CN110489665A - A kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks - Google Patents
A kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks Download PDFInfo
- Publication number
- CN110489665A CN110489665A CN201910759011.0A CN201910759011A CN110489665A CN 110489665 A CN110489665 A CN 110489665A CN 201910759011 A CN201910759011 A CN 201910759011A CN 110489665 A CN110489665 A CN 110489665A
- Authority
- CN
- China
- Prior art keywords
- microblogging
- user
- word
- contextual model
- convolutional neural
- 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
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 48
- 239000000284 extract Substances 0.000 claims abstract description 30
- 230000004927 fusion Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 3
- 230000009193 crawling Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 16
- 241001269238 Data Species 0.000 abstract description 2
- 238000001914 filtration Methods 0.000 description 11
- 238000002474 experimental method Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 230000001537 neural effect Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- YOETUEMZNOLGDB-UHFFFAOYSA-N 2-methylpropyl carbonochloridate Chemical compound CC(C)COC(Cl)=O YOETUEMZNOLGDB-UHFFFAOYSA-N 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000002996 emotional effect Effects 0.000 description 2
- 239000004744 fabric Substances 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 210000003813 thumb Anatomy 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 244000097202 Rathbunia alamosensis Species 0.000 description 1
- 235000009776 Rathbunia alamosensis Nutrition 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000011157 data evaluation Methods 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000007115 recruitment Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Economics (AREA)
- Primary Health Care (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Strategic Management (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks, carries out scene modeling to user from time and the dimension of region two first, extracts the time contextual model and region contextual model of user；Then, building contextual model similarity calculating method is extended the contextual model of user, captures the interested contextual model tendency of user, constructs user individual contextual model library；Finally, constructing personalized microblogging recommended models using convolutional neural networks, the personalized recommendation of microblog users is realized.The building of scenario models and the introducing of convolutional neural networks, the interest for capturing user provide a great help.Finally the present invention is based on true microblog datas and existing algorithm to compare, it was demonstrated that the model has good recommendation effect, compared to existing model all improves 3%-4% in user satisfaction and mean absolute error.
Description
Technical field
The present invention relates to a kind of microblogging personalized recommendation methods, and in particular to one kind is based on scene modeling and convolutional Neural net
The microblogging personalized recommendation method of network, belongs to technical field of information retrieval.
Background technique
Microblogging is as emerging internet social platform, with its unique real-time, opening, interactivity and convenience
People carry out opinion expression and information interchange and provide good medium, beyond tradition media as new information aggregation,
The information for affecting society at a terrific speed propagates pattern.
Currently, people microblogging obtain information by way of be broadly divided into it is following several: first is that by concern good friend hair
The micro-blog information of cloth；Second is that the related hot spot microblogging recommended by " hot topic " of microblog；Third is that the inspection for passing through microblogging
Suo Gongneng retrieval includes the microblogging of particular keywords.Information above acquisition modes are recommended towards all microblog users
, lack a kind of personalized recommendation function for specific user.Simultaneously as microblogging enormous amount, is that user is timely, effective
The interested content of microblog of acquisition oneself bring great difficulty.Therefore, for the individual info service technology of microblogging
The extensive concern of domestic and foreign scholars has been obtained, the hot spot of Social Media area research is become.
Personalized recommendation passes through the information of user and article in acquisition system, using a series of computation model, to user
Information selection and decision provide support.Current personalized recommendation algorithm be broadly divided into proposed algorithm based on collaborative filtering and
Content-based recommendation algorithm.The proposed algorithm (CollaborativeFiltering, CF) of collaborative filtering is earliest recommendation
Model, the connection of discovery user and article, constitutes scoring mainly from historical data (to the scoring of article before such as user)
Matrix carries out personalized recommendation to the scoring of unknown article by prediction user.In the base of the proposed algorithm of collaborative filtering
On plinth, another common recognition model (COnsensusModel, COM) based on probability is formd, by the life for studying group activity
The proposed algorithm based on group is constructed according to the behavioural characteristic of member each in group at process.To solve collaborative filtering
In Sparse Problem, a kind of collaboration knowledge base is embedded in (CollaborativeKnowledgeBaseEmbedding, CKE)
Integrated framework opened, using stacking denoising autocoder and convolution autocoder extract article text representation and
Visual representation, and the real data collection in two different practice scenes closes and demonstrates the applicability of algorithm.
The problem of recommendation in proposed algorithm in order to solve the problems, such as collaborative filtering personalized insufficient, cold start-up and phase
Like the confinement problems of user group, there is content-based recommendation model.Content-based recommendation model have it is stronger can
Explanatory, the recommendation results of each user are determined by its previous behavior, and the personalized aspect of recommendation results has phase
When advantage；Meanwhile when forming recommendation results, directly compare the similitude of Candidate Recommendation object and user interest model, no
There are problems that cold start-up.But there is also recommendation diversity deficiencies to become at any time with user interest for content-based recommendation method
The deficiency of change.
Scholars put forth effort to explore more effective recommended technology in recent years, have also carried out fusion to conventional recommendation algorithm, have improved.
If collaborative filtering is combined with LDA topic model, filter method is coordinated in the mixing of building LDA-MF and LDA_CF；It will association
It is blended with filter algorithm and information filtering algorithm, proposes a kind of mixed recommendation method for merging collaborative filtering and information filtering.
Meanwhile proposed algorithm be not it is self-existent, need to be optimized according to the characteristic of various platforms, scholars
The particularity of social platform data is constantly explored, and studies the proposed algorithm for being suitble to microblog data.It is such as true micro- by research
Rich data analyze the influence of micro-blog information and community information to recommendation results, demonstrate community information to the weight of personalized recommendation
The property wanted；By modifying each stage parameter in traditional collaborative filtering, the influence of community information is added in the algorithm, demonstrates
Have using the SNCF-RM that community information similarity is revised and preferably recommends efficiency；According to label association and user social contact relationship into
Row models, for identification the interest of user；The collaborative filtering based on probabilistic model is devised, is analyzed in the text of tweet
Hold the interactive relation between user, for recommending interested user and microblogging for user.These researchs are on recommending efficiency
Certain success is all achieved, but due to the uniqueness of the complexity of microblogging environment and microblog data, recommendation effect performance
Room for promotion there are also very big.
