CN109815309A - A kind of user information recommended method and system based on personalization - Google Patents
A kind of user information recommended method and system based on personalization Download PDFInfo
- Publication number
- CN109815309A CN109815309A CN201811574681.7A CN201811574681A CN109815309A CN 109815309 A CN109815309 A CN 109815309A CN 201811574681 A CN201811574681 A CN 201811574681A CN 109815309 A CN109815309 A CN 109815309A
- Authority
- CN
- China
- Prior art keywords
- information
- user
- neural network
- algorithm
- recommendation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000013528 artificial neural network Methods 0.000 claims abstract description 130
- 238000004458 analytical method Methods 0.000 claims abstract description 22
- 238000005457 optimization Methods 0.000 claims abstract description 9
- 230000000306 recurrent effect Effects 0.000 claims description 57
- 238000010276 construction Methods 0.000 claims description 17
- 238000013517 stratification Methods 0.000 claims description 13
- 230000002708 enhancing effect Effects 0.000 claims description 7
- 230000003313 weakening effect Effects 0.000 claims description 4
- 241000208340 Araliaceae Species 0.000 claims description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 2
- 235000008434 ginseng Nutrition 0.000 claims description 2
- 230000006870 function Effects 0.000 description 32
- 238000007726 management method Methods 0.000 description 12
- 238000013461 design Methods 0.000 description 11
- 238000005516 engineering process Methods 0.000 description 11
- 238000011160 research Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000001965 increasing effect Effects 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000007704 transition Effects 0.000 description 3
- 239000002360 explosive Substances 0.000 description 2
- 210000004218 nerve net Anatomy 0.000 description 2
- 230000007115 recruitment Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of based on personalized user information recommended method, which comprises scoring of the information and user that acquisition user has focused on to information of interest constructs the sequence data of a plurality of information and multiple scorings of user's mistake of interest；The sequence data is subjected to calculating analysis by neural network algorithm, constructs the neural network recommendation algorithm of user information；Information in Extracting Information archives calculates the information using neural network recommendation algorithm, using the highest information of probability as user's recommendation information；According to user to the feedback of user's recommendation information, the parameter of the neural network recommendation algorithm is adjusted, the neural network recommendation algorithm after obtaining optimization.
Description
Technical field
The present invention relates to technical field of information processing, are recommended more particularly, to a kind of based on personalized user information
Method and system.
Background technique
Archival Informationization is to be believed under the unified planning and system organization of National archives administrative service division using the modern times
Technological transformation traditional archive business is ceased, the movable development and change of digital environment Archives is constantly adapted to, meets society to the maximum extent
The process of construction of meeting archives information requirement, it is the organic component of national information system, is that national information strategy exists
The specific implementation in archives field, construction content very abundant simultaneously constantly change, and have epochal character and social.
With the fast development of Internet technology, the explosive growth of data, file data information construction has welcome big number
According to the epoch, files in China informatization is obviously accelerated, and the ratio of archives Digital Archives Resources at different levels greatly improves, entirely
State's digital archive resource is obviously accelerated up to 22,430,000 GB, national archival digitalization process.In recent years, with network and information skill
Art is fast-developing, and archives work is also facing a deep reform.In face of magnanimity file data, although domestic Archival Informationization water
It is flat to step up, but Now Domestic Market Review class management software is irregular, and most of rest on traditional management, clear
It lookes at, query function, and as information technology is universal, user's more and more diversified, intelligence to file administration and using the demand of service
Energyization, standard also step up, and traditional archive service and traditional search engines have been insufficient for the demand of user.It realizes
Archives user personalized service, is the developing direction of big data era archives certainty, and realizes archives personalized service, most closes
Key be the obtained information content of user personalization.Therefore, the digital archives management system of construction personalized recommendation engine is
Construct the core and trend of personalized file data service.
Archives work is still within the situation based on manual management profile entity, and informatization is slow, sends out between area
It opens up extremely unbalanced.Reinforce the rising as the work of Guangxi Information Construction of Archive such as electronic records filing and electronic records management work
Step gradually carries out Information Construction of Archive work.
Therefore, archival digitalization, informationization, intelligentized study on construction are extremely urgent, carry out personalized recommendation engine
Digital archives management system research, foundation meet user to the technological system and service mechanism of file data individual demand, incite somebody to action
Various limitations of traditional archive service are broken through, the value of Archive Resource is provided to the maximum extent, promote files department to active
Service change is accurately classified and is positioned to user using modern technologies, has comprehensive, accurate assurance to its demand,
Attract user to greatest extent, provide the service of high-efficiency high-quality for society, promote the development of archives undertaking, simultaneously for promotion shelves
Case digitlization process, the transition of quickening archives work and archives scientific research work sutdy have great importance.
Therefore, it is necessary to a kind of technologies, to realize based on personalized user information recommended technology.
Summary of the invention
Technical solution of the present invention provide it is a kind of based on personalized user information recommended method and system, with solve how
Based on personalization, the problem of recommendation user information.
