoring of the maximum article that scored of article similarity, calculate and the maximum article of scoring of article similarity to be scored is treated
The factor of influence that the prediction of article of scoring is scored, so as to improve the accuracy of factor of influence.
It is further preferred that treating the pre- of scoring article calculating the scored article maximum with article similarity to be scored
During the factor of influence of test and appraisal point, formula is usedCalculate and article similarity to be scored is maximum
Scoring article treats the factor of influence of the prediction scoring of scoring article, wherein,For user U in the history score data of user
To article SkScoring, si,sk∈ S={ s1,s2,...,sj, S={ s1,s2,...,sjFor user U the article that scored collection
Close.
In step s 102, the limited Boltzmann machine model that factor of influence input calculating obtained is pre-established, leads to
Cross limited Boltzmann machine model and calculate prediction scoring of the user to article to be recommended, generate recommendation list.
In embodiments of the present invention, history score data composition rating matrix, will calculate obtained factor of influence input
Before the limited Boltzmann machine model pre-established, by the use of rating matrix as input, the training of RBM models is carried out, is passed through
RBM models after training calculate user and the initial predicted of article to be recommended are scored.Preferably, the RBM moulds after according to training
When type calculating user scores the initial predicted of article to be recommended, formula is usedUser is calculated to be recommended
The initial predicted scoring of article, wherein, l represents to be used to carry out article scoring to be recommended in the neutral net visible layer of RBM models
Softmax neural units number, l={ 1,2 ... ..., K },Represent the neutral net hidden layer h from RBM models
The probability of k-th of softmax neural unit of i-th of article to be recommended of visible layer is calculated, article to be recommended is entered so as to realize
Row initial predicted scores.
In embodiments of the present invention, after calculating obtains initial predicted scoring of the user to article to be recommended, calculate and use
Prediction of the family to article to be recommended is scored.
Preferably, when calculating user scores the prediction of article to be recommended, according to maximum with article similarity to be scored
The article that scored treat scoring article prediction scoring factor of influence θ to prediction scoring influence, by θ assign one
Weight, uses final prediction score calculation formulaCalculate prediction of the user to article to be recommended
Scoring, so that the influence for the article that both considers to have scored when being predicted scoring to article to be recommended is realized, it is further contemplated that in article
The influence of similarity between appearance, improves the accuracy of prediction scoring.In addition, if article to be recommended is new article, then new thing
Although product can not be predicted scoring by collaborative filtering, it is available with similarity between item contents and is predicted to comment
Point, so that the problem of solving new article cold start-up.
In embodiments of the present invention, after prediction scoring of the user to article to be recommended is obtained, recommendation list is generated.It is excellent
Selection of land, when generating recommendation list, positive sequence sequence, choosing are carried out according to the size of prediction scoring to the prediction scoring of article to be recommended
The recommendation list that scoring highest N article generation size to be recommended is N is taken, so as to improve the accuracy rate of recommendation.
In step s 103, according to the recommendation list of generation, article is recommended to user's output.
In embodiments of the present invention, the N items article to be recommended in recommendation list is exactly that commending system is the thing that user u recommends
Product, therefore, recommendation list are set to the recommendation results of commending system, and pushing away for this N article to be recommended composition is exported to user
List is recommended, it is achieved thereby that the recommendation to user, improves the accuracy rate of recommendation.
Can be with one of ordinary skill in the art will appreciate that realizing that all or part of step in above-described embodiment method is
The hardware of correlation is instructed to complete by program, described program can be stored in a computer read/write memory medium,
Described storage medium, such as ROM/RAM, disk, CD.
Embodiment two:
Fig. 2 shows the structure for the article recommendation apparatus that the embodiment of the present invention two is provided, and for convenience of description, illustrate only
The part related to the embodiment of the present invention, including:
Factor calculating unit 21, for the similarity according to scored in history score data article and article to be scored,
Calculate the factor of influence that history score data treats the prediction scoring of scoring article.
In embodiments of the present invention, factor calculating unit has scored article and article to be scored according to the user being previously obtained
Similarity, calculate history score data in scored article treat scoring article prediction scoring factor of influence.