Microblog is a complicated social environment, the generation of information with to exchange all be that there are specific scenes
Mode, effective capture to contextual model have great importance for promoting microblogging personalized recommendation.
Summary of the invention
It is of the existing technology in order to solve the problems, such as, improve the Sparse Problem of user, promotes the performance of recommendation, this hair
It is bright to construct a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks, from time dimension and region dimension
The microblogging text that two aspects issue user, comment on, forwarding, thumbing up carries out scene modeling, and it is of interest to extract microblog users
Then certain scenarios mode is extended the personalized feelings to form microblog users to the contextual model of user using microblogging corpus
Scape pattern base constructs user-customized recommended model using convolutional neural networks on the basis of personalized contextual model library, real
Personalized recommendation now is carried out to the hot spot microblogging in microblog system.
The present invention adopts the following technical scheme that realize above-mentioned technical purpose.
A kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks, includes the following steps:
S1, the scene modeling of microblog users；
The microblogging progress scene modeling for issuing to user from time and region dimension, commenting on, forward, thumbing up, extracts microblogging
The certain scenarios mode of user's concern, is extended using contextual model of the microblogging corpus to user, forms microblog users
Personalized contextual model library；
S2 constructs microblogging Personalization recommendation model；
User personality recommended models are constructed using convolutional neural networks, the hot spot microblogging in microblog system are carried out personalized
Recommend.
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, wherein S1,
The scene of microblog users models, and carries out as follows:
S11 extracts contextual model；
S12, contextual model it is extensive；
S13 constructs user individual contextual model library.
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, wherein
S11 extracts contextual model and carries out as follows:
S111: carrying out participle pretreatment to microblogging text, if being mentioned in microblogging text there are time word and place word
It takes；
S112: the descriptor and its weight of microblogging are extracted；
S113: according to the extraction of time word and place word progress contextual model in microblogging text；
S114: the when and where of microblogging publication is extracted.
Specifically, S112 extracts the descriptor of microblogging and its weight uses the TI-E algorithm based on topic label information entropy.
Whether the different traditional TF-IDF algorithm of this algorithm, the frequency occurred in microblogging text according to theme related term and appear in micro-
In rich label, descriptor and weight are extracted, shown in method such as formula (1):
Wherein, TIE (wij) indicate word wiTI-E value in microblogging j, TI (wij) indicate word wiTF- in microblogging j
IDF value, shown in calculation method such as formula (2), TagE (wi) indicate word wiLabel information entropy, calculation method such as formula (3) institute
Show.
Wherein, | wij| indicate word wiThe frequency occurred in microblogging j, ∑k|wkj| indicate microblogging j in all word numbers it
With, | D | indicate all microblogging numbers in microblogging corpus, | { dj∶wi∈dj| indicate word w occur in microblogging corpusiMicroblogging
Number, | T (wi) | indicate word w in microblogging corpusiAppear in the microblogging number in topic label.
By calculating the TI-E value of each word in microblogging text, the Topic word of acquisition microblogging text and its corresponding power
Every microblogging, according to the size of weight, is expressed as the set of descriptor and its weight, i.e. WeiBo by weighti={ t1∶w1, t2∶
w2..., tn∶wn, wherein tiBased on write inscription, wiFor its corresponding weight.
Specifically, S113, according in microblogging text time word and place word carry out in the extraction of contextual model, time feelings
The extracting method of scape is as follows:
According in microblogging text time word and microblogging publication time, by the time according to every one period of n hour into
Row divides, and value is respectively { 0, n, 2n, 3n ... (24/n-1) n }, and building time scene extracts shown in model such as formula (4).
Wherein, WeiBoiIt is indicated for the descriptor of microblogging, time is the time value of microblogging publication.
As a preferred solution, the time was divided according to every three hours as a period, value difference
For { 0,3,6,9,12,15,18,21 }, constructs time scene and extract shown in model such as formula (4).
The extracting method of region scene is as follows:
According to the place of place word and microblogging publication in microblogging text, place is divided according to region, value
For the title in region, construct shown in region scenario models such as formula (5).
Wherein, WeiBoiIt is indicated for the descriptor of microblogging, location is the zone name of microblogging publication.
As a preferred mode, place is divided according to province, value is the title in province, constructs region
Scene extracts shown in model such as formula (5).
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, S12, scene
The extensive of mode carries out as follows:
From three time, place, personage dimensions, existing contextual model is generalized for general contextual model, realization pair
User situation model it is extensive.
Abstraction rule is as follows:
Personage: male/female → any people
Place: the subway/public transport → vehicles → anywhere
Family → area → city → province → anywhere
Office → company → anywhere
Time: Monday → working day → any time
So situation s=(man, subway, Monday) can be generalized for s1=(any people, subway, working day), s2=(male
Scholar, anywhere, working day) or s3=(man, subway, any time).
The microblogging for including in user individual contextual model library is microblogging related to user, and reflection is that user is interested
Contextual model tendency, but user's microblogging number for issue, comment on, forward and thumbing up is compared to the microblogging in microblogging library
It is to have accounted for seldom ratio, if only user individual contextual model library is constructed using this part microblogging, as user
The primary data that propertyization is recommended can have serious Sparse phenomenon, in order to further obtain the interested microblogging of more users
Data need on the basis of existing contextual model, by means of microblogging corpus, extract the interested microblogging of more users.
