CN108665311A - A kind of electric business user's time varying characteristic Similarity measures recommendation method based on deep neural network - Google Patents
A kind of electric business user's time varying characteristic Similarity measures recommendation method based on deep neural network Download PDFInfo
- Publication number
- CN108665311A CN108665311A CN201810429983.9A CN201810429983A CN108665311A CN 108665311 A CN108665311 A CN 108665311A CN 201810429983 A CN201810429983 A CN 201810429983A CN 108665311 A CN108665311 A CN 108665311A
- Authority
- CN
- China
- Prior art keywords
- user
- feature
- neural network
- brand
- deep 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
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
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- 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
- 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)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- Development Economics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The electric business user's time varying characteristic Similarity measures and recommend method that the invention discloses a kind of based on deep neural network.The present invention is by analyzing user characteristics, and using neural network model, other users similar with its time of the act evolution Feature are found for target user.On this basis, the present invention further constructs a commending system, and the brand being likely to purchase according to similar users is recommended to be done for the user, improves the diversity of recommendation.
Description
Technical field
The present invention relates to a kind of, and electric business user's time varying characteristic Similarity measures based on deep neural network recommend method, belong to
In software technology field.
Background content
Most of existing goods suggested design is only simple to extract user behavior data, is tied with user's master data
It closes, the buying behavior in user's future is predicted using collaborative filtering.This technology can not find the fine granularity between user
Time-evolution behavior similarity relation more buys possibility to excavate user.
Explanation of nouns：It refers to that his historical behavior data are analyzed for some user to buy prediction model, obtains him
Behavior to some brand and his property feature, then by these feature construction model predictions futures to this brand
Purchase intention model.
Invention content
The present invention overcomes the shortcomings of the prior art, and the invention discloses one kind being based on deep neural network electric business user
Time varying characteristic Similarity measures recommend method.The present invention can use user behavior information, be found and its row for target user
For the similar other users of temporal evolution feature.On this basis, this patent further constructs a commending system, according to phase
Recommend to be done for the user like the brand that user once bought, to improve the diversity of recommendation.
In order to solve the above technical problems, the technical solution adopted in the present invention is：
A kind of electric business user's time varying characteristic Similarity measures recommendation method based on deep neural network, including walk as follows
Suddenly：
Step 1: user's similitude prediction model is established by user behavior characteristics and user personality feature, and to user
Similitude prediction model is trained to obtain trained user's similitude prediction model；
Step 2：Build Collaborative Filtering Recommendation System；
One), according to trained model in step 1, feature extraction is carried out to all user data first, it then, will
The feature of user u and user v input neural network calculate, and the result of neural network output is the purchase phase in their futures
Like the predicted value of degree；For n user, user u carries out similarity prediction with other users respectively, and then sequence is found and used
The highest k user of family u similarities；
Two) acquisition and the higher preceding k user of target user's behavior similarity, had user behavior to the k user
The brand product of characteristic behavior carries out purchase prediction, i.e.,：According to uj, j=1 ..., k, behavioural characteristic and property feature, prediction
ujThe brand being likely to purchase, then the product that k user is likely to purchase are bought to the possibility of product and are ranked up, it is higher to obtain possibility
Preceding m brand, this m brand is put into recommendation list and recommends user u；M is natural number, is set according to contrast experiment's adjustment
It is fixed.
Further to improve, the user behavior characteristics include when clicking tendency feature, collection tendency feature, collection behavior
Between evolution Feature, repeat buying feature, at least one of feature will be bought；The user personality feature includes user's click
Frequecy characteristic, user's collection frequecy characteristic, user's purchase frequency feature, user enliven situation feature, age of user feature, user
At least one of sex character.
It is further to improve, the user behavior characteristics include click tendency feature, feature will be bought, collection behavior is drilled
Change feature；It is special that the user personality feature includes user's purchase frequency feature, user collects frequecy characteristic, user's click frequency
Sign, user's online hours feature, age of user feature, user's sex character.
Further to improve, the value of k is determined jointly with desired recommendation effect by the user volume in data set.General k
Value is not more than n*0.01, and setting is adjusted according to experiment.