To solve the above-mentioned problems, the present invention provides a kind of based on personalized user information recommended method, the side
Method includes:
Scoring of the information and user that acquisition user has focused on to information of interest, building user's mistake of interest
The sequence data of a plurality of information and multiple scorings；
The sequence data is subjected to calculating analysis by neural network algorithm, constructs the neural network recommendation of user information
Algorithm；
Information in Extracting Information archives calculates the information using neural network recommendation algorithm, most by probability
High information is as user's recommendation information；
According to user to the feedback of user's recommendation information, the parameter of the neural network recommendation algorithm is adjusted
Whole, after obtaining optimization neural network recommendation algorithm.
Preferably, the neural network algorithm is Recognition with Recurrent Neural Network algorithm.
Preferably, described that the sequence data is subjected to calculating analysis by neural network algorithm, construct user information
Neural network recommendation algorithm, comprising:
The sequence data is input to the input layer of the Recognition with Recurrent Neural Network algorithm；
The hidden layer of the Recognition with Recurrent Neural Network algorithm is designed, by the hidden layer of the Recognition with Recurrent Neural Network algorithm
The input value of the latter number of plies is added in the output valve of the hidden layer preceding layer of the Recognition with Recurrent Neural Network algorithm；
The sequence data is divided into forward data and negative sense data, the forward data is for enhancing Generalization bounds；Institute
Negative sense data are stated for weakening Generalization bounds.
Preferably, described that the sequence data is subjected to calculating analysis by neural network algorithm, construct user information
Neural network recommendation algorithm, comprising:
Door and input gate are forgotten in setting in the hidden layer of the Recognition with Recurrent Neural Network algorithm, and the forgetting door is used for basis
Scoring of the user to the information having focused on retains the information that the scoring is higher than first threshold；The input gate passes through
Threshold function table sigmoid function layer calculates, the scoring according to user to information, determines the quantity that information is added；
Output layer is constructed using stratification exponential function Softmax function, the Recognition with Recurrent Neural Network algorithm carries out level
Change exponential function Softmax function.
Preferably, the sequence data of a plurality of information and multiple scorings of building user's mistake of interest, the sequence
Column data includes:
Message file number, user's scoring, archives kind.
Based on another aspect of the present invention, provide a kind of based on personalized user information recommender system, the system packet
It includes:
Acquiring unit, for obtaining the scoring of information and user that user has focused on to information of interest, building is used
The sequence data of a plurality of information and multiple scorings of family mistake of interest；
Construction unit constructs user information for the sequence data to be carried out calculating analysis by neural network algorithm
Neural network recommendation algorithm；
Computing unit carries out the information using neural network recommendation algorithm for the information in Extracting Information archives
It calculates, using the highest information of probability as user's recommendation information；
Optimize unit, for the feedback according to user to user's recommendation information, to the neural network recommendation algorithm
Parameter be adjusted, obtain optimization after neural network recommendation algorithm.
Preferably, the neural network algorithm is Recognition with Recurrent Neural Network algorithm.
Preferably, the construction unit is used for: the sequence data is carried out calculating analysis, structure by neural network algorithm
Build the neural network recommendation algorithm of user information, comprising:
The sequence data is input to the input layer of the Recognition with Recurrent Neural Network algorithm；
The hidden layer of the Recognition with Recurrent Neural Network algorithm is designed, by the hidden layer of the Recognition with Recurrent Neural Network algorithm
The input value of the latter number of plies is added in the output valve of the hidden layer preceding layer of the Recognition with Recurrent Neural Network algorithm；
The sequence data is divided into forward data and negative sense data, the forward data is for enhancing Generalization bounds；Institute
Negative sense data are stated for weakening Generalization bounds.
Preferably, the construction unit is used for: the sequence data is carried out calculating analysis, structure by neural network algorithm
Build the neural network recommendation algorithm of user information, comprising:
Door and input gate are forgotten in setting in the hidden layer of the Recognition with Recurrent Neural Network algorithm, and the forgetting door is used for basis
Scoring of the user to the information having focused on retains the information that the scoring is higher than first threshold；The input gate passes through
Threshold function table sigmoid function layer calculates, the scoring according to user to information, determines the quantity that information is added；
Output layer is constructed using stratification exponential function Softmax function, the Recognition with Recurrent Neural Network algorithm carries out level
Change exponential function Softmax function.
Preferably, the acquiring unit is used for: the sequence of a plurality of information and multiple scorings of building user's mistake of interest
Column data, the sequence data include: message file number, user's scoring, archives kind.
Technical solution of the present invention provides a kind of user information recommended method and system based on personalization, wherein method packet
Include: scoring of the information and user that acquisition user has focused on to information of interest constructs a plurality of letter of user's mistake of interest
The sequence data of breath and multiple scorings；Sequence data is subjected to calculating analysis by neural network algorithm, constructs user information
Neural network recommendation algorithm；Information in Extracting Information archives calculates information using neural network recommendation algorithm, will be general
The highest information of rate is as user's recommendation information；According to the feedback of user to user recommendation information, to neural network recommendation algorithm
Parameter be adjusted, obtain optimization after neural network recommendation algorithm.Technical solution of the present invention be intended to make full use of cloud computing,
The generation information technologies tool such as big data, artificial intelligence, in conjunction with the original archive management system of company and all kinds of file datas
Resource studies the AI personalized recommendation algorithm for being suitble to digital archive system, AI personalized recommendation system platform is developed, from archives
The information for meeting users ' individualized requirement is extracted in big data, realizes that artificial intelligence technology is mutually tied with file data information management
It closes, its application field is extended into all trades and professions, push archives research work and archive management system transition.