Based on this, S13, building user individual contextual model library is carried out as follows:
S131: crawling the microblogging that specific user issues, comments on, forwarding, thumbing up, and constructs user's microblogging corpus；
S132: using the microblog users of specific number as starting point, user and its follower and the person's of being concerned publication is crawled, is commented
By, the microblogging that forwards, thumb up, microblogging corpus is constructed；
S133: the time contextual model and region scene mould of each microblogging in user's microblogging library and microblogging corpus are extracted
Formula；
S134: the microblogging in microblogging corpus and user's microblogging library is divided according to the value of contextual model；
S135: each microblogging in microblogging and the user's microblogging library in the microblogging corpus under corresponding contextual model value is calculated
Similarity value, if maximum similarity be greater than a certain threshold value, by this microblogging be added user individual contextual model library.
Wherein, similarity calculating method is as follows:
In microblogging corpus in microblogging and user's microblogging library the contextual model of microblogging be expressed as and, wherein t is microblogging
In corresponding descriptor, the weight write inscription based on w, s is the value of corresponding contextual model, then the similarity calculating method of sum is as follows:
Microblogging W in microblogging corpuscWith microblogging W in user's microblogging libraryuContextual model be expressed as Wc={ tc1∶wc1,
tc2∶wc2..., tcn∶wcn；scAnd Wu={ tu1∶wu1, tu2∶wu2..., tun∶wun；su, wherein t is corresponding theme in microblogging
Word, the weight write inscription based on w, s are the value of corresponding contextual model, then WcAnd WuSimilarity calculating method it is as follows:
Appoint and i ∈ { 1,2 ..., n } is taken to calculate word tciAnd tuiBetween semantic similarity, the calculating of semantic similarity uses
Word2Vec [14] model is calculated, and the building of model experimental section will be introduced below.Then, according to semantic phase
Like degree value by WcAnd WuDescriptor be divided into n group, every group is WcAnd WuThe middle most similar word of descriptor semanteme, finally, using formula
(6) calculate every group of descriptor weight weighted average and, result is WcAnd WuSimilarity value.
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, S2, building
Microblogging Personalization recommendation model carries out as follows.
On the basis of user individual microblogging library, convolutional neural networks model is introduced, using the thought of emotional semantic classification, structure
Build microblogging Personalization recommendation model.
Specifically, using the microblogging in user individual microblogging library as the positive example in disaggregated model, from microblogging corpus
In randomly select with the microblogging of positive example number equivalent as negative example, form the training data of user's microblogging Personalization recommendation model,
By the training of model, learns the contextual model tendency of user's microblogging interested, construct user's microblogging Personalization recommendation model.Mould
Type mainly uses the convolutional Neural of multichannel to carry out feature extraction, and carries out feature sampling using pond layer, and by connecting entirely
Layer and Softmax layers of progress semantic classification.
Specifically, the following steps are included:
S21: convolution operation is carried out to input word vector matrix using the filter of the h × k in multiple channels；
S22: in pond, layer is obtained most important feature in characteristic pattern and is grasped as current convolution using the strategy in maximum pond
The feature of work exports；
S23: will extract a plurality of types of features, by being combined to various types of features, be input to full articulamentum
Carry out Fusion Features.
Specific structure is as shown in the microblogging Personalization recommendation model part of Fig. 1.
Specifically, about S21: in convolutional neural networks, using multiple channels h × k filter to input word to
Moment matrix carries out in convolution operation, obtains the local feature in corresponding window, obtains the characteristic pattern of input microblogging text, calculates
As shown in formula (7).
ci=f (wXi∶i+h-1+b) (7)
Wherein, ciIth feature value in the characteristic pattern extracted is represented, f is convolution kernel function,For filter, h
For sliding window size, k is the size of term vector, and b is bias, Xi∶i+h-1Indicate the i-th row to the i-th+h-1 row of input matrix
The local feature matrix of composition.The global feature figure C that input can be obtained in the feature that comprehensive each sliding window extracts is formula
(8) shown in.
C=[c1, c2..., cn-h+1] (8)
Specifically, about S22: in pond, layer is obtained most important feature in characteristic pattern and is made using the strategy in maximum pond
It is exported as shown in formula (9) for the feature of current convolution operation.
Specifically, about S23: a plurality of types of features will be extracted, by being combined to various types of features, inputted
Fusion Features are carried out to full articulamentum.Since the present invention is using the convolutional neural networks of multichannel, multiple types will be extracted
Feature be input to full articulamentum and carry out Fusion Features by being combined to various types of features.In the base of fusion feature
On plinth, by Softmax output category result, the positive example in the result of output will be as the alternative micro- of user-customized recommended
It is rich.
After obtaining alternative microblogging, the time scenario models and region scenario models of every microblogging are extracted according to the method for S1,
Using the construction method in the user individual microblogging library of S13, microblogging of the microblogging as user-customized recommended of TopN is chosen.
The present invention uses above-mentioned technical solution, obtains following technical effect.
1, a kind of micro-blog recommendation method based on scene modeling and convolutional neural networks, by the time for extracting user's concern
Scenario models and region scenario models construct user individual microblogging library, and convolutional Neural is used on the basis of personalized microblogging library
Network carries out classification to hot spot microblogging and realizes that the personalized microblogging of user is recommended.The introducing of scenario models is for obtaining the emerging of user
Interest tendency has huge help, meanwhile, the disaggregated model based on convolutional neural networks is also that the performance boost recommended is brought
Very big help.
2, the extraction of time scenario models and region scenario models brings the interest tendency of acquisition user very big
It helps.Inclined by the scene that the user individual contextual model library of scene modeling building sufficiently covers user's microblogging interested
To.
3, the introducing of the disaggregated model based on convolutional neural networks also brings for the performance boost of recommended models very big
Contribution.The Research on Classification Model of convolutional neural networks has compared sufficiently, and technology is relatively mature, mentions to recommendation performance
It rises very useful.