Preferably, the article that scored in history score data is calculated treats the factor of influence of the prediction scoring of scoring article
Before, scored the content of text of article in the history score data that article to be scored and user are received first, to these texts
Content carries out participle, gone after stop words pretreatment operation, as the input content of default Word2vec models, to Word2vec
Model is trained, after Word2vec model trainings are finished, and obtains the term vector of the word of each in content of text, using training
Word2vec models obtain after scoring article and the article that scored content in each word term vector represent after, by right
The term vector of all words in item contents, which is overlapped, to be obtained the term vectors of each item contents and represents, so that convenient pushing away
Article is represented during recommending.
Preferably, after item contents are represented using term vector, formula is used
The similarity between the content of each article in the history score data of article to be scored and user is calculated, wherein, Ti、Tj
The term vector of two item contents of expression of progress Similarity Measure is represented, n represents the size of term vector, Ti,i、Tj,iRepresent Ti
And TjComponent.
Preferably, between the content of each article in obtaining the history score data of article to be scored and user
After similarity, according to similar between article to be scored and the content of each article in the history score data of user
Degree, is selected and the maximum scoring article of article similarity to be scored, to improve the accuracy of factor of influence.
Preferably, select and the maximum article of scoring of article similarity to be scored after, according to user couple with it is to be evaluated
Divide the scoring of the maximum article that scored of article similarity, calculate and the maximum article of scoring of article similarity to be scored is treated
The factor of influence that the prediction of article of scoring is scored, so as to improve the accuracy of factor of influence.
It is further preferred that treating the pre- of scoring article calculating the scored article maximum with article similarity to be scored
During the factor of influence of test and appraisal point, formula is usedCalculate and article similarity to be scored is maximum
Scoring article treats the factor of influence of the prediction scoring of scoring article, wherein,For user U in the history score data of user
To article SkScoring, si,sk∈ S={ s1,s2,...,sj, S={ s1,s2,...,sjFor user U the article that scored collection
Close.
List generation unit 22, for the limited Boltzmann machine mould that the input of obtained factor of influence is pre-established will to be calculated
Type, by limited Boltzmann machine model, calculates prediction of the user to article to be recommended and scores, generate recommendation list.
In embodiments of the present invention, history score data composition rating matrix, list generation unit will calculate what is obtained
Before the limited Boltzmann machine model that factor of influence input is pre-established, by the use of rating matrix as input, RBM models are carried out
Training, user is calculated by RBM models after training the initial predicted of article to be recommended is scored.Preferably, according to instruction
When RBM models calculating user after white silk scores the initial predicted of article to be recommended, formula is usedCalculate
User scores the initial predicted of article to be recommended, wherein, l is represented in the neutral net visible layers of RBM models for being treated
The number of the softmax neural units of recommendation article scoring, l={ 1,2 ... ..., K },Represent the god from RBM models
The probability of k-th of softmax neural unit of i-th of article to be recommended of visible layer is calculated through network hidden layer h, so as to realize pair
Article to be recommended carries out initial predicted scoring.
In embodiments of the present invention, list generation unit obtains initial predicted scoring of the user to article to be recommended in calculating
Afterwards, prediction of the user to article to be recommended is calculated to score.
Preferably, when calculating user scores the prediction of article to be recommended, according to maximum with article similarity to be scored
The article that scored treat scoring article prediction scoring factor of influence θ to prediction scoring influence, by θ assign one
Weight, uses final prediction score calculation formulaCalculate prediction of the user to article to be recommended
Scoring, so that the influence for the article that both considers to have scored when being predicted scoring to article to be recommended is realized, it is further contemplated that in article
The influence of similarity between appearance, improves the accuracy of prediction scoring.In addition, if article to be recommended is new article, then new thing
Although product can not be predicted scoring by collaborative filtering, it is available with similarity between item contents and is predicted to comment
Point, so that the problem of solving new article cold start-up.
In embodiments of the present invention, list generation unit is raw after prediction scoring of the user to article to be recommended is obtained
Into recommendation list.Preferably, when generating recommendation list, according to prediction of the size to article to be recommended of prediction scoring score into
Row positive sequence sorts, and the recommendation list that scoring highest N article generation size to be recommended is N is chosen, so as to improve the standard of recommendation
True rate.
List recommendation unit 23, for the recommendation list according to generation, article is recommended to user's output.