Detailed description of the invention
By reference to the following drawings, exemplary embodiments of the present invention can be more fully understood by:
Fig. 1 is according to a kind of based on personalized user information recommended method flow chart of the preferred embodiment for the present invention；
Fig. 2 is the Recognition with Recurrent Neural Network algorithm flow chart according to the preferred embodiment for the present invention；
Fig. 3 is the Recognition with Recurrent Neural Network algorithm hidden layer structure chart according to the preferred embodiment for the present invention；
Fig. 4 is the Recognition with Recurrent Neural Network algorithm hidden layer algorithm flow chart according to the preferred embodiment for the present invention；And
Fig. 5 is according to a kind of based on personalized user information recommender system structure chart of the preferred embodiment for the present invention.
Specific embodiment
Exemplary embodiments of the present invention are introduced referring now to the drawings, however, the present invention can use many different shapes
Formula is implemented, and is not limited to the embodiment described herein, and to provide these embodiments be at large and fully disclose
The present invention, and the scope of the present invention is sufficiently conveyed to person of ordinary skill in the field.Show for what is be illustrated in the accompanying drawings
Term in example property embodiment is not limitation of the invention.In the accompanying drawings, identical cells/elements use identical attached
Icon note.
Unless otherwise indicated, term (including scientific and technical terminology) used herein has person of ordinary skill in the field
It is common to understand meaning.Further it will be understood that with the term that usually used dictionary limits, should be understood as and its
The context of related fields has consistent meaning, and is not construed as Utopian or too formal meaning.
Fig. 1 is according to a kind of based on personalized user information recommended method flow chart of the preferred embodiment for the present invention.
Is difficult to file retrieval in industry, the application embodiment makes full use of cloud computing, big intelligent the problem of reaching purpose user
The generation information technologies tool such as data, artificial intelligence is ground in conjunction with original archive management system and all kinds of file data resources
Study carefully the AI personalized recommendation algorithm of suitable digital archive system, AI personalized recommendation system platform is developed, from archives big data
It is middle to extract the information for meeting users ' individualized requirement, realize that artificial intelligence technology is combined with file data information management, by it
Application field extends to all trades and professions, pushes archives research work and archive management system transition.Intelligent product when
In generation, does not mean only that product has very high " data " and " information " feature, enterprise is more required to need the angle from user,
Using the demand of user as the highest standard of intelligence manufacture, more personalized service is provided for user.The application is based on user's sheet
The AI personalized recommendation algorithm of body constructs AI personalized recommendation system, analyzes user data and its to file data information
Property demand, realize file data intelligently reach target user, provide meet demand, the information of quick response for user,
Both customer personalized service had been met, the maximizing the benefits of resource is also achieved.
As shown in Figure 1, embodiment of the present invention provides a kind of user information recommended method based on personalization, method packet
It includes:
Preferably, in step 101: scoring of the information and user that acquisition user has focused on to information of interest, structure
Build a plurality of information of user's mistake of interest and the sequence data of multiple scorings.Preferably, a plurality of letter of user's mistake of interest is constructed
The sequence data of breath and multiple scorings, sequence data include: message file number, user's scoring, archives kind.
Preferably, in step 102: sequence data being carried out calculating analysis by neural network algorithm, constructs user information
Neural network recommendation algorithm.Preferably, neural network algorithm is Recognition with Recurrent Neural Network algorithm.Preferably, sequence data is led to
It crosses neural network algorithm and carries out calculating analysis, construct the neural network recommendation algorithm of user information, comprising:
Sequence data is input to the input layer of Recognition with Recurrent Neural Network algorithm；
The hidden layer of Recognition with Recurrent Neural Network algorithm is designed, by the latter number of plies of the hidden layer of Recognition with Recurrent Neural Network algorithm
Input value be added in the output valve of hidden layer preceding layer of Recognition with Recurrent Neural Network algorithm；
Sequence data is divided into forward data and negative sense data, forward data is for enhancing Generalization bounds；Negative sense data are used
In decrease Generalization bounds.
Preferably, sequence data is subjected to calculating analysis by neural network algorithm, constructs the neural network of user information
Proposed algorithm, comprising:
Door and input gate are forgotten in setting in the hidden layer of Recognition with Recurrent Neural Network algorithm, forget door and are used for according to user to
The scoring of the information of concern retains the information that scoring is higher than first threshold；Input gate passes through threshold function table sigmoid letter
Several layers of calculating, the scoring according to user to information determine the quantity that information is added；
Output layer is constructed using stratification exponential function Softmax function, Recognition with Recurrent Neural Network algorithm carries out stratification and refers to
Number function Softmax function.