4, the micro-blog recommendation method of the invention based on scene modeling and convolutional neural networks has good recommendation effect,
3%-4% is all improved in user satisfaction and mean absolute error compared to existing model.
Detailed description of the invention
Fig. 1 is microblogging Personalization recommendation model schematic diagram of the present invention；
Fig. 2 is the relational graph of similarity threshold values α, MAE and AUS of the invention；
Fig. 3 is the comparison diagram of time scenario models and region scenario models of the invention；
Specific embodiment
To keep the purpose of the present invention, technical scheme and beneficial effects clearer, below in conjunction in the embodiment of the present invention
And attached drawing, technical solution of the present invention is clearly and completely described, it is clear that described embodiment is the present invention one
Divide embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making
Every other embodiment obtained, shall fall within the protection scope of the present invention under the premise of creative work.
Embodiment
The present invention provides a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks, including as follows
Step:
S1, the scene modeling of microblog users；
The microblogging progress scene modeling for issuing to user from time and region dimension, commenting on, forward, thumbing up, extracts microblogging
The certain scenarios mode of user's concern, is extended using contextual model of the microblogging corpus to user, forms microblog users
Personalized contextual model library；
S2 constructs microblogging Personalization recommendation model；
User personality recommended models are constructed using convolutional neural networks, the hot spot microblogging in microblog system are carried out personalized
Recommend.
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, wherein S1,
The scene of microblog users models, and carries out as follows:
S11 extracts contextual model；
S12, contextual model it is extensive；
S13 constructs user individual contextual model library.
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, wherein
S11 extracts contextual model and carries out as follows:
S111: carrying out participle pretreatment to microblogging text, if being mentioned in microblogging text there are time word and place word
It takes；
S112: the descriptor and its weight of microblogging are extracted；
S113: according to the extraction of time word and place word progress contextual model in microblogging text；
S114: the when and where of microblogging publication is extracted；
Specifically, S112 extracts the descriptor of microblogging and its weight uses the TI-E algorithm based on topic label information entropy.
Whether the different traditional TF-IDF algorithm of this algorithm, the frequency occurred in microblogging text according to theme related term and appear in micro-
In rich label, descriptor and weight are extracted, shown in method such as formula (1):
Wherein, TIE (wij) indicate word wiTI-E value in microblogging j, TI (wij) indicate word wiTF- in microblogging j
IDF value, shown in calculation method such as formula (2), TagE (wi) indicate word wiLabel information entropy, calculation method such as formula (3) institute
Show.
Wherein, | wij| indicate word wiThe frequency occurred in microblogging j, ∑k|wkj| indicate microblogging j in all word numbers it
With, | D | indicate all microblogging numbers in microblogging corpus, | { dj∶wi∈dj| indicate word w occur in microblogging corpusiMicroblogging
Number, | T (wi) | indicate word w in microblogging corpusiAppear in the microblogging number in topic label.
By calculating the TI-E value of each word in microblogging text, the Topic word of acquisition microblogging text and its corresponding power
Every microblogging, according to the size of weight, is expressed as the set of descriptor and its weight, i.e. WeiBo by weighti={ t1∶w1, t2∶
w2..., tn∶wn, wherein tiBased on write inscription, wiFor its corresponding weight.
Specifically, S113, according in microblogging text time word and place word carry out in the extraction of contextual model, time feelings
The extracting method of scape is as follows:
According in microblogging text time word and microblogging publication time, by the time according to every one period of n hour into
Row divides, and value is respectively { 0, n, 2n, 3n ... (24/n-1) n }, and building time scene extracts shown in model such as formula (4).
Wherein, WeiBoiIt is indicated for the descriptor of microblogging, time is the time value of microblogging publication.
As a preferred solution, the time was divided according to every three hours as a period, value difference
For { 0,3,6,9,12,15,18,21 }, constructs time scene and extract shown in model such as formula (4).
The extracting method of region scene is as follows:
According to the place of place word and microblogging publication in microblogging text, place is divided according to region, value
For the title in region, construct shown in region scenario models such as formula (5).
Wherein, WeiBoiIt is indicated for the descriptor of microblogging, location is the zone name of microblogging publication.
As a preferred mode, place is divided according to province, value is the title in province, constructs region
Scene extracts shown in model such as formula (5).
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, S12, scene
The extensive of mode carries out as follows:
From three time, place, personage dimensions, existing contextual model is generalized for general contextual model, realization pair
User situation model it is extensive.
Abstraction rule is as follows:
Personage: male/female → any people
Place: the subway/public transport → vehicles → anywhere
Family → area → city → province → anywhere
Office → company → anywhere
Time: Monday → working day → any time
So situation s=(man, subway, Monday) can be generalized for s1=(any people, subway, working day), s2=(male
Scholar, anywhere, working day) or s3=(man, subway, any time).
The microblogging for including in user individual contextual model library is microblogging related to user, and reflection is that user is interested
Contextual model tendency, but user's microblogging number for issue, comment on, forward and thumbing up is compared to the microblogging in microblogging library
It is to have accounted for seldom ratio, if only user individual contextual model library is constructed using this part microblogging, as user
The primary data that propertyization is recommended can have serious Sparse phenomenon, in order to further obtain the interested microblogging of more users
Data need on the basis of existing contextual model, by means of microblogging corpus, extract the interested microblogging of more users.
Based on this, S13, building user individual contextual model library is carried out as follows:
S131: crawling the microblogging that specific user issues, comments on, forwarding, thumbing up, and constructs user's microblogging corpus；
S132: using the microblog users of specific number as starting point, user and its follower and the person's of being concerned publication is crawled, is commented
By, the microblogging that forwards, thumb up, microblogging corpus is constructed；
S133: the time contextual model and region scene mould of each microblogging in user's microblogging library and microblogging corpus are extracted
Formula；
S134: the microblogging in microblogging corpus and user's microblogging library is divided according to the value of contextual model；
S135: each microblogging in microblogging and the user's microblogging library in the microblogging corpus under corresponding contextual model value is calculated
Similarity value, if maximum similarity be greater than a certain threshold value, by this microblogging be added user individual contextual model library.