In embodiments of the present invention, the N items article to be recommended in recommendation list is exactly that commending system is the thing that user u recommends
Recommendation list is set to the recommendation results of commending system by product, therefore, list recommendation unit, to user export this N it is to be recommended
The recommendation list of article composition, it is achieved thereby that the recommendation to user, improves the accuracy rate of recommendation.
It is therefore preferred that as shown in figure 3, the factor calculating unit 21 includes:
Scored article in vector representation unit 311, the history score data for receiving article to be scored and user
Content of text, is trained to default Word2vec models, obtains the term vector of the word of each in content of text, using obtaining
Term vector represent the content of article to be scored and the article that scored;
Similarity calculated 312, for usingCalculate article to be scored and use
Similarity between the content of each article in the history score data at family, wherein, Ti、TjRepresent to carry out Similarity Measure
Two item contents of expression term vector, n be term vector size, Ti,i、Tj,iRepresent TiAnd TjComponent;
Similarity-rough set unit 313, for each thing in the history score data according to article to be scored and user
Similarity between the content of product, is obtained and the maximum scoring article of article similarity to be scored;And
Factor computation subunit 314, for according to user couple the scored article maximum with article similarity to be scored
Scoring, is calculated and the maximum article that scored of article similarity to be scored treats the factor of influence that the prediction of scoring article is scored;
Preferably, the list generation unit 22 includes:
Initial computation unit 321, for using formulaUser is calculated to the initial of article to be recommended
Prediction scoring, wherein, l represents to be used to carry out article scoring to be recommended in the neutral net visible layer of limited Boltzmann machine model
Softmax neural units number,Represent to calculate from the neutral net hidden layer h of limited Boltzmann machine model
The probability of k-th of softmax neural unit of i-th of article to be recommended of visible layer;
Score calculation unit 322, for using formulaUser is calculated to article to be recommended
Prediction scoring, wherein, α be factor of influence θ weight;
List generates subelement 323, and the prediction scoring of article to be recommended is carried out just for the size according to prediction scoring
Sequence sorts, the recommendation list that the article generation size to be recommended of preceding N in selected and sorted is N;
Preferably, the factor computation subunit 314 includes:
Formula computing unit, for using formulaCalculate and article similarity to be scored most
The big article that scored treats the factor of influence of the prediction scoring of scoring article, wherein,For in the history score data of user
User U is to article SkScoring, si,sk∈ S={ s1,s2,...,sj, S={ s1,s2,...,sjBe user U scoring thing
The set of product.
In embodiments of the present invention, each unit of article recommendation apparatus can be realized by corresponding hardware or software unit, respectively
Unit can be independent soft and hardware unit, a soft and hardware unit can also be integrated into, herein not to limit the present invention.
The embodiment of each unit refers to the description of previous embodiment one, will not be repeated here.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
Any modifications, equivalent substitutions and improvements made within refreshing and principle etc., should be included in the scope of the protection.
Claims (10)
1. a kind of item recommendation method, it is characterised in that methods described comprises the steps:
According to the similarity of scored in history score data article and article to be scored, the history score data is calculated to institute
State the factor of influence of the prediction scoring of article to be scored;
The limited Boltzmann machine model that the factor of influence input that the calculating is obtained is pre-established, passes through the limited bohr
Hereby graceful machine model calculates prediction of the user to the article to be recommended and scored, and generates recommendation list;
According to the recommendation list of the generation, article is recommended to user output.
2. the method as described in claim 1, it is characterised in that the user is calculated by the limited Boltzmann machine model
The step of scoring the prediction of the article to be recommended, including:
Use formulaCalculate the user to score to the initial predicted of the article to be recommended, the l tables
Showing is used for the softmax nerves for carrying out article scoring to be recommended in the neutral net visible layer of the limited Boltzmann machine model
The number of unit, it is describedRepresenting can described in the neutral net hidden layer h calculating from the limited Boltzmann machine model
See the probability of softmax neural units described in k-th of i-th of article to be recommended of layer;
Use formulaCalculate prediction of the user to the article to be recommended to score, the α
For the weight of the factor of influence θ.