Preferably, in step 103: the information in Extracting Information archives carries out information using neural network recommendation algorithm
It calculates, using the highest information of probability as user's recommendation information.
The application embodiment is RNN (Recognition with Recurrent Neural Network) proposed algorithm based on file data information, in face of big number
According to the information explosive growth in epoch, to solve the problems, such as that information overload and file data searching field encounter, for user's sheet
Body individual demand, research are based on the personalized recommendation algorithm of RNN (Recognition with Recurrent Neural Network), according to customer attribute information and to shelves
Personalized archives recommendation information is recommended in the evaluation of case data information, and algorithm basic structure is as shown in fig. 2.
The Recognition with Recurrent Neural Network algorithm that the application proposes is different from other neural network algorithms, RNN (Recognition with Recurrent Neural Network)
In introduce the structure of directed circulation, solve the problems, such as that input information is interrelated, user to the scoring of file data even
Once interested files all recommend file data related to user to active user, can better binding time it is longer
Historical information is predicted that RNN algorithm model is designed and is described as follows come file data information interested in user:
1.1 data input layers
File data information enters data input layer after pretreatment, and the data input layer of proposed algorithm is by user
The dossier information that browsed, user form the score information of file data and the classification information of file data, by this
A little information carry out quantization and are ranked up according to the scoring time, and then are abstracted as one section of sequence data, i.e., the data of user=
{ (files number, user's scoring, archives kind) ..., (files number, user's scoring, archives kind) }, this sequence
Data are the input data of RNN input layer.By being converted into sequence to the pretreatment of data and by subscriber profile data information above
The step of column data, realizes the preliminary classification to file data information and selects, and is the basis that each step calculates below.
The design of 1.2RNN hidden layer
Traditional RNN (Recognition with Recurrent Neural Network) algorithm is designed to processing sequence data, is different from the calculation of other neural networks
The place of method is that the current output of a sequence in RNN is related with the output of front, and the input value of the number of plies below is added to
In the output valve of front layer.There is connection between RNN hidden layer, shown in network structure Fig. 3.
According to above-mentioned RNN Recognition with Recurrent Neural Network structure chart, wherein (W, U, V) indicates a group model parameter, user data
Information input collection is denoted as Xt={ x0, x1, x2..., xt-1, xt, xt+1... }, archives recommending data output collection is then denoted as Yt=
{y0, y1, y2..., yt-1, yt, yt+1... }, the output set of hidden unit i.e. the memory representation of t moment are set St=
{s0, s1, s2..., st-1, st, st+1... }.As shown in figure 3, the information of single direction is transmitted to output list from hiding unit
Member, the output at current time are obtained by the outgoing event of past memory and current time as a result, RNN algorithm recorded
It removes the interested archive information of user and predicts that future customer may interested archive information.This is the RNN on basis
Algorithm structure, but since RNN is consistent with the processing of current information for past information, so that for the archives number of different scorings
It is believed that breath all generates identical influence to the calculating of model, hence it is evident that the algorithm is not enough bonded the individual demand of user, in order into
One step innovatory algorithm improves algorithm accuracy and more meets the requirement of personalized recommendation, we to traditional RNN improved with
Convenient for being preferably applied to recommender system.
Forward data and negative sense data can be divided into the sequence data of input layer user, be respectively intended to indicate that scoring is higher
File data information and the lower file data information of scoring, the output data at each moment should have any different, therefore be
Enhancing forward data has an impact subsequent Generalization bounds and weakens the influence of negative sense data, and RNN algorithm is added simultaneously
Rowization design.
(3) RNN (Recognition with Recurrent Neural Network) paralell design.
By the research to basic RNN algorithm, a part is operated and realizes paralell design, is increased in RNN hidden layer
The concept and algorithm design of two classes gate: forget door and input gate, reach and a part memory is screened, to improve nerve net
The efficiency of network quickly calculates the purpose for meeting the personalized recommendation file data information of user demand.
Forgeing door is utilized while using customer attribute information in the neural network training process of each t moment
User is determined influence of the current file data to Generalization bounds below by score information to the score information of file data, will
Some once interested files or certain archives kind remain before user, are calculated by sigmoid function
The output between 0 and 1 (including 0 and 1) is obtained wherein to abandon by the memory that sigmoid is calculated as 1 as a result, retaining.One
Memory of the partial results between 0 and 1, output result all abandon for 0 then remember before.
Increase input gate, further determines which memory can retain, calculated by sigmoid function layer, according to user
Score the recommendation archives and determine, if score it is higher if be added more information, it is on the contrary then a small amount of information is added.Finally, will
Old memory is given up by reflecting unworthy information after calculating in conjunction with forgetting door, current time user is browsed archives
Information is combined with input gate layer result, is able to for newest user browsing archive information being added in neural network.
Based on calculating before, output is updated by remembering, where calculates determination data by sigmoid layers and tanh layer functions
Part is remembered, and determines the memory of suitable back-propagation.Its design cycle such as Fig. 4.