Wherein, similarity calculating method is as follows:
In microblogging corpus in microblogging and user's microblogging library the contextual model of microblogging be expressed as and, wherein t is microblogging
In corresponding descriptor, the weight write inscription based on w, s is the value of corresponding contextual model, then the similarity calculating method of sum is as follows:
Microblogging W in microblogging corpuscWith microblogging W in user's microblogging libraryuContextual model be expressed as Wc={ tc1∶wc1,
tc2∶wc2..., tcn∶wcn；scAnd Wu={ tu1∶wu1, tu2∶wu2..., tun∶wun；su, wherein t is corresponding theme in microblogging
Word, the weight write inscription based on w, s are the value of corresponding contextual model, then WcAnd WuSimilarity calculating method it is as follows:
Appoint and i ∈ { 1,2 ..., n } is taken to calculate word tciAnd tuiBetween semantic similarity, the calculating of semantic similarity uses
Word2Vec [14] model is calculated, and the building of model experimental section will be introduced below.Then, according to semantic phase
Like degree value by WcAnd WuDescriptor be divided into n group, every group is WcAnd WuThe middle most similar word of descriptor semanteme, finally, using formula
(6) calculate every group of descriptor weight weighted average and, result is WcAnd WuSimilarity value.
According to a kind of above-mentioned microblogging personalized recommendation method based on scene modeling and convolutional neural networks, S2, building
Microblogging Personalization recommendation model carries out as follows.
On the basis of user individual microblogging library, convolutional neural networks model is introduced, using the thought of emotional semantic classification, structure
Build microblogging Personalization recommendation model.
Specifically, using the microblogging in user individual microblogging library as the positive example in disaggregated model, from microblogging corpus
In randomly select with the microblogging of positive example number equivalent as negative example, form the training data of user's microblogging Personalization recommendation model,
By the training of model, learns the contextual model tendency of user's microblogging interested, construct user's microblogging Personalization recommendation model.Mould
Type mainly uses the convolutional Neural of multichannel to carry out feature extraction, and carries out feature sampling using pond layer, and by connecting entirely
Layer and Softmax layers of progress semantic classification.
Specifically, the following steps are included:
S21: convolution operation is carried out to input word vector matrix using the filter of the h × k in multiple channels；
S22: in pond, layer is obtained most important feature in characteristic pattern and is grasped as current convolution using the strategy in maximum pond
The feature of work exports；
S23: will extract a plurality of types of features, by being combined to various types of features, be input to full articulamentum
Carry out Fusion Features.
Specific structure is as shown in the microblogging Personalization recommendation model part of Fig. 1.
Specifically, about S21: in convolutional neural networks, using multiple channels h × k filter to input word to
Moment matrix carries out in convolution operation, obtains the local feature in corresponding window, obtains the characteristic pattern of input microblogging text, calculates
As shown in formula (7).
ci=f (wXi∶i+h-1+b) (7)
Wherein, ciIth feature value in the characteristic pattern extracted is represented, f is convolution kernel function,For filter, h
For sliding window size, k is the size of term vector, and b is bias, Xi∶i+h-1Indicate the i-th row to the i-th+h-1 row of input matrix
The local feature matrix of composition.The global feature figure C that input can be obtained in the feature that comprehensive each sliding window extracts is formula
(8) shown in.
C=[c1, c2..., cn-h+1] (8)
Specifically, about S22: in pond, layer is obtained most important feature in characteristic pattern and is made using the strategy in maximum pond
It is exported as shown in formula (9) for the feature of current convolution operation.
Specifically, about S23: a plurality of types of features will be extracted, by being combined to various types of features, inputted
Fusion Features are carried out to full articulamentum.Since the present invention is using the convolutional neural networks of multichannel, multiple types will be extracted
Feature be input to full articulamentum and carry out Fusion Features by being combined to various types of features.In the base of fusion feature
On plinth, by Softmax output category result, the positive example in the result of output will be as the alternative micro- of user-customized recommended
It is rich.
After obtaining alternative microblogging, the time scenario models and region scenario models of every microblogging are extracted according to the method for S1,
Using the construction method in the user individual microblogging library of S13, microblogging of the microblogging as user-customized recommended of TopN is chosen.
Experimental example:
In order to verify the effect of the microblogging personalized recommendation method of the invention based on scene modeling and convolutional neural networks,
Carry out following experiment.
Experimental method
In order to construct microblogging corpus, a microblogging crawler journey is realized using the crawler frame WebCollecter of open source
Sequence crawls user information and its relevant microblog data in Sina weibo.1261967 users and its related letter are crawled altogether
Breath, 182672450 microblog datas, user include ordinary user, celebrity, network marketing number, certification authority, official etc.,
The microblogging quantity and its number of fans and attention number of publication are very uneven, and the constant interval of user's issuing microblog is [3,17382],
The constant interval of the number of fans of user is [12,12006518].It is such as complicated and simple by having carried out a series of microblogging Text Pretreatment
Conversion, url replacement, rejecting of short and small meaningless microblogging etc. finally construct the microblogging corpus that one includes 104 652 972
Library.Microblogging in microblogging corpus includes word 3 334 763 247 altogether, this experiment has trained one with these corpus
Word2Vec model, Skip-gram model is used in trained process, and other relevant parameters are all made of default setting.By instruction
Practice, finally obtained the term vector comprising 850599 words, the dimension of the term vector of each word is 200.