Further to improve, the similitude prediction model is that user's similitude based on deep neural network predicts mould
Type, specific establishment step are as follows：
One) feature for being higher than given threshold in user behavior characteristics and user personality feature to purchase predicted impact, is obtained
As selected feature；
Two) it, is based on selected feature, the model of prediction user's similitude is built using deep neural network；Depth nerve net
The input layer of the structure of network inputs selected feature respectively, and each neuron of hidden layer uses phase to the selected feature of each user
Same computational methods obtain the comprehensive characteristics of user respectively；The calculation formula of user comprehensive characteristics σ (z) is as follows：
Wherein z=Σ wixi+b
Wherein, xiFor the corresponding value of i-th of selected feature, wiFor the corresponding weight coefficient of i-th of selected feature；B is each
The corresponding biasing of neuron；E is the nature truth of a matter, and i indicates to select the sequence number of feature；
" output layer is to two user characteristics of hidden layer into the calculating in line (1), i.e. xiPass through for i-th of user defeated
Enter the value that layer is calculated, wiFor the corresponding weight coefficient of i-th of user.The similitude σ of two users is calculated with this
(U1U2)；When training, using the Jaccard similarities that user brand is bought as training fit object, formula is as follows：
Wherein J (u1, u2) indicate user u1With user u2Brand Buying similarity；WithRespectively user u1With
User u2The brand all bought；
Three) model training, is carried out：
Use σ (U1U2) and J (u1, u2) mean square error as loss function, using gradient descent method to neural network into
Row training.So that loss function is dropped in a smaller level as soon as constantly carrying out Gradient Iteration, can determine each choosing at this time
Determine the corresponding weight coefficient w of featureiThe value of biasing b corresponding with each neuron, to obtain trained similitude prediction
Model.
It is further to improve, prediction model is bought by foundation and is obtained in user behavior characteristics and user personality feature to purchase
Predicted impact is bought higher than the feature of given threshold as selected feature；Wherein purchase prediction model is that two sorting machines learn
Model.
Description of the drawings
Fig. 1 is characterized the flow chart of screening；
Fig. 2 is the step schematic diagram of neural computing user's similarity；
Fig. 3 is according to the step schematic diagram that user's similarity is user's Recommendations.
Specific implementation mode
Such as Fig. 1-3, step of the invention is as follows：
One, feature construction
Verification algorithm uses the data set of day cat, extracts 5,000,000 behavior numbers for wherein containing a user more than 30,000
According to.Pass through the analysis to data, it has been found that user behavior in time can be conceptualized as the feature with more expressiveness, knot
Share the essential attribute at family, we define user behavior and user personality of both feature come Brand Buying for user into
Row prediction, and these features are screened, it obtains and the higher feature of similarity contribution is bought to user.It is specific as follows：
1, user behavior characteristics
User behavior characteristics refer to the feature obtained to a series of behaviors of some brand by analyzing user.This patent
It is classified as behavior quantative attribute, time of the act evolution Feature etc..
1.1 behavior quantative attributies
Certain behavior quantity of the counting user on some period first, such as click, collection.Again with specific brand
Behavior number is compared per capita, obtains the tendency quantative attribute of user behavior in the corresponding period.It can be with by these information
Weigh the purchase possibility of user.
1.2 time of the act evolution Features
In addition to the influence that behavior quantity buys user, the time of origin of behavior also has to the purchase of user certain related
Property.Such as user is in the commodity for having collected certain brand on November 10, then its double ten once buying the possibility of this commodity than this
User collected this commodity higher November 1.Therefore, we can establish the pass at collection time point and user's purchase conversion ratio
System, formula are as follows：
Wherein, RminIt is minimum conversion ratio, TPIt is purchase forecast date, TbIt it is time when behavior record occurs, k is to adjust
The value of parameter, k ∈ [0,1], k will be debugged by specific scene, ensure the collect at a time pointconversionEnergy
Enough conversion ratios for being stowed to purchase for preferably describing user.
Since user may have a plurality of collection behavior to brand, it is therefore desirable to which each behavior is calculated
collectconversionIt adds up, obtains the synthesis evolution Feature of the collection behavior in some period, i.e.,：
collectevolving=∑ collectconversion#2
Wherein, collectcoversionUser on the specific time point obtained for formula 1 collects conversion ratio.
collectevolvingThe overall conversion feature of collection behavior in some period.