1.4 file data output layers
For file data output layer, prediction result can be exported using Softmax function modelling, but with the increasing of user
More, file data information is also constantly increasing, and when output data quantity is excessively huge, undoubtedly increases to personalized recommendation system
Very big calculating pressure is added, this can have been solved the problems, such as using stratification Softman function thus.Based on Huffman tree
Thought establishes the Huffman tree about file data, and every a file data file has corresponding Huffman to compile in data set
Code is calculated based on this thought implementation level Softman function.
According to the characteristic of Huffman tree, for the recommendation information arbitrarily exported, all exist in Huffman tree and there is only
Exclusive path, each branch point are equivalent to one two classification, and the probability value that each two classification generates mutually is obtained at convenience
The prediction probability value of required each file data information, the highest file data information of the final output probability of recommender system are
The most possible interested files of user.
Stratification Softmax is compared with Softmax, and stratification Softmax is reduced model computation complexity, especially
For huge file data information, training effectiveness is able to greatly promote undoubtedly extremely important and crucial for it, more has
Conducive to current ever-increasing file data amount is coped with, more accurate, efficient personalization is provided for user in big data era
Recommendation service.
2. constructing file data AI personalized recommendation system of the AI with " three libraries, two module " for core
The design of personalized recommendation system should be while user oriented towards administrative staff, therefore fully considers the two
Demand be designed, existing user data information acquisition also include administrative staff Modifying model or historical information are added
Add.Thus design AI personalized recommendation system integrates file data, user information and AI with " three libraries, two module " for basic structure
Personalized Recommendation Strategy and algorithm, the design framework specify the division and function of each section module database.
(1) " three libraries " function division and design
1. customer data base is a part the most basic, the information input about user is stored in the database,
Therefore scoring and shelves of the dossier, user that the database purchase customer attribute information, user browsed to file data
The classification of case data and Privacy Preservation Mechanism to relative users file data information not only store magnanimity file data letter
Breath increases secret protection to prevent information leakage, ensures that data safety is reliable.
2. Generalization bounds database mainly stores AI personalized recommendation algorithm and correlation based on RNN (Recognition with Recurrent Neural Network)
The file data information needed, Generalization bounds database calling and obtaining user historical viewings mistake from customer data base and archive database
The scoring to browsed files of files and user, then pass through RNN algorithm analytical calculation in the database
It obtains the highest user's recommendation information of probability and is stored, opening interface is exported from the database convenient for data.It can
To need to input historical data information and Adjusted Option information according to the actual situation.
3. archive database stores file data and archives metadata.The input of all files data information is stored in the number
According to library, Generalization bounds database can therefrom transfer archive information.
(2) " two modules " function division and design
In file data AI personalized recommendation system, generation module is recommended to extract file data letter from three databases
Breath, by the AI Personalized Recommendation Strategy in Generalization bounds database and based on the personalized recommendation of RNN (Recognition with Recurrent Neural Network)
Algorithm generates recommendation information.The module data processing step is substantially are as follows: extracts user from customer data base and archive database
User data and file data are merged in Generalization bounds database and classify and be abstracted as sequence number by data and file data
According to by data prediction, being calculated by RNN hidden layer, provide probability vector by output layer, analysis user may be interested
File data information is simultaneously made prediction, and is finally generated to target user and is met the recommendation results of user demand and be stored in recommendation plan
Slightly in database.
Preferably, in step 104: according to the feedback of user to user recommendation information, to the ginseng of neural network recommendation algorithm
Number is adjusted, the neural network recommendation algorithm after obtaining optimization.
The recruitment evaluation module of the application is responsible for carrying out the recommendation file data information that recommender system provides according to user
It feeds back to be adjusted optimization to model, administrative staff, which can therefrom input, needs modified parameter information, can also increase more
More archive informations are adjusted the accuracy of model, and making every effort to more fitting user needs the personalization of file data information
It asks.
The application embodiment is based on the proposed algorithm of RNN (Recognition with Recurrent Neural Network), since traditional archive service cannot
Meet the needs of active user is personalized to file data information, to meet user to file data information resources utilization requirement,
It is user demand oriented, study RNN (Recognition with Recurrent Neural Network) proposed algorithm based on digital archives.RNN (circulation nerve net
Network) it is that the model structure of deep learning a kind of introduces in Recognition with Recurrent Neural Network (RNN) different from conventional neural network model
The structure of directed circulation can solve the relevant issues that are mutually related between those input information, that is to say, that for each
Position user demand, the recommendation information that each moment is calculated all with the browsing of the past user record, archives scoring and on
The recommendation information that one moment obtained is related.The algorithm handles file data information analysis, calculates by collecting user information
To user to the preference and demand of file data, AI personalized recommendation information is provided to target user, meets user to archives number
It is believed that breath diversification and multi-level demand.