3. the method as described in claim 1, it is characterised in that according to article and the thing to be scored of having been scored in history score data
The similarity of product, the step of calculating the factor of influence that prediction of the history score data to the article to be scored is scored, bag
Include:
Scored the content of text of article in the history score data for receiving the article to be scored and the user, to default
Word2vec models are trained, and obtain the term vector of each word in the content of text, use the obtained term vector
Represent the content of the article to be scored and the article that scored;
Use formulaCalculate the article to be scored and the history scoring number of the user
Similarity between the content of each article in, the Ti、TjRepresent two things of expression of progress Similarity Measure
The term vector of product content, the n represents the size of the term vector, the Ti,i、Tj,iRepresent the TiWith the TjComponent;
According to similar between the article to be scored and the content of each article in the history score data of the user
Degree, obtains the scoring article maximum with the article similarity to be scored;
According to the scoring of the user couple and the maximum article that scored of the article similarity to be scored, calculate it is described with it is described
Maximum factor of influence of the article to the prediction scoring of the article to be scored that scored of article similarity to be scored.
4. method as claimed in claim 3, it is characterised in that calculate described maximum with the article similarity to be scored
The step of factor of influence that prediction of the article to the article to be scored of scoring is scored, including:
Use formulaCalculate the scored thing maximum with the article similarity to be scored
The factor of influence that prediction of the product to the article to be scored is scored, it is describedFor user U in the history score data of the user
To article SkScoring, the si,sk∈ S={ s1,s2,...,sj, the S={ s1,s2,...,sjFor the user U
The set of scoring article.
5. the method as described in claim 1, it is characterised in that the step of generating recommendation list, including:
Positive sequence sequence is carried out to the prediction scoring of the article to be recommended according to the size of the prediction scoring, the sequence is selected
In the preceding N articles to be recommended generation sizes be N the recommendation list.
6. a kind of article recommendation apparatus, it is characterised in that described device includes:
Factor calculating unit, for the similarity according to scored in history score data article and article to be scored, calculates institute
State factor of influence of the history score data to the prediction scoring of the article to be scored;
List generation unit, for calculating the limited Boltzmann machine mould that the input of obtained factor of influence is pre-established by described
Type, calculates prediction of the user to the article to be recommended by the limited Boltzmann machine model and scores, generation is recommended
List;And
List recommendation unit, for the recommendation list according to the generation, article is recommended to user output.
7. device as claimed in claim 6, it is characterised in that the list generation unit includes:
Initial computation unit, for using formulaThe user is calculated to the initial of the article to be recommended
Prediction scoring, the l represents to be used to carry out article to be recommended in the neutral net visible layer of the limited Boltzmann machine model
The number of the softmax neural units of scoring, it is describedRepresent the neutral net from the limited Boltzmann machine model
Hidden layer h calculates the probability of softmax neural units described in k-th of i-th of article to be recommended of visible layer;
Score calculation unit, for using formulaThe user is calculated to the article to be recommended
Prediction scoring, the α be the factor of influence θ weight.
8. device as claimed in claim 6, it is characterised in that the factor calculating unit includes:
Scored article in vector representation unit, the history score data for receiving the article to be scored and the user
Content of text, is trained to default Word2vec models, obtains the term vector of each word in the content of text, use
The obtained term vector represents the content of the article to be scored and the article that scored;
Similarity calculated, for usingThe article to be scored is calculated to use with described
Similarity between the content of each article in the history score data at family, the Ti、TjRepresent to carry out Similarity Measure
The term vector of two item contents is represented, the n is the size of the term vector, the Ti,i、Tj,iRepresent the TiWith
The TjComponent;
Similarity-rough set unit, for according to the article to be scored and each thing in the history score data of the user
Similarity between the content of product, obtains the scoring article maximum with the article similarity to be scored;And
Factor computation subunit, for according to the user couple the scored article maximum with the article similarity to be scored
Scoring, calculates the scored article prediction to the to be scored article maximum with the article similarity to be scored and scores
Factor of influence.
9. device as claimed in claim 8, it is characterised in that the factor computation subunit includes:
Formula computing unit, for using formulaCalculate with the article similarity to be scored most
The big factor of influence of the article to the prediction scoring of the article to be scored that scored, it is describedFor going through for the user
User U is to article S in commentary on historical events or historical records divided datakScoring, the si,sk∈ S={ s1,s2,...,sj, the S={ s1,s2,...,
sjFor the user U the article that scored set.