1.3 user's special behaviors --- buying behavior correlated characteristic
Finally, due to which user may buy the commodity with brand, and buying behavior pair before the time point of prediction
For user, there is the expense (clicking, collection cost free) in cost, therefore we need to buying behavior independent analysis.
By analysis, we have two such discovery：User goes over the brand that repeat buying is crossed, and user is in future purchase
Possibility is higher；In addition, user is more likely to disposably buy all preferences of this brand when browsing the commodity of some brand
Commodity, therefore, within a short period of time, user's maximum probability will not buy this brand again.We can obtain following two spies as a result,
Sign：Repeat buying feature and feature will be bought.The user of some brand of repeat buying has often established ratio to brand
Preferable understanding, the possibility bought again are larger；And due to net purchase environment, user generally will not be right in a short time
The result of brand repeat buying, data analysis shows Most users in repeat buying apart from upper primary purchase 10 days or so.
2, user personality feature
User personality feature is divided into two classes, respectively behavioral trait feature and Demographics.They are by dividing
Analyse whole related datas of user, the feature that the data such as usage behavior frequency, age of user obtain.
2.1 behavioral trait features
User behavior frequency refers in timing statistics section, and the three behaviors (click, collect and buy) of user account for institute respectively
There is the proportion of behavior number, can reflect the active degree of user's difference behavior.
User's enlivens situation.User has the ratio of the time and entire time segment length of behavior, can reflect user
Overall active degree.
2.2 Demographics
Demographics includes age and the gender of user.Data analysis the results show that age, the gender of user
All may the buying behavior of user have an impact, as the purchasing power of women of the age between 20-30 may be stronger.
According to foregoing description, the feature used is as shown in table 1：
The whole features of table 1
Two, it predicts to carry out feature selecting using purchase
Front has analyzed two major classes to the influential user characteristics of purchase, but the influence degree of these features is respectively not
It is identical.Therefore it needs to carry out importance ranking to features described above, finds out and buy the highest feature of relevance to calculating user brand.
The purchase that this patent carries out user using Logic Regression Models predicts that (purchase is predicted：Whether user can be bought in future
Some brand regards two classification problems as.Then user's history data are analyzed, the system that may be influenced on this problem is obtained
Row feature trains a machine learning model using feature, and whether this model, which can buy future, carries out classification prediction.Together
When, trained machine learning model by way of transferring correlation technique, can obtain the weight system of each feature of input
Number, the bigger feature of weight coefficient is more important, therefore just obtains the sequence of feature) experiment, the weight of each feature of purchase evaluation
The property wanted therefrom filters out most important 9 features, the selected feature as neural network.
The larger problem of user data generally existing sparsity, in order to avoid this problem, this patent uses degree of rarefication phase
Data are bought as prediction data to smaller advertising campaign (i.e. double 11) user, carries out user and buys prediction.According to training
Good prediction model obtains the importance of wherein different characteristic, is ranked up.Sequence is as follows (more top in table
Importance it is higher)：
2 feature ordering of table
| User personality feature | User behavior characteristics |
| User's purchase frequency feature | User clicks tendency feature |
| User collects frequecy characteristic | Feature will be bought |
| User's click frequency feature | Collect time of the act evolution Feature |
| User's online hours feature | / |
| Age of user feature | / |
| User's sex character | / |
9 features obtained in table 2, by the input as next step neural network.
Three, deep neural network model is built
Neural network model has protrusion excellent on the field that the data dimensions such as image and text are high, data redundancy information is more
Gesture, reason are that neural network has outstanding data " purification ", abstract ability.
So this patent uses previously obtained feature, it is proposed that when a kind of user using deep neural network (DNN)
Become similarity calculation method, for target user's positioning and the most like user of its action evolution feature.The model is by largely counting
According to training, importance of the different user feature on Brand Buying more can be meticulously captured than traditional simple collaborative filtering.Cause
This, the brand commending system based on this method design also can more make the stronger recommendation of diversity.
1, feature selecting
It is characterized in weighing importance by purchase prediction listed by table 2, due to the essence of this patent user's similarity calculation
It is the similitude for weighing user in purchase intention, therefore this patent carries out user's similarity calculation using same feature.