The application building is the personalized recommendation engine of basic structure with " three libraries, two module ", mainly includes that " three libraries " is used
User data library, Generalization bounds database and archive database, respectively to store customer attribute information, based on RNN (circulation nerve
Network) personalized recommendation algorithm and file data and metadata.Generalization bounds database is from customer data base and archives number
According to RNN algorithm analytical calculation is carried out in calling and obtaining user essential information in library and file data information and Generalization bounds database, obtain
To AI personalized recommendation as a result, being stored in the Generalization bounds database；" two modules " recommends generation module and recruitment evaluation
Module, the former reads information from " three libraries " and generates recommendation results, and the latter collects user feedback and according to feedback information to recommendation
AI personalization algorithm in policy database carries out parameter adjustment." three libraries, two module " is the personalized recommendation engine of basic structure
Technology realizes the combination of AI personalized recommendation algorithm and digital archive system, realizes that personalized recommendation engine technique and user are believed
The Information Resources Integration of breath and personalized recommendation algorithm, the final storage for realizing mass users information and file data information, shelves
The target of case data AI personalization and intelligent recommendation, it is established that meet the service mechanism of file data individual demand.
The digital archives management system for the AI personalized recommendation engine that the application proposes will meet user to archives personalization
The demand of service is significant to pushing Archival Informationization process, expansion AI industrial application, promotion upgrading of industry industry structure change etc. to bring
Benefit.File data personalization service engine is built upon on digitized records application environmental basis, and the application not only exists
The promotion of service quality has been carried out in existing file administration, has more incorporated technology of new generation, is scientific research and popularization of knowledge etc.
Intelligent Service is provided, the efficiency of archive information utilization is improved, information is made to play maximum benefit.Meanwhile can once put into, it is more
Secondary output pushes the innovation and development of the digitalization construction of archives.
Fig. 5 is according to a kind of based on personalized user information recommender system structure chart of the preferred embodiment for the present invention.
As shown in figure 5, the application provides a kind of user information recommender system based on personalization, system includes:
Acquiring unit 501, for obtaining the scoring of information and user that user has focused on to information of interest, structure
Build a plurality of information of user's mistake of interest and the sequence data of multiple scorings.Preferably, acquiring unit 501 is used for: building user
The a plurality of information of mistake of interest and the sequence data of multiple scorings, sequence data include: message file number, user's scoring, shelves
Case classification.
Construction unit 502 constructs user information for sequence data to be carried out calculating analysis by neural network algorithm
Neural network recommendation algorithm.Preferably, neural network algorithm is Recognition with Recurrent Neural Network algorithm.Preferably, construction unit 502 is used
In: sequence data is subjected to calculating analysis by neural network algorithm, constructs the neural network recommendation algorithm of user information, is wrapped
It includes:
Sequence data is input to the input layer of Recognition with Recurrent Neural Network algorithm；
The hidden layer of Recognition with Recurrent Neural Network algorithm is designed, by the latter number of plies of the hidden layer of Recognition with Recurrent Neural Network algorithm
Input value be added in the output valve of hidden layer preceding layer of Recognition with Recurrent Neural Network algorithm；
Sequence data is divided into forward data and negative sense data, forward data is for enhancing Generalization bounds；Negative sense data are used
In decrease Generalization bounds.
Preferably, construction unit 502 is used for: sequence data being carried out calculating analysis by neural network algorithm, building is used
The neural network recommendation algorithm of family information, comprising:
Door and input gate are forgotten in setting in the hidden layer of Recognition with Recurrent Neural Network algorithm, forget door and are used for according to user to
The scoring of the information of concern retains the information that scoring is higher than first threshold；Input gate passes through threshold function table sigmoid letter
Several layers of calculating, the scoring according to user to information determine the quantity that information is added；
Output layer is constructed using stratification exponential function Softmax function, Recognition with Recurrent Neural Network algorithm carries out stratification and refers to
Number function Softmax function.
Computing unit 503 counts information using neural network recommendation algorithm for the information in Extracting Information archives
It calculates, using the highest information of probability as user's recommendation information；
Optimize unit 504, for the feedback according to user to user recommendation information, to the parameter of neural network recommendation algorithm
It is adjusted, the neural network recommendation algorithm after obtaining optimization.
A kind of the user information recommender system 500 and the present invention based on personalization of the preferred embodiment for the present invention are another excellent
The a kind of corresponding based on personalized user information recommended method 100 of embodiment is selected, is no longer repeated herein.
The present invention is described by reference to a small amount of embodiment.However, it is known in those skilled in the art, as
Defined by subsidiary Patent right requirement, in addition to the present invention other embodiments disclosed above equally fall in it is of the invention
In range.
Normally, all terms used in the claims are all solved according to them in the common meaning of technical field
It releases, unless in addition clearly being defined wherein.All references " one // be somebody's turn to do [device, component etc.] " are explained with being all opened
For at least one example in device, component etc., unless otherwise expressly specified.The step of any method disclosed herein, does not all have
Necessity is run with disclosed accurate sequence, unless explicitly stated otherwise.
Claims (10)
1. a kind of based on personalized user information recommended method, which comprises
Scoring of the information and user that acquisition user has focused on to information of interest, constructs the described of user's mistake of interest
The sequence data of a plurality of information and multiple scorings；
The sequence data is subjected to calculating analysis by neural network algorithm, the neural network recommendation for constructing user information is calculated
Method；
Information in Extracting Information archives calculates the information using neural network recommendation algorithm, and probability is highest
Information is as user's recommendation information；
According to user to the feedback of user's recommendation information, the parameter of the neural network recommendation algorithm is adjusted, is obtained
Neural network recommendation algorithm after taking optimization.