Experimental result
Since there is presently no the common data sets that microblogging recommends aspect, the result of personalized recommendation is also to vary with each individual
, evaluation metrics are difficult directly to evaluate with accuracy rate and recall rate, therefore, using mean absolute error (Mean
Absolute Error, MAE) and user satisfaction (Average User Satisfaction, AUS) evaluated, participate in commenting
The user of survey is the volunteer invited, and evaluation and test person participates in the case where knowing nothing to model evaluating and testing, wherein MAE and AUS
Shown in calculation method such as formula (10) and formula (11):
Wherein, m is the number for participating in the volunteer of evaluation and test, and n is the number for the microblogging that each user recommends, Sim (wi, wij)
It is expressed as the j-th strip microblogging of user i recommendation and the similarity in user individual microblogging library, calculation method is shown in Section 3.3,
feedbackijBe user i to the feedback of the j-th strip microblogging of recommendation, feedback result be divided into three grades " do not like, it is noninductive,
Like ", value is { -1,0,1 }.
On the basis of above data collection and evaluation metrics, the present invention devises following several experiments:
(1) determination of microblogging contextual model similarity threshold values α；
(2) recruitment evaluation of recommended models of the present invention；
(3) comparison of time scenario models and region scenario models.
Experimental result and analysis
The determination of microblogging contextual model similarity threshold values
It is emerging to the sense of user by the similarity calculation of contextual model in the building process of user's microblogging personalization microblogging library
The microblogging of interest is extended, if carrying out the determination of similarity threshold values using the microblogging of all extensions, can be brought to volunteer
Therefore a large amount of feedback operation amount only selects 50 at random from the microblogging of extension and user is allowed to assess.It is adopted in evaluation process
It is assessed with feedback of the AUS index to user.In general, similarity threshold values α value is bigger, and the value of AUS is bigger, but
It is that the microblogging number that excessive threshold values will lead to extension is smaller and smaller, this just loses the meaning for carrying out microblogging extension, performance
It is smaller and smaller for the value of MAE.Therefore, it is necessary to found by the variation tendency of AUS and MAE extension number and similarity threshold it
Between equalization point.
During the experiment, the relationship of similarity threshold values α, MAE and AUS are as shown in Figure 2.
From Fig. 2 it will be seen that with similarity threshold values α increase, MAE is smaller and smaller, and AUS is increasing, in α
When=0.80, the variation of MAE and AUS tend towards stability, and therefore, choose 0.80 value as similarity threshold values α.
The Performance Evaluation of recommendation effect
In order to verify effectiveness of the invention and advance, in selected parameter alpha=0.8, select fusion tag relationship with
Customer relationship proposed algorithm (ILCAUSR)), the microblogging personalized recommendation algorithm (RA-CD) based on community discovery, user interaction words
The microblogging proposed algorithm (IBCF) of topic and the recommended models (SM-CNN) of the invention based on scene modeling and convolutional neural networks
It compares, the parameter in each contrast model is all made of the optimized parameter selected in paper, and experimental result is as shown in table 1.
The recommendation effect comparison of the different models of table 1
| Model | MAE | AUS |
| ILCAUSR | 0.3523 | 0.8043 |
| RA-CD | 0.3323 | 0.8567 |
| IBCF | 0.3242 | 0.8456 |
| SM-CNN | 0.3106 | 0.8842 |
From table 1 it follows that SM-DL model of the invention has reached most no matter in MAE index or AUS index
Excellent effect.The superiority of SM-DL model of the invention is mainly reflected in several aspects:
(1) extraction of time scenario models and region scenario models brings the interest tendency of acquisition user very big
It helps.Inclined by the scene that the user individual contextual model library of scene modeling building sufficiently covers user's microblogging interested
To.
(2) introducing of the disaggregated model based on convolutional neural networks also brings for the performance boost of recommended models very big
Contribution.The Research on Classification Model of convolutional neural networks has compared sufficiently, and technology is relatively mature, mentions to recommendation performance
It rises very useful.
The comparison of time scenario models and region scenario models
Influence for further search time scenario models and region scenario models to recommendation effect, present invention setting
Three groups of control experiments, the recommendation respectively based on time scenario models, the recommendation based on region scene, based on time scene and
The experiment effect of the recommendation that region scene combines, three is as shown in Figure 3.
The comparison of Fig. 3 time scenario models and region scenario models
The recommendation effect of binding time scene and region scene is best as can be seen from Figure 3, while being based on time scene mould
The effect of type is better than based on region scenario models.It is considered herein that in terms of reason mainly has following two: first is that user is sending out
When cloth microblogging, the mark for ground point location is not very comprehensive with specifically, is only positioned to a part of microblogging therein
Mark, a big chunk microblogging all lack specific regional information, so that many microbloggings can not all extract its region contextual model；
Second is that fixation is generally compared in the region of User Activity, few people can frequently replace geographical location, so that its region scene mould
Ground thresholding in formula is relatively simple.The above two o'clock is that the extraction of user region scene brings certain obstacle, also results in base
It is not so good as the recommendation effect based on time scene in the recommendation effect of region scene.
Technical solution provided by the invention, is not restricted to the described embodiments, it is all by structure of the invention and in the way of,
By converting and replacement is formed by technical solution, all within the scope of the present invention.
Claims (10)
1. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks, it is characterised in that including walking as follows
It is rapid:
S1, the scene modeling of microblog users；
The certain scenarios mode that microblog users are paid close attention to is extracted from time and region dimension, forms the personalized scene mould of microblog users
Formula library；
S2 constructs microblogging Personalization recommendation model；
User personality recommended models are constructed using convolutional neural networks, personalization is carried out to the hot spot microblogging in microblog system and is pushed away
It recommends；
Wherein, S1, the scene of microblog users, which models, to be included the following steps,
S11 extracts contextual model；
S12, contextual model it is extensive；
S13 constructs user individual contextual model library；
Wherein, S2, building microblogging Personalization recommendation model carry out as follows,
S21: convolution operation is carried out to input word vector matrix using the filter of the h × k in multiple channels；
S22: in pond, layer uses the strategy in maximum pond, and most important feature is as current convolution operation in acquisition characteristic pattern
Feature output；
S23: will extract a plurality of types of features, by being combined to various types of features, be input to full articulamentum and carry out
Fusion Features.
2. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks as described in claim 1,
It is characterized in that:
Wherein, S11 extracts contextual model and carries out as follows:
S111: carrying out participle pretreatment to microblogging text, if extracting in microblogging text there are time word and place word；
S112: the descriptor and its weight of microblogging are extracted；
S113: according to the extraction of time word and place word progress contextual model in microblogging text；
S114: the when and where of microblogging publication is extracted.
3. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks as claimed in claim 2,
Be characterized in that: the descriptor and its weight that S112 extracts microblogging are using the TI-E algorithm based on topic label information entropy, according to master
It inscribes in the related term frequency occurred in microblogging text and the label for whether appearing in microblogging, extracts descriptor and weight, side
Method is as follows:
Wherein, TIE (wij) indicate word wiTI-E value in microblogging j, TI (wij) indicate word wiTF-IDF in microblogging j
Value, calculation method are shown below:
TagE(wi) indicate word wiLabel information entropy, calculation method is shown below:
Wherein, | wij| indicate word wiThe frequency occurred in microblogging j, ∑k|wkj| indicate the sum of all word numbers in microblogging j, |
D | indicate all microblogging numbers in microblogging corpus, | { dj: wi∈djIndicate word w occur in microblogging corpusiMicroblogging number,
|T(wi) | indicate word w in microblogging corpusiAppear in the microblogging number in topic label；
By calculating the TI-E value of each word in microblogging text, the Topic word of acquisition microblogging text and its corresponding weight,
According to the size of weight, every microblogging is expressed as to the set of descriptor and its weight, i.e. WeiBoi={ t1: w1, t2: w2...,
tn: wn, wherein tiBased on write inscription, wiFor its corresponding weight.
4. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks as claimed in claim 3,
It is characterized in that:
S113, according in microblogging text time word and place word carry out contextual model extraction in, the extraction side of time scene
Method is as follows:
According to the time of time word and microblogging publication in microblogging text, the time is drawn according to every one period of n hour
Point, value is respectively { 0, n, 2n, 3n, (24/n-1) n }, and building time scene extracts model and is shown below；
Wherein, WeiBoiIt is indicated for the descriptor of microblogging, time is the time value of microblogging publication；
The extracting method of region scene is as follows:
According to the place of place word and microblogging publication in microblogging text, place is divided according to region, value is area
The title in domain, building region scenario models are as follows；
Wherein, WeiBoiIt is indicated for the descriptor of microblogging, location is the zone name of microblogging publication.
5. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks according to claim 4,
It is characterized by:
S12, the extensive of contextual model carry out as follows: from three time, place, personage dimensions, by existing feelings
Border mode is generalized for general contextual model, realizes to the extensive of user situation model.
6. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks as claimed in claim 5,
It is characterized in that:
S13, building user individual contextual model library are carried out as follows:
S131: crawling the microblogging that specific user issues, comments on, forwarding, thumbing up, and constructs user's microblogging corpus；
S132: using the microblog users of specific number as starting point, crawl user and its follower and the person's of being concerned publication, comment,
The microblogging forward, thumbed up constructs microblogging corpus；
S133: the time contextual model and region contextual model of each microblogging in user's microblogging library and microblogging corpus are extracted；
S134: the microblogging in microblogging corpus and user's microblogging library is divided according to the value of contextual model；
S135: the phase of the microblogging and each microblogging in user's microblogging library in the microblogging corpus under corresponding contextual model value is calculated
Like angle value, if maximum similarity is greater than a certain threshold value, user individual contextual model library is added in this microblogging.
7. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks as claimed in claim 6,
It is characterized in that: the similarity calculating method in S135 are as follows:
Microblogging W in microblogging corpuscWith microblogging W in user's microblogging libraryuContextual model be expressed as Wc={ tc1: wc1, tc2:
wc2..., tcn: wcn；scAnd Wu={ tu1: wu1, tu2: wu2..., tun: wun；su, wherein t is corresponding descriptor in microblogging, w
Based on the weight write inscription, s is the value of corresponding contextual model, then WcAnd WuSimilarity calculating method it is as follows:
Appoint and i ∈ { 1,2 ..., n } is taken to calculate word tc1And tuiBetween semantic similarity, the calculating of semantic similarity uses
Word2Vec model is calculated, then, according to the value of semantic similarity by WcAnd WuDescriptor be divided into n group, every group is WcWith
WuThe middle most similar word of descriptor semanteme, finally, using following formula calculate every group of descriptor weight weighted average and, result is
For WcAnd WuSimilarity value,
8. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks as described in claim 1,
It is characterized in that:
S21: in convolutional neural networks, convolution behaviour is carried out to input word vector matrix using the filter of the h × k in multiple channels
In work, the local feature in corresponding window is obtained, the characteristic pattern of input microblogging text is obtained, calculates the c that is shown belowi=f
(w·XI:i+h-1+ b),
Wherein, ciIth feature value in the characteristic pattern extracted is represented, f is convolution kernel function,For filter, h is to slide
Dynamic window size, k are the size of term vector, and b is bias, XI:i+h-1Indicate that the i-th row of input matrix is formed to the i-th+h-1 row
Local feature matrix；
The global feature figure C that input can be obtained in the feature that comprehensive each sliding window extracts is to be shown below:
C=[c1, c2..., cn-h+1]。
9. special such as a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks that claim 8 is stated
Sign is:
S22: in pond, layer uses the strategy in maximum pond, and most important feature is as current convolution operation in acquisition characteristic pattern