When calculating user's similarity, the quantity n for the brand for considering that two user's cooperating contacts are crossed, each brand is needed to correspond to
Above-mentioned three behaviors feature (i.e. user clicks tendency feature, will buy feature and collection time of the act evolution Feature).By
Different in the trademark quantity that user contacted, to ensure that the time of model training is unlikely to long, the value of n is needed according to tool
Volume data is judged.By data analysis, there is most users and purchases jointly in the brand number that counting user is bought jointly, selection
Buy value of the brand number as n.After the value that n is determined, when the brand that the cooperating contact of two users is crossed is less than n, with 0 completion
Correlated characteristic.
2, similarity model framework
Based on features described above, the model of prediction user's similitude is built using deep neural network.The structure of neural network
Such as Fig. 2.Wherein bottom one layer is input layer, using above-mentioned feature as input；Intermediate one layer is hidden layer, uses two god
Through member to user u1With user u2Feature carry out conformity calculation；Last layer is output layer, for calculating the similar of two users
Degree.
Each neuron of hidden layer uses same computational methods, and user characteristics σ (z) calculation formula of hidden layer are such as
Under：
Wherein z=Σ wixi+b
Wherein, xiFor the corresponding value of i-th of behavioural characteristic, wiFor the corresponding weight coefficient of i-th of behavioural characteristic；B is inclined
It sets；E is the nature truth of a matter.
Output layer does two user characteristics obtained above and similarity σ (U similarly is calculated1U2).When training, phase
The purchase data using user are calculated like degree, carry out the Jaccard similarity calculations of user, formula is as follows：
Wherein J (u1, u2) indicate user u1With user u2Similarity；WithRespectively user u1With user u2All purchase
The brand bought.
This patent uses σ (U1U2) and J (u1, u2) mean square error as loss function, using gradient descent method to nerve
Network is trained.Weight coefficient w in hidden layer neuron is determined in trainingiWith the value of biasing b, the two parameters determine
Afterwards, this neural network model has been determined that.Training process is illustrated：There is u1And u2Two users, existing u1And u2In the number of the 2-3 months
According to using this data prediction u1And u2Similitude in the purchase in April.User u can be used at this time1And u2In 2 months data
As input, training objective is purchase similitude of two users in March.We carry out iteration in training of judgement using mean square error
Obtained two users in March buy the similitude of two users in similitude and truthful data, after the certain number of iteration, observe this
When square mean error amount, the deconditioning when meeting the requirements just obtained weight coefficient w in hidden layer neuroniWith biasing b's
Value.So far, we have just obtained trained neural network model.
3, Collaborative Filtering Recommendation System is built
To a target user u, all customer data is carried out feature extraction by this patent first.Then, by the spy of user u
The feature of other users v of seeking peace inputs trained neural network model and is predicted, can obtain their similitude；
Similarly, the similitude for calculating user u and other all users successively, is then ranked up by similitude, to build recommendation list
It prepares.
For the structure of recommendation list, the first step needs to estimate to be that user u selects how many a similar users, is set as k.Its
It is secondary, it is thus necessary to determine that each user had the possibility that the brand of behavior is bought by the user in selected k user, it would be possible to property compared with
Big brand is put into recommendation list and recommends user u, to increase the diversity of recommendation, as shown in Figure 3.
The value of k is determined jointly with desired recommendation effect by the user volume in data set.General k value is not more than n*
0.01, setting is adjusted according to experiment.The data set that this patent has been 30000 or so using a number of users is tested,
By experiment, the value of k is determined 20.
Above-described embodiment is only the specific embodiment of the present invention, is also existed to its simple transformation, replacement etc.
In the protection domain of invention.