2. according to the method described in claim 1, the neural network algorithm is Recognition with Recurrent Neural Network algorithm.
3. according to the method described in claim 2, described carry out calculating analysis by neural network algorithm for the sequence data,
Construct the neural network recommendation algorithm of user information, comprising:
The sequence data is input to the input layer of the Recognition with Recurrent Neural Network algorithm；
The hidden layer of the Recognition with Recurrent Neural Network algorithm is designed, the hidden layer of the Recognition with Recurrent Neural Network algorithm is latter
The input value of the number of plies is added in the output valve of the hidden layer preceding layer of the Recognition with Recurrent Neural Network algorithm；
The sequence data is divided into forward data and negative sense data, the forward data is for enhancing Generalization bounds；It is described negative
To data for weakening Generalization bounds.
4. according to the method described in claim 2, described carry out calculating analysis by neural network algorithm for the sequence data,
Construct the neural network recommendation algorithm of user information, comprising:
Door and input gate are forgotten in setting in the hidden layer of the Recognition with Recurrent Neural Network algorithm, and the forgetting door is used for according to user
Scoring to the information having focused on retains the information that the scoring is higher than first threshold；The input gate passes through threshold value
Function sigmoid function layer calculates, the scoring according to user to information, determines the quantity that information is added；
Output layer is constructed using stratification exponential function Softmax function, the Recognition with Recurrent Neural Network algorithm carries out stratification and refers to
Number function Softmax function.
5. according to the method described in claim 1, a plurality of information and multiple scorings of building user's mistake of interest
Sequence data, the sequence data include:
Message file number, user's scoring, archives kind.
6. a kind of user information recommender system based on personalization, the system comprises:
Acquiring unit constructs user institute for obtaining the scoring of information and user that user has focused on to information of interest
The sequence data of a plurality of information and multiple scorings paid close attention to；
Construction unit constructs the mind of user information for the sequence data to be carried out calculating analysis by neural network algorithm
Through network recommendation algorithm；
Computing unit calculates the information using neural network recommendation algorithm for the information in Extracting Information archives,
Using the highest information of probability as user's recommendation information；
Optimize unit, for the feedback according to user to user's recommendation information, to the ginseng of the neural network recommendation algorithm
Number is adjusted, the neural network recommendation algorithm after obtaining optimization.
7. system according to claim 6, the neural network algorithm is Recognition with Recurrent Neural Network algorithm.
8. system according to claim 7, the construction unit is used for: the sequence data is passed through neural network algorithm
Calculating analysis is carried out, the neural network recommendation algorithm of user information is constructed, comprising:
The sequence data is input to the input layer of the Recognition with Recurrent Neural Network algorithm；
The hidden layer of the Recognition with Recurrent Neural Network algorithm is designed, the hidden layer of the Recognition with Recurrent Neural Network algorithm is latter
The input value of the number of plies is added in the output valve of the hidden layer preceding layer of the Recognition with Recurrent Neural Network algorithm；
The sequence data is divided into forward data and negative sense data, the forward data is for enhancing Generalization bounds；It is described negative
To data for weakening Generalization bounds.
9. system according to claim 7, the construction unit is used for: the sequence data is passed through neural network algorithm
Calculating analysis is carried out, the neural network recommendation algorithm of user information is constructed, comprising:
Door and input gate are forgotten in setting in the hidden layer of the Recognition with Recurrent Neural Network algorithm, and the forgetting door is used for according to user
Scoring to the information having focused on retains the information that the scoring is higher than first threshold；The input gate passes through threshold value
Function sigmoid function layer calculates, the scoring according to user to information, determines the quantity that information is added；
Output layer is constructed using stratification exponential function Softmax function, the Recognition with Recurrent Neural Network algorithm carries out stratification and refers to
Number function Softmax function.