Feature output is shown below:
10. a kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks as claimed in claim 9,
It is characterized in that:
In S23, a plurality of types of features will be extracted, by being combined to various types of features, be input to full articulamentum into
Row Fusion Features, positive example conduct on the basis of fusion feature, by Softmax output category result, in the result of output
The alternative microblogging of user-customized recommended；
After obtaining alternative microblogging, the time scenario models and region scenario models of every microblogging are extracted according to the method for S1, are used
The construction method in the user individual microblogging library of S13, chooses microblogging of the microblogging as user-customized recommended of TopN.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910759011.0A CN110489665B (en) | 2019-08-16 | 2019-08-16 | Microblog personalized recommendation method based on scene modeling and convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910759011.0A CN110489665B (en) | 2019-08-16 | 2019-08-16 | Microblog personalized recommendation method based on scene modeling and convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110489665A true CN110489665A (en) | 2019-11-22 |
| CN110489665B CN110489665B (en) | 2023-11-14 |
Family
ID=68551449
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910759011.0A Active CN110489665B (en) | 2019-08-16 | 2019-08-16 | Microblog personalized recommendation method based on scene modeling and convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110489665B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111651672A (en) * | 2020-05-27 | 2020-09-11 | 福建博思软件股份有限公司 | Time-interval user activity recommendation method and device based on deep learning |
| CN112734519A (en) * | 2021-01-06 | 2021-04-30 | 河北工业大学 | Commodity recommendation method based on convolution self-encoder network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150261773A1 (en) * | 2012-07-04 | 2015-09-17 | Qatar Foundation | System and Method for Automatic Generation of Information-Rich Content from Multiple Microblogs, Each Microblog Containing Only Sparse Information |
| CN105183833A (en) * | 2015-08-31 | 2015-12-23 | 天津大学 | User model based microblogging text recommendation method and recommendation apparatus thereof |
| CN105740366A (en) * | 2016-01-26 | 2016-07-06 | 哈尔滨工业大学深圳研究生院 | Inference method and device of MicroBlog user interests |
| CN106776881A (en) * | 2016-11-28 | 2017-05-31 | 中国科学院软件研究所 | A kind of realm information commending system and method based on microblog |
-
2019
- 2019-08-16 CN CN201910759011.0A patent/CN110489665B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150261773A1 (en) * | 2012-07-04 | 2015-09-17 | Qatar Foundation | System and Method for Automatic Generation of Information-Rich Content from Multiple Microblogs, Each Microblog Containing Only Sparse Information |
| CN105183833A (en) * | 2015-08-31 | 2015-12-23 | 天津大学 | User model based microblogging text recommendation method and recommendation apparatus thereof |
| CN105740366A (en) * | 2016-01-26 | 2016-07-06 | 哈尔滨工业大学深圳研究生院 | Inference method and device of MicroBlog user interests |
| CN106776881A (en) * | 2016-11-28 | 2017-05-31 | 中国科学院软件研究所 | A kind of realm information commending system and method based on microblog |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111651672A (en) * | 2020-05-27 | 2020-09-11 | 福建博思软件股份有限公司 | Time-interval user activity recommendation method and device based on deep learning |
| CN112734519A (en) * | 2021-01-06 | 2021-04-30 | 河北工业大学 | Commodity recommendation method based on convolution self-encoder network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110489665B (en) | 2023-11-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tso-Sutter et al. | Tag-aware recommender systems by fusion of collaborative filtering algorithms | |
| Khrouf et al. | Hybrid event recommendation using linked data and user diversity | |
| CN103823844A (en) | Question forwarding system and question forwarding method on the basis of subjective and objective context and in community question-and-answer service | |
| CN107562947A (en) | A kind of Mobile Space-time perceives the lower dynamic method for establishing model of recommendation service immediately | |
| CN112598438A (en) | Outdoor advertisement recommendation system and method based on large-scale user portrait | |
| CN110083764A (en) | A kind of collaborative filtering cold start-up way to solve the problem | |
| CN113806630A (en) | Attention-based multi-view feature fusion cross-domain recommendation method and device | |
| Nguyen et al. | Check out this place: Inferring ambiance from airbnb photos | |
| CN110489665A (en) | A kind of microblogging personalized recommendation method based on scene modeling and convolutional neural networks | |
| Widayat et al. | Bibliometric analysis and visualization articles on presidential election in social media indexed in Scopus by Indonesian authors | |
| Wang et al. | Hybrid variable-scale clustering method for social media marketing on user generated instant music video | |
| KR102225795B1 (en) | Influential Factor Based Hybrid Recommendation System with Deep Neural Network-Based Data Supplement and Method thereof | |
| CN112380455A (en) | Method for directionally and covertly acquiring data of international and foreign internet based on backtracking security controlled network access channel | |
| Gao et al. | Mining tourist preferences and decision support via tourism-oriented knowledge graph | |
| Kim et al. | Analysis of consumer value using semantic network: The comparison of hierarchical and nonhierarchical value structures | |
| Cui et al. | Identification of Micro-blog Opinion Leaders based on User Features and Outbreak Nodes. | |
| Ramos et al. | Analyzing national film based on social media tweets input using topic modelling and data mining approach | |
| Zhang et al. | The Users Emotional Study of Netease Cloud Music Based on LDA Model | |
| Kwon et al. | Sentiment trend analysis in social web environments | |
| Lin et al. | Personalized music recommendation algorithm based on tag information | |
| Chen et al. | Study on collaborative filtering recommendation algorithm based on web user clustering | |
| CN111078888A (en) | Method for automatically classifying comment data of social network users | |
| Li | Text mining model for virtual community user portrait based on social network analysis | |
| Liu et al. | Small Data Fusion Algorithm for Personalized Library Recommendations | |
| Zhenyu | Research on Text and Community Marketing of Little Red Book UGC Platform under Natural Language Processing |
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 |