Claims (6)
1. a kind of electric business user's time varying characteristic Similarity measures based on deep neural network recommend method, which is characterized in that packet
Include following steps：
Step 1: user's similitude prediction model is established by user behavior characteristics and user personality feature, and it is similar to user
Property prediction model is trained to obtain trained user's similitude prediction model；
Step 2：Build Collaborative Filtering Recommendation System；
One), according to trained model in step 1, feature extraction is carried out to all user data first, then, by user
The feature of u and user v input neural network calculate, and the result of neural network output is the purchase similarity in their futures
Predicted value；For n user, user u carries out similarity prediction with other users respectively, and then sequence is found and user u
The highest k user of similarity；
Two) acquisition and the higher preceding k user of target user's behavior similarity, had user behavior characteristics to the k user
The brand product of behavior carries out purchase prediction, i.e.,：According to uj, j=1 ..., k, behavioural characteristic and property feature, predict ujIt can
Can purchase brand, then the product that k user is likely to purchase are bought to the possibility of product and are ranked up, before acquisition possibility is higher
This m brand is put into recommendation list and recommends user u by m brand；M is natural number, is adjusted and is set according to contrast experiment.
2. electric business user's time varying characteristic Similarity measures based on deep neural network recommend method as described in claim 1,
It is inclined to feature, collection tendency feature, collection time of the act evolution spy it is characterized in that, the user behavior characteristics include click
Sign, will buy at least one of feature at repeat buying feature；The user personality feature includes that user's click frequency is special
Sign, user's collection frequecy characteristic, user's purchase frequency feature, user enliven situation feature, age of user feature, user gender spy
At least one of sign.
3. electric business user's time varying characteristic Similarity measures based on deep neural network recommend method as claimed in claim 2,
It is characterized in that, the user behavior characteristics include clicking tendency feature, will buying feature, collection behavior evolution feature；Institute
It includes that collect frequecy characteristic, user's click frequency feature, user online by user's purchase frequency feature, user to state user personality feature
Duration characteristics, age of user feature, user's sex character.
4. electric business user's time varying characteristic Similarity measures based on deep neural network recommend method as described in claim 1,
It is characterized in that, the value of k is not more than n*0.01.
5. electric business user's time varying characteristic Similarity measures based on deep neural network recommend method as described in claim 1,
It is characterized in that, the similitude prediction model is user's similitude prediction model based on deep neural network, it is specific to establish
Steps are as follows：
One) the feature conduct for being higher than given threshold in user behavior characteristics and user personality feature to purchase predicted impact, is obtained
Selected feature；
Two) it, is based on selected feature, the model of prediction user's similitude is built using deep neural network；Deep neural network
The input layer of structure inputs selected feature respectively, and each neuron of hidden layer uses the selected feature of each user identical
Computational methods obtain the comprehensive characteristics of user respectively；The calculation formula of user comprehensive characteristics σ (z) is as follows：
Wherein z=∑s wixi+b (1)
Wherein, xiFor the corresponding value of i-th of selected feature, wiFor the corresponding weight coefficient of i-th of selected feature；B is each nerve
The corresponding biasing of member；E is the nature truth of a matter, and i indicates to select the sequence number of feature；
" output layer is to two user characteristics of hidden layer into the calculating in line (1), i.e. xiPass through input layer meter for i-th of user
Obtained value, wiFor the corresponding weight coefficient of i-th of user.Similitude σ (the U of two users are calculated with this1 U2)；Instruction
When practicing, using the Jaccard similarities that user brand is bought as training fit object, formula is as follows：
Wherein J (u1, u2) indicate user u1With user u2Brand Buying similarity；WithRespectively user u1With user u2
The brand all bought；
Three) model training, is carried out：
Use σ (U1U2) and J (u1, u2) mean square error as loss function, neural network is instructed using gradient descent method
Practice.So that loss function is dropped in a smaller level as soon as constantly carrying out Gradient Iteration, can determine each selected spy at this time
Levy corresponding weight coefficient wiThe value of biasing b corresponding with each neuron, to obtain trained similitude prediction model.