10. system according to claim 6, the acquiring unit is used for: a plurality of letter of building user's mistake of interest
The sequence data of breath and multiple scorings, the sequence data include:
Message file number, user's scoring, archives kind.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811574681.7A CN109815309A (en) | 2018-12-21 | 2018-12-21 | A kind of user information recommended method and system based on personalization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811574681.7A CN109815309A (en) | 2018-12-21 | 2018-12-21 | A kind of user information recommended method and system based on personalization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109815309A true CN109815309A (en) | 2019-05-28 |
Family
ID=66602238
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811574681.7A Pending CN109815309A (en) | 2018-12-21 | 2018-12-21 | A kind of user information recommended method and system based on personalization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109815309A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110598766A (en) * | 2019-08-28 | 2019-12-20 | 第四范式（北京）技术有限公司 | Training method and device for commodity recommendation model and electronic equipment |
| CN111274331A (en) * | 2020-01-15 | 2020-06-12 | 中国建设银行股份有限公司 | Relational data management maintenance system and method |
| CN112784165A (en) * | 2021-01-29 | 2021-05-11 | 北京百度网讯科技有限公司 | Training method of incidence relation estimation model and method for estimating file popularity |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7069256B1 (en) * | 2002-05-23 | 2006-06-27 | Oracle International Corporation | Neural network module for data mining |
| CN106779921A (en) * | 2016-11-30 | 2017-05-31 | 广州市万表科技股份有限公司 | Recommend method and device |
| CN107507054A (en) * | 2017-07-24 | 2017-12-22 | 哈尔滨工程大学 | A kind of proposed algorithm based on Recognition with Recurrent Neural Network |
| CN108537373A (en) * | 2018-03-27 | 2018-09-14 | 黄晓鸣 | Travel information recommends method and apparatus |
| CN108563755A (en) * | 2018-04-16 | 2018-09-21 | 辽宁工程技术大学 | A kind of personalized recommendation system and method based on bidirectional circulating neural network |
| CN108984680A (en) * | 2018-06-29 | 2018-12-11 | 北京百度网讯科技有限公司 | A kind of information recommendation method, device, server and storage medium |
-
2018
- 2018-12-21 CN CN201811574681.7A patent/CN109815309A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7069256B1 (en) * | 2002-05-23 | 2006-06-27 | Oracle International Corporation | Neural network module for data mining |
| CN106779921A (en) * | 2016-11-30 | 2017-05-31 | 广州市万表科技股份有限公司 | Recommend method and device |
| CN107507054A (en) * | 2017-07-24 | 2017-12-22 | 哈尔滨工程大学 | A kind of proposed algorithm based on Recognition with Recurrent Neural Network |
| CN108537373A (en) * | 2018-03-27 | 2018-09-14 | 黄晓鸣 | Travel information recommends method and apparatus |
| CN108563755A (en) * | 2018-04-16 | 2018-09-21 | 辽宁工程技术大学 | A kind of personalized recommendation system and method based on bidirectional circulating neural network |
| CN108984680A (en) * | 2018-06-29 | 2018-12-11 | 北京百度网讯科技有限公司 | A kind of information recommendation method, device, server and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 李光: "基于循环神经网络的推荐算法研究" * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110598766A (en) * | 2019-08-28 | 2019-12-20 | 第四范式（北京）技术有限公司 | Training method and device for commodity recommendation model and electronic equipment |
| CN110598766B (en) * | 2019-08-28 | 2022-05-10 | 第四范式（北京）技术有限公司 | Training method and device for commodity recommendation model and electronic equipment |
| CN111274331A (en) * | 2020-01-15 | 2020-06-12 | 中国建设银行股份有限公司 | Relational data management maintenance system and method |
| CN112784165A (en) * | 2021-01-29 | 2021-05-11 | 北京百度网讯科技有限公司 | Training method of incidence relation estimation model and method for estimating file popularity |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102567464B (en) | Based on the knowledge resource method for organizing of expansion thematic map | |
| CN103605665B (en) | Keyword based evaluation expert intelligent search and recommendation method | |
| Karim et al. | Decision tree and naive bayes algorithm for classification and generation of actionable knowledge for direct marketing | |
| CN110162593A (en) | A kind of processing of search result, similarity model training method and device | |
| CN109255586B (en) | Online personalized recommendation method for e-government affairs handling | |
| CN105808590B (en) | Search engine implementation method, searching method and device | |
| CN112396108A (en) | Service data evaluation method, device, equipment and computer readable storage medium | |
| CN113254659A (en) | File studying and judging method and system based on knowledge graph technology | |
| CN110188346A (en) | A kind of network security bill part intelligence analysis method based on information extraction | |
| CN102708130A (en) | Scalable engine that computes user micro-segments for offer matching | |
| CN110990567A (en) | Electric power audit text classification method for enhancing domain features | |
| CN109947902B (en) | Data query method and device and readable medium | |
| CN109815309A (en) | A kind of user information recommended method and system based on personalization | |
| CN109471944A (en) | Training method, device and the readable storage medium storing program for executing of textual classification model | |
| CN110287292B (en) | Judgment criminal measuring deviation degree prediction method and device | |
| Cong | Personalized recommendation of film and television culture based on an intelligent classification algorithm | |
| Chen et al. | A deep learning method for judicial decision support | |
| CN110059190A (en) | A kind of user's real-time point of view detection method based on social media content and structure | |
| CN113298318A (en) | Novel overload prediction method for distribution transformer | |
| CN113761359A (en) | Data packet recommendation method and device, electronic equipment and storage medium | |
| Ramkumar et al. | A survey on mining multiple data sources | |
| CN115577080A (en) | Question reply matching method, system, server and storage medium | |
| Al Fatta et al. | Captioning image using convolutional neural network (CNN) and long-short term memory (LSTM) | |
| WO2023246849A1 (en) | Feedback data graph generation method and refrigerator | |
| Li et al. | Automatic classification algorithm for multisearch data association rules in wireless networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190528 |
|
| RJ01 | Rejection of invention patent application after publication |