10. device as claimed in claim 6, it is characterised in that the list generation unit includes:
List generates subelement, and the prediction scoring of the article to be recommended is carried out just for the size according to the prediction scoring
Sequence sorts, and selects the recommendation list that preceding N in the sequence article generation sizes to be recommended are N.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710352695.3A CN107256494B (en) | 2017-05-18 | 2017-05-18 | Article recommendation method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710352695.3A CN107256494B (en) | 2017-05-18 | 2017-05-18 | Article recommendation method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107256494A true CN107256494A (en) | 2017-10-17 |
CN107256494B CN107256494B (en) | 2020-10-27 |
Family
ID=60027262
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710352695.3A Active CN107256494B (en) | 2017-05-18 | 2017-05-18 | Article recommendation method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107256494B (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108052670A (en) * | 2017-12-29 | 2018-05-18 | 北京奇虎科技有限公司 | A kind of recommendation method and device of camera special effect |
CN108198045A (en) * | 2018-01-30 | 2018-06-22 | 东华大学 | The design method of mixing commending system based on e-commerce website data mining |
CN108629665A (en) * | 2018-05-08 | 2018-10-09 | 北京邮电大学 | A kind of individual commodity recommendation method and system |
CN108763318A (en) * | 2018-04-27 | 2018-11-06 | 达而观信息科技(上海)有限公司 | item recommendation method and device |
CN108846479A (en) * | 2018-07-13 | 2018-11-20 | 河海大学 | The training method and device of recommended method, RBM model based on RBM model |
CN109460512A (en) * | 2018-10-25 | 2019-03-12 | 腾讯科技(北京)有限公司 | Recommendation information processing method, device, equipment and storage medium |
CN109783738A (en) * | 2019-01-22 | 2019-05-21 | 东华大学 | A kind of double extreme learning machine mixing collaborative filtering recommending methods based on more similarities |
CN110110230A (en) * | 2019-04-26 | 2019-08-09 | 华南理工大学 | A kind of recommended method to be scored based on user with comment |
CN110633418A (en) * | 2019-09-20 | 2019-12-31 | 曹严清 | Commodity recommendation method and device |
CN112052388A (en) * | 2020-08-20 | 2020-12-08 | 深思考人工智能科技(上海)有限公司 | Method and system for recommending gourmet stores |
CN112215680A (en) * | 2020-10-21 | 2021-01-12 | 中国银行股份有限公司 | Product recommendation method and device, electronic equipment and storage medium |
CN113743972A (en) * | 2020-08-17 | 2021-12-03 | 北京沃东天骏信息技术有限公司 | Method and device for generating article information |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103324690A (en) * | 2013-06-03 | 2013-09-25 | 焦点科技股份有限公司 | Mixed recommendation method based on factorization condition limitation Boltzmann machine |
CN104424247A (en) * | 2013-08-28 | 2015-03-18 | 北京闹米科技有限公司 | Product information filtering recommendation method and device |
CN106600302A (en) * | 2015-10-19 | 2017-04-26 | 玺阅信息科技(上海)有限公司 | Hadoop-based commodity recommendation system |
-
2017
- 2017-05-18 CN CN201710352695.3A patent/CN107256494B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103324690A (en) * | 2013-06-03 | 2013-09-25 | 焦点科技股份有限公司 | Mixed recommendation method based on factorization condition limitation Boltzmann machine |
CN104424247A (en) * | 2013-08-28 | 2015-03-18 | 北京闹米科技有限公司 | Product information filtering recommendation method and device |
CN106600302A (en) * | 2015-10-19 | 2017-04-26 | 玺阅信息科技(上海)有限公司 | Hadoop-based commodity recommendation system |
Non-Patent Citations (1)
Title |
---|
霍淑华: "基于协同过滤的评分预测推荐算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108052670A (en) * | 2017-12-29 | 2018-05-18 | 北京奇虎科技有限公司 | A kind of recommendation method and device of camera special effect |
CN108198045A (en) * | 2018-01-30 | 2018-06-22 | 东华大学 | The design method of mixing commending system based on e-commerce website data mining |
CN108763318A (en) * | 2018-04-27 | 2018-11-06 | 达而观信息科技(上海)有限公司 | item recommendation method and device |
CN108763318B (en) * | 2018-04-27 | 2022-04-19 | 达而观信息科技(上海)有限公司 | Item recommendation method and device |
CN108629665B (en) * | 2018-05-08 | 2021-07-16 | 北京邮电大学 | Personalized commodity recommendation method and system |
CN108629665A (en) * | 2018-05-08 | 2018-10-09 | 北京邮电大学 | A kind of individual commodity recommendation method and system |
CN108846479A (en) * | 2018-07-13 | 2018-11-20 | 河海大学 | The training method and device of recommended method, RBM model based on RBM model |
CN109460512A (en) * | 2018-10-25 | 2019-03-12 | 腾讯科技(北京)有限公司 | Recommendation information processing method, device, equipment and storage medium |
CN109783738A (en) * | 2019-01-22 | 2019-05-21 | 东华大学 | A kind of double extreme learning machine mixing collaborative filtering recommending methods based on more similarities |
CN109783738B (en) * | 2019-01-22 | 2023-02-28 | 东华大学 | Multi-similarity-based hybrid collaborative filtering recommendation method for double-pole-limit learning machine |
CN110110230A (en) * | 2019-04-26 | 2019-08-09 | 华南理工大学 | A kind of recommended method to be scored based on user with comment |
CN110633418A (en) * | 2019-09-20 | 2019-12-31 | 曹严清 | Commodity recommendation method and device |
CN113743972A (en) * | 2020-08-17 | 2021-12-03 | 北京沃东天骏信息技术有限公司 | Method and device for generating article information |
CN112052388A (en) * | 2020-08-20 | 2020-12-08 | 深思考人工智能科技(上海)有限公司 | Method and system for recommending gourmet stores |
CN112215680A (en) * | 2020-10-21 | 2021-01-12 | 中国银行股份有限公司 | Product recommendation method and device, electronic equipment and storage medium |
CN112215680B (en) * | 2020-10-21 | 2023-09-08 | 中国银行股份有限公司 | Product recommendation method and device, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN107256494B (en) | 2020-10-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107256494A (en) | A kind of item recommendation method and device | |
Zuo et al. | Tag-aware recommender systems based on deep neural networks | |
CN107273438B (en) | Recommendation method, device, equipment and storage medium | |
CN108763362B (en) | Local model weighted fusion Top-N movie recommendation method based on random anchor point pair selection | |
Kaminskas et al. | Location-aware music recommendation using auto-tagging and hybrid matching | |
CN103870973B (en) | Information push, searching method and the device of keyword extraction based on electronic information | |
CN108334592B (en) | Personalized recommendation method based on combination of content and collaborative filtering | |
CN110245299A (en) | Sequence recommendation method and system based on dynamic interaction attention mechanism | |
US11636251B2 (en) | Content aware font recommendation | |
CN112131469A (en) | Deep learning recommendation method based on comment text | |
CN110825850B (en) | Natural language theme classification method and device | |
CN106610970A (en) | Collaborative filtering-based content recommendation system and method | |
CN110263257A (en) | Multi-source heterogeneous data mixing recommended models based on deep learning | |
CN106971200A (en) | A kind of iconic memory degree Forecasting Methodology learnt based on adaptive-migration | |
CN110428295A (en) | Method of Commodity Recommendation and system | |
CN106127506A (en) | A kind of recommendation method solving commodity cold start-up problem based on Active Learning | |
CN108920521A (en) | User's portrait-item recommendation system and method based on pseudo- ontology | |
CN109598586A (en) | A kind of recommended method based on attention model | |
CN113961705A (en) | Text classification method and server | |
CN110210933A (en) | A kind of enigmatic language justice recommended method based on generation confrontation network | |
CN111523940B (en) | Deep reinforcement learning-based recommendation method and system with negative feedback | |
JP2023024932A (en) | System for multi-modal transformer-based item categorization, data processing system, data processing method, and computer implemented method | |
Wu et al. | Predicting latent structured intents from shopping queries | |
Maaloe et al. | Deep belief nets for topic modeling | |
Pan et al. | ChefGAN: Food image generation from recipes |
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 |