6. electric business user's time varying characteristic Similarity measures based on deep neural network recommend method as claimed in claim 5,
It is obtained in user behavior characteristics and user personality feature it is characterized in that, buying prediction model by foundation to buying predicted impact
Feature is selected in the conduct high compared with other features；Wherein buying prediction model is：Whether user can be bought into some brand in future
Regard two classification problems as.Then user's history data are analyzed, the series of features that may be influenced on this problem is obtained, makes
A machine learning model is trained with feature, whether this model, which can buy future, carries out classification prediction.Meanwhile it training
Machine learning model by way of transferring correlation technique, can obtain input each feature weight coefficient, weight system
The bigger feature of number is more important, therefore just obtains the sequence of feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810429983.9A CN108665311B (en) | 2018-05-08 | 2018-05-08 | Electric commercial user time-varying feature similarity calculation recommendation method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810429983.9A CN108665311B (en) | 2018-05-08 | 2018-05-08 | Electric commercial user time-varying feature similarity calculation recommendation method based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108665311A true CN108665311A (en) | 2018-10-16 |
| CN108665311B CN108665311B (en) | 2022-02-25 |
Family
ID=63778804
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810429983.9A Active CN108665311B (en) | 2018-05-08 | 2018-05-08 | Electric commercial user time-varying feature similarity calculation recommendation method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108665311B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109547814A (en) * | 2018-12-13 | 2019-03-29 | 北京达佳互联信息技术有限公司 | Video recommendation method, device, server and storage medium |
| CN109784979A (en) * | 2018-12-19 | 2019-05-21 | 重庆邮电大学 | A kind of supply chain needing forecasting method of big data driving |
| CN109800325A (en) * | 2018-12-26 | 2019-05-24 | 北京达佳互联信息技术有限公司 | Video recommendation method, device and computer readable storage medium |
| CN109801100A (en) * | 2018-12-26 | 2019-05-24 | 北京达佳互联信息技术有限公司 | Advertisement placement method, device and computer readable storage medium |
| CN110930203A (en) * | 2020-02-17 | 2020-03-27 | 京东数字科技控股有限公司 | Information recommendation model training method and device and information recommendation method and device |
| CN111652664A (en) * | 2019-03-04 | 2020-09-11 | 富士通株式会社 | Apparatus and method for training mixed element learning network |
| CN112269937A (en) * | 2020-11-16 | 2021-01-26 | 加和（北京）信息科技有限公司 | Method, system and device for calculating user similarity |
| CN112819570A (en) * | 2021-01-21 | 2021-05-18 | 东北大学 | Intelligent commodity collocation recommendation method based on machine learning |
| CN118071468A (en) * | 2024-04-25 | 2024-05-24 | 沈阳曼得科技有限公司 | Intelligent marketing system and method based on quick-service product industry |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631535A (en) * | 2015-12-17 | 2016-06-01 | 东软集团股份有限公司 | Method and device for predicting scoring data |
| CN106022865A (en) * | 2016-05-10 | 2016-10-12 | 江苏大学 | Goods recommendation method based on scores and user behaviors |
| US20170270593A1 (en) * | 2016-03-21 | 2017-09-21 | The Procter & Gamble Company | Systems and Methods For Providing Customized Product Recommendations |
| CN107577782A (en) * | 2017-09-14 | 2018-01-12 | 国家计算机网络与信息安全管理中心 | A kind of people-similarity depicting method based on heterogeneous data |
| CN107808278A (en) * | 2017-10-11 | 2018-03-16 | 河海大学 | A kind of Github open source projects based on sparse self-encoding encoder recommend method |
-
2018
- 2018-05-08 CN CN201810429983.9A patent/CN108665311B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631535A (en) * | 2015-12-17 | 2016-06-01 | 东软集团股份有限公司 | Method and device for predicting scoring data |
| US20170270593A1 (en) * | 2016-03-21 | 2017-09-21 | The Procter & Gamble Company | Systems and Methods For Providing Customized Product Recommendations |
| CN106022865A (en) * | 2016-05-10 | 2016-10-12 | 江苏大学 | Goods recommendation method based on scores and user behaviors |
| CN107577782A (en) * | 2017-09-14 | 2018-01-12 | 国家计算机网络与信息安全管理中心 | A kind of people-similarity depicting method based on heterogeneous data |
| CN107808278A (en) * | 2017-10-11 | 2018-03-16 | 河海大学 | A kind of Github open source projects based on sparse self-encoding encoder recommend method |
Non-Patent Citations (2)
| Title |
|---|
| 张宁 等: ""一种基于RFM模型的新型协同过滤个性化推荐算法"", 《电信科学》 * |
| 武玲梅 等: ""基于降噪自动编码器的推荐算法"", 《计算机与现代化》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109547814A (en) * | 2018-12-13 | 2019-03-29 | 北京达佳互联信息技术有限公司 | Video recommendation method, device, server and storage medium |
| CN109784979A (en) * | 2018-12-19 | 2019-05-21 | 重庆邮电大学 | A kind of supply chain needing forecasting method of big data driving |
| CN109800325A (en) * | 2018-12-26 | 2019-05-24 | 北京达佳互联信息技术有限公司 | Video recommendation method, device and computer readable storage medium |
| CN109801100A (en) * | 2018-12-26 | 2019-05-24 | 北京达佳互联信息技术有限公司 | Advertisement placement method, device and computer readable storage medium |
| CN109800325B (en) * | 2018-12-26 | 2021-10-26 | 北京达佳互联信息技术有限公司 | Video recommendation method and device and computer-readable storage medium |
| CN111652664A (en) * | 2019-03-04 | 2020-09-11 | 富士通株式会社 | Apparatus and method for training mixed element learning network |
| CN110930203A (en) * | 2020-02-17 | 2020-03-27 | 京东数字科技控股有限公司 | Information recommendation model training method and device and information recommendation method and device |
| CN112269937A (en) * | 2020-11-16 | 2021-01-26 | 加和（北京）信息科技有限公司 | Method, system and device for calculating user similarity |
| CN112819570A (en) * | 2021-01-21 | 2021-05-18 | 东北大学 | Intelligent commodity collocation recommendation method based on machine learning |
| CN112819570B (en) * | 2021-01-21 | 2023-09-26 | 东北大学 | Intelligent commodity collocation recommendation method based on machine learning |
| CN118071468A (en) * | 2024-04-25 | 2024-05-24 | 沈阳曼得科技有限公司 | Intelligent marketing system and method based on quick-service product industry |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108665311B (en) | 2022-02-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108665311A (en) | A kind of electric business user's time varying characteristic Similarity measures recommendation method based on deep neural network | |
| Chen et al. | Learning to rank features for recommendation over multiple categories | |
| Gao et al. | Identifying preferred management options: An integrated agent-based recreational fishing simulation model with an AHP-TOPSIS evaluation method | |
| CN109241424A (en) | A kind of recommended method | |
| CN110427560A (en) | A kind of model training method and relevant apparatus applied to recommender system | |
| CN111695042B (en) | User behavior prediction method and system based on deep walking and ensemble learning | |
| CN109034960B (en) | Multi-attribute inference method based on user node embedding | |
| CN106897911A (en) | A kind of self adaptation personalized recommendation method based on user and article | |
| CN109544197A (en) | A kind of customer churn prediction technique and device | |
| CN117829914B (en) | Digital media advertisement effect evaluation system | |
| Qin et al. | Comprehensive helpfulness of online reviews: A dynamic strategy for ranking reviews by intrinsic and extrinsic helpfulness | |
| Frischknecht et al. | A simple method for estimating preference parameters for individuals | |
| CN115204985A (en) | Shopping behavior prediction method, device, equipment and storage medium | |
| CN110110372A (en) | A kind of user's timing behavior automatic segmentation prediction technique | |
| Rahmani Seryasat et al. | Predicting the number of comments on Facebook posts using an ensemble regression model | |
| KR102609681B1 (en) | Method for determining product planning reflecting user feedback and Apparatus thereof | |
| Wang et al. | Computer supported data-driven decisions for service personalization: a variable-scale clustering method | |
| CN115115403A (en) | Method and device for classifying customers in target customer group, electronic equipment and storage medium | |
| CN115187312A (en) | Customer loss prediction method and system based on deep learning | |
| Masoudi et al. | The effect of web interface features on consumer online purchase intentions | |
| Trithipkaiwanpon et al. | Sensitivity Analysis of Random Forest Hyperparameters | |
| Prihatmono et al. | Application of the KNN Algorithm for Predicting Data Card Sales at PT. XL Axiata Makassar | |
| Wenji | User Interest Discovery and Prediction Service Model in E-commerce Recommendation | |
| Jiang | Prediction of Consumer Behavior Based on Machine Learning Algorithm | |
| Zheng et al. | Context-dependent self-exciting point processes: models, methods, and risk bounds in high dimensions |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Jiang Wenjun Inventor after: Dong Yunqi Inventor before: Jiang Wenjun |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |