CN106375369B - The business recommended method of mobile Web and Collaborative Recommendation system based on user behavior analysis - Google Patents
The business recommended method of mobile Web and Collaborative Recommendation system based on user behavior analysis Download PDFInfo
- Publication number
- CN106375369B CN106375369B CN201610688083.7A CN201610688083A CN106375369B CN 106375369 B CN106375369 B CN 106375369B CN 201610688083 A CN201610688083 A CN 201610688083A CN 106375369 B CN106375369 B CN 106375369B
- Authority
- CN
- China
- Prior art keywords
- web
- user
- feature vector
- target user
- module
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/02—Protocols based on web technology, e.g. hypertext transfer protocol [HTTP]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/51—Discovery or management thereof, e.g. service location protocol [SLP] or web services
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of business recommended method of mobile Web based on user behavior analysis and Collaborative Recommendation systems, Web service prediction model is constructed first, and pass through intelligent terminal side user browse data and wearable device side user's physiological data, searching and two optimal association users that target user's long-term habits are most like and short-term emotional is most like.In turn, using two optimal association user data rich target user prediction model sample databases, the sample-rich mechanism for introducing minimum is realized.The web resource memory mechanism based on feature vector is designed, according to model prediction as a result, realizing the accurate recommendation of mobile Web business.The program passes through interactive collaboration between intelligent terminal and wearable device, and wearable device side physiological data is carried out from multi-angle of view, multi dimensional analysis user behavior, to realize Accurate Prediction and the recommendation of business, and then promote the usage experience of user using reconstruct.
Description
Technical field
The present invention relates to a kind of business recommended method of mobile Web based on user behavior analysis and Collaborative Recommendation systems, belong to
In recommended technology field.
Background technique
In response in the explosive growth trend of mobile interchange business, business recommended engine is overloaded as user information is solved
And the effective tool of the isotropic problem of information is sieved from massive information by obtaining and predicting the potential preference of mobile subscriber for user
Business interested is selected, and realizes business pre-download, reach balance network load and reduces the effect of service delay.It is business recommended to draw
It holds up and makes the premise of Accurate Prediction and need a large number of users behavior sample as support, but be limited to the sample size of single user not
The problems such as foot, type are limited, it is low to result in traffic forecast accuracy, recommends quality bad.
Although having the relevant technologies to solve the above problems, the collaborative filtering based on experience and the association based on model are such as used
Target user's sample database is enriched using relevant user data by finding association user with the method for filtering.But there are still some
Problem, such method are typically all that the prediction result of all association users is weighted fusion, this will introduce other users
Sample noise, cause precision of prediction not high, or even become lower, good business experience can not be also provided for user.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of business recommended sides of the mobile Web based on user behavior analysis
Method and Collaborative Recommendation system, this method constructs Web service prediction model first, and passes through intelligent terminal side user browse data
With wearable device side user's physiological data, find with target user's long-term habits are most like and short-term emotional is most like two
Optimal association user.In turn, it using two optimal association user data rich target user prediction model sample databases, realizes and introduces
The sample-rich mechanism of minimum.Design the web resource memory mechanism based on feature vector, according to model prediction as a result,
Realize the accurate recommendation of mobile Web business.The program passes through interactive collaboration between intelligent terminal and wearable device, sets wearable
Standby side physiological data is carried out using reconstruct, from multi-angle of view, multi dimensional analysis user behavior, is realized the Accurate Prediction of business and is pushed away
It recommends, and then promotes the usage experience of user.
The present invention uses following technical scheme to solve above-mentioned technical problem:
On the one hand, the present invention provides a kind of business recommended method of the mobile Web based on user behavior analysis, periodically to mesh
Mark user carry out mobile Web it is business recommended, this method comprising the following specific steps
Step 1, Web feature vector is designed, the mapping table between Web URL and Web feature vector is established, specifically:
1.1, the domain name of Web URL is parsed, extracts the first and second characteristic values of the corresponding Web, wherein first
Characteristic value is enterprise or organization's title, and Second Eigenvalue is type of service;The resource path name of Web URL is solved
Analysis, extracts the third feature value of the corresponding Web, and third feature value is the specific form of service under type of service;To obtain Web
Feature vector, element are first to third feature value;
1.2, according to the Web URL information of extraction, establish the mapping table between Web URL and Web feature vector;
Step 2, the web resource library based on Web feature vector is established, specifically:
2.1, all URL informations for prestoring web resource are extracted, the mapping between Web URL and Web feature vector is inquired
Table generates corresponding Web feature vector;
2.2, according to the characteristic value in Web feature vector, classification storage is carried out to all web resources that prestore, i.e., it is same to deposit
Storage corresponds to the same type of service of same company or organization under path, to establish web resource library;
Step 3, it is recorded according to the history web browsing of target user, constructs Web service prediction model, specifically:
3.1, it is recorded according to the history web browsing of target user, inquires the mapping between Web URL and Web feature vector
Table, the history web browsing for obtaining target user record corresponding several Web feature vectors;
3.2, the tree structure of characterization target user's web browsing record is established, which is not present root node, and appoints
Two different nodes of anticipating can subsequent node each other, wherein each node characterizes a feature vector, and is stored in current node
The number that current node is shifted to other nodes;The tree structure is Web service prediction model;
Step 4, it is recorded according to the browsing in target user's current period, carries out Web service recommendation to target user, specifically
Are as follows:
4.1, the browsing record at target user's current period start time to current time is extracted, Web URL and Web are inquired
Mapping table between feature vector generates corresponding Web feature vector;
4.2, according to browsing timing, the Web service prediction model in Web feature vector and step 3 that 4.1 are generated is carried out
Matching, if all Web feature vectors and mutual transfer relationship can find corresponding node in Web service prediction model
And transfer path, then with the Web feature vector at target user's nearest moment, the highest Web of transfer number is special in corresponding node
Vector is levied as prediction result, executes 4.3;Otherwise, prediction is not executed, next period is waited;
4.3, according to prediction result, inquire web resource library, find with the most matched store path of prediction result, this is deposited
Newest web resource under storage path recommends target user.
As a further optimization solution of the present invention, web resource library is periodically updated, and records renewal time.
As a further optimization solution of the present invention, this method further includes periodically carrying out more to Web service prediction model
Newly, specifically:
If in current period, target user has browsing to record, then record, inquiry are browsed according to target user in current period
Mapping table between Web URL and Web feature vector generates corresponding Web feature vector, utilizes generation according to step 3.2
Web feature vector and mutual transfer relationship are updated Web service prediction model;Meanwhile current period is compared respectively
The phase of browsing record between interior target user and the most like association user of its long-term habits, the most like association user of short-term emotional
Like degree, if between target user and the most like association user of long-term habits browsing record similarity be higher than target user with it is short
The similarity recorded is browsed between the most like association user of phase mood, then long-term habits important factor w1Add 0.1, short-term emotional
Important factor w2Subtract 0.1, otherwise short-term emotional important factor w2Add 0.1, long-term habits important factor w1Subtract 0.1;
If in current period, target user does not browse record, then according to the long-term habits of target user in current period
The browsing of most like association user and the most like association user of short-term emotional record, is inquired between Web URL and Web feature vector
Mapping table, generate corresponding Web feature vector, according to step 3.2 using generate Web feature vector and mutual transfer
Relationship is updated Web service prediction model, wherein the Web feature vector of the corresponding most like association user of long-term habits it
Between transfer number add w1, the transfer number between the Web feature vector of the corresponding most like association user of short-term emotional adds w2。
As a further optimization solution of the present invention, w1、w2Initial value be all 0.5, and 0≤w1≤ 1,0≤w2≤1。
As a further optimization solution of the present invention, this method further includes periodically acquiring the long-term habits of target user most
Analogous relationship user and the most like association user of short-term emotional, specifically:
1) the association user collection Ψ of target user is determined, wherein association user is the mobile social circle of target user
Contact person;
2) the association user u most like with the long-term habits of target user m is periodically found*, specifically:
In a cycle, the association user for having browsing to record integrates as Λ, whereinFirstly, count the period it
Preceding N1The browsing of target user m and any association user u records in × T time section, and generates corresponding feature vector set,
Wherein, u ∈ Λ, N1>=100, T are the period;Secondly, according to browsing timing, statistics target user m and association user u feature to
The frequency of occurrences of feature vector transfer pair in duration set, and statistical result is ranked up;Again, according to the statistics knot after sequence
Fruit calculates separately the long-term habits similarity R of any association user u Yu target user m(m,u);Finally, the length with target user m
It is the most like association user u of long-term habits that phase, which is accustomed to the highest association user of similarity,*;
3) it periodically finds and the most like association user v of target user m short-term emotional*, specifically:
In a cycle, the association user collection for having browsing to record isWherein,Firstly, count the period it
Preceding N2Every physiological data set of target user m and any association user v in × T time section, whereinN2≤
10;Secondly, calculating separately target use to the physiological data in the jth kind physiological data set of target user m and association user v
The similarity ρ of jth kind physiological data between family m and any association user vj(m,v);Again, target user m and any association are used
The short-term emotional similarity ρ of family v(m,v)ForWherein, 1≤j≤L, L is are counted physiological data kind
Class;Finally, the highest association user of short-term emotional similarity with target user m is the most like association user of short-term emotional
v*。
As a further optimization solution of the present invention, using Kendall rank correlation coefficient or Spearman rank correlation coefficient
Calculate separately the long-term habits similarity R of any association user u Yu target user m(m,u)。
As a further optimization solution of the present invention, it is counted respectively using Pearson correlation coefficient or cosine similarity function
Calculate the similarity ρ of the jth kind physiological data between target user m and any association user vj(m,v)。
On the other hand, the mobile Web business cooperation recommender system based on user behavior analysis that the present invention also provides a kind of, packet
It includes: wearable terminal, user terminal, gateway, server, wherein
Wearable terminal includes physiological data collection module and the first wireless transport module, wherein physiological data collection mould
Block acquires every physiological data of user in real time, is transmitted to user terminal by the first wireless transport module;
User terminal includes the second wireless transport module, browser module, Web feature vector generation module, local data
Cache module, recommending module, third wireless transport module, wherein the second wireless transport module is used for and the first wireless transmission mould
Block carries out wireless communication;Browser module is used to collect the browsing record of user, shows recommendation business;Web feature vector generates
Module is used to carry out feature extraction to the browsing record of user, generates Web feature vector;Local data cache module, for delaying
Deposit the physiological data that user browses record and its corresponding Web feature vector, user;Recommending module is for receiving from gateway
Recommendation business, and shown in browser module in the form of pop-up after the format conversion;Third wireless transport module is used for user
The wireless communication of terminal and gateway;
Gateway includes: the 4th wireless transport module, User IP management module, recommends service buffer module, wherein the 4th nothing
Line transmission module is used for the wireless communication of user terminal and gateway;User IP management module, for distributing IP for institute accessing user
Address;Recommend service buffer module, recommendation business is sent to use for caching the recommendation business from server, and periodically
Family terminal;
Server includes database module, prediction model constructs and update module, business recommended module, web resource library,
In, database module, for storing the feature vector for representing user browsing behavior and every physiological data;Prediction model building and
Update module is periodically updated for constructing Web service prediction model, and to model;Business recommended module is for executing
Web service prediction, and the high web resource of similarity is found from web resource library according to prediction result and is recommended;Web resource
Library, for being stored according to Web feature vector to web resource.
As a further optimization solution of the present invention, it is connected between gateway and server by wired interface module.
As a further optimization solution of the present invention, wearable terminal further includes data preprocessing module, for acquisition
Physiological data cleaned and filtered, removal acquisition mistake and error.
The invention adopts the above technical scheme compared with prior art, has following technical effect that proposed by the present invention one
Mobile Web business recommended method and wearable Collaborative Recommendation engine of the kind based on user behavior analysis, pass through and construct long-term habits
Index and short-term emotional index find the two optimal association users most like with target user's long-term habits and short-term emotional,
It realizes and introduces the smallest sample-rich mechanism of noise, in addition, carrying out multi-dimension feature extraction to Web, not only improve prediction model
Storage organization, also improve the recommendation quality of engine.The mobile Web recommended method and recommended engine provided through the invention,
On the one hand user is able to solve under mobile interchange environment, on the other hand information overload and isotropic problem can be realized movement
Web service pre-download with balance network load and reduces service delay, and then promotes the business experience of user.
Detailed description of the invention
Fig. 1 is a kind of business recommended method flow diagram of mobile Web based on user behavior analysis provided by the invention.
Fig. 2 is a kind of Web service prediction model data more new strategy schematic diagram provided by the invention.
Fig. 3 is a kind of Web service prediction model schematic diagram provided by the invention.
Fig. 4 is the mobile Web business cooperation recommender system modular construction provided by the invention based on user behavior analysis
Schematic diagram.
Specific embodiment
With reference to the accompanying drawing and specific embodiment is described in further detail technical solution of the present invention:
It is specific as follows the present invention provides a kind of business recommended method of the mobile Web based on user behavior analysis:
One, Web feature vector is designed, webpage URL information is extracted, establishes the mapping between Web URL and Web feature vector
Table;
1.1, the domain name of Web URL is parsed, extracts the first and second characteristic values of the corresponding Web, wherein first
Characteristic value is enterprise or organization's title, and Second Eigenvalue is type of service;The resource path name of Web URL is solved
Analysis, extracts the third feature value of the corresponding Web, and third feature value is the specific form of service under type of service;To obtain Web
Feature vector, element are first to third feature value;
1.2, according to the Web URL information of extraction, establish the mapping table between Web URL and Web feature vector.
Two, the web resource library based on Web feature vector is established, specifically:
2.1, all URL informations for prestoring web resource are extracted, the mapping between Web URL and Web feature vector is inquired
Table generates corresponding Web feature vector;
2.2, according to the characteristic value in Web feature vector, classification storage is carried out to all web resources that prestore, i.e., it is same to deposit
Storage corresponds to the same type of service of same company or organization under path, to establish web resource library.It simultaneously can be with the period
Property update resources bank, and add renewal time label.
Three, it is recorded according to target user's history web browsing, constructs Web service prediction model.
The Web service prediction model is a kind of special tree structure, which is not present root node, and any
Two different nodes subsequent node, each of tree node can represent a feature vectors, store and work as in node each other
Transfer case of the preceding service feature vector to another service feature vector.
3.1, it is recorded according to the history web browsing of target user, inquires the mapping between Web URL and Web feature vector
Table, the history web browsing for obtaining target user record corresponding several Web feature vectors.
3.2, to obtained several Web feature vectors, according to each feature vector timing, successively searching in prediction model is
It is no to include corresponding node;If comprising, in corresponding node, the transfer feelings of record this feature vector to next group of feature vector
Condition, corresponding transfer number add 1;If not including, corresponding node is created, and records this feature vector to next group of feature vector
Transfer case, corresponding transfer number add 1.
Four, (cycle T < 60min) periodically is updated to Web service prediction model:
The most like association user u of long-term habits of target user m is introduced first*, the most like association user v of short-term emotional*,
As its name suggests, the two optimal users are in a cycle, with target user's m long-term habits most like association user and
The most like association user with target user's m short-term emotional.Wherein, the long-term habits similarity of different user, passes through user's
The similarity of web browsing behavior is measured, the similarity that the short-term emotional similarity of different user passes through the physiological parameter of user
To measure.Since the web browsing behavior of user depends on two factors of long-term habits and short-term emotional, i.e., do not have in target user m
Have in the period of browsing behavior, the browsing of the most like association user of long-term habits and the most like association user of short-term emotional can be used
Behavior represents the browsing behavior of target user m.At the end of the time of the update of Web service prediction model is each period
It carves.
4.1, periodically find the most like association user u of long-term habits of target user*, the most like association of short-term emotional uses
Family v*, specifically:
Firstly, finding the contact person user of association user collection Ψ: target user m mobile social circle, as association user
The association user of concentration.Here the method for searching association user collection Ψ, including but not limited to: mobile telephone registration mode passes through
Associated rights authorization, to obtain target user handset address book contact as target user's association user collection;Mobile agency is handed over the accounts
Number mode is logined, by associated rights authorization, forges a good relationship friend as target user's association user collection to obtain target user's mobile agency.
Secondly, periodically finding and the most like association user u of target user m long-term habits*: in a cycle, have clear
The association user that behavior of looking at generates integrates as Λ, whereinFirstly, counting the N before the period1In × T time section, mesh
It marks the Web service that user m and any association user u is generated and browses record, and generate corresponding feature vector set, wherein u ∈
Λ, N1≥100;Secondly, statistics target user m and any association user u browse each feature vector in record according to browsing timing
To the frequency of occurrences, i.e., previous feature vector is transferred to the transfer number of later feature vector for transfer;In turn, to target user
The statistical result of m and association user u are ranked up;Association is calculated using similarity function finally, for two groups of ranking results
The long-term habits similarity R of user u and target user m(m,u);Similarity function described here includes but is not limited to, such as
Kendall rank correlation coefficient, Spearman rank correlation coefficient etc..It calculates in association user collection Λ, all association users and target
The similarity of user m, and the similar user's optimal model of long-term habits is established, it finds and target user m long-term habits most phase
As association user u*, model is as follows:
Finally, periodically finding and the most like association user v of target user m short-term emotional*: in a cycle, have clear
Behavior of looking at generate association user collection beWherein,Firstly, counting the N before the period2In × T time section, mesh
Mark every physiological data set of user m and association user v, wherein v ∈ Λ, N2≤10;Secondly, to target user m and association
Physiological data in the jth kind physiological data set of user v calculates the jth of user u and target user v using similarity function
Kind physiological data similarity ρj(m,v), similarity function described here includes but is not limited to Pearson correlation coefficient, cosine
Similarity function etc.;Finally, obtaining the short-term emotional of target user m Yu user v to every physiological data similarity statistical average
Similarity ρ(m,v)ForWherein, 1≤j≤L, L is are counted physiological data type;Calculate association user
CollectionIn, the short-term emotional similarity of all users and target user m, and establish the similar user of short-term emotional and optimize mould
Type is found and the most like association user v of target user m short-term emotional*, model is as follows:
4.2, judge in current period, whether target user m has browsing behavior generation;If so, executing step 4.3, otherwise
Execute step 4.4.
4.3, in current period, target user m has web browsing behavior generation, then carries out following two operations: first, benefit
Web service prediction model is updated with the browsing record of target user: being recorded, is looked into the browsing of current period according to target user m
Web URL and Web maps feature vectors table are ask, generates feature vector, and mould is predicted to Web service using the feature vector generated
Type is updated, and renewal process adds 1 with step 3.2, corresponding transfer number;Second, update long-term habits important factor w1And it is short
Phase mood important factor w2: the comparison user m and most like association user u of long-term habits*, the most like association user v of short-term emotional*
The similarity of generated Web feature vector in current period, if m and u*Feature vector similarity it is higher, then long-term habits
Important factor w1Increase by 0.1, short-term emotional important factor w2Reduce 0.1;If m and v*Similarity is higher, then w1Reduce 0.1, w2Increase
Add 0.1.Wherein, w1、w2Initial value be all 0.5, and 0≤w1≤ 1,0≤w2≤1。
4.4, in current period, target user m does not have web browsing behavior generation, then utilizes the most like association of long-term habits
User u*, the most like association user v of short-term emotional*Browsing record update Web service prediction model: firstly, according to long-term habits
Most like association user u*, the most like association user v of short-term emotional*Browsing record within the period, query webpage URL and Web
Maps feature vectors table generates character pair vector;Finally, the feature vector using generation is updated model sample library,
Renewal process adds w with step 3.2, corresponding transfer number respectively1、w2。
Five, it is recorded according to the browsing in target user's current period, carries out Web service recommendation to target user, realize and move
Dynamic Web service is recommended.
Firstly, extracting the browsing record in user's current period start time to current time, query webpage URL and Web
Maps feature vectors table generates individual features vector;Secondly, to the feature vector of generation according to timing, successively and in model
Node matched;If all feature vectors can find corresponding node and transfer path in a model, that is, browse road
Diameter exact matching, then in node corresponding to last group of feature vector, the highest feature vector of transfer number is as prediction knot
Fruit;If browse path Incomplete matching, does not execute prediction.
The mobile Web business cooperation recommender system based on user behavior analysis that the present invention also provides a kind of, such as Fig. 4 institute
Show, comprising:
Wearable terminal is sent to for acquiring every physiological data of user, and after the rejecting of gross error data
Intelligent terminal specifically includes: wireless transport module, for transmitting data using wireless technologys such as WIFI, bluetooths, using UDP points
Group transmission signaling, establishes the communication for coordination with user terminal;Physiological data collection module acquires use using various sensors in real time
Every physiological data at family, the sensor include but is not limited to, such as body temperature transducer, pulse transducer, and heart rate sensor etc.,
Corresponding physiological parameter includes but is not limited to, such as body temperature, pulse, heart rate;Data preprocessing module, for being adopted
The physiological data of collection is cleaned and is filtered, removal acquisition mistake and error.
User terminal, for obtaining user service data and physiological data, being sent to gateway and receiving pushing away from gateway
Recommend information, specifically include: wireless transport module is grouped by wireless technologys such as WIFI, bluetooths using UDP, establish with it is wearable
The communication for coordination of terminal;Browser module, the browsing for collecting user records, and shows the recommendation business from server;
Web feature vector generation module carries out feature extraction for every browsing record to user, generates Web feature vector;It is local
Data cache module, for cache user in a cycle T, a complete browsing record included feature vector and
The physiological data of user;Recommending module carries out the conversion of information format, to conversion for receiving the recommendation information from gateway
Recommendation information is shown in the form of pop-up in browser interface afterwards;Wireless transport module, for using wireless skills such as WiFi, 3G or 4G
Art carries out user data with gateway and recommends the interaction of business.
Gateway receives business datum and physiology number from each intelligent terminal for carrying out IP management to each intelligent terminal
According to being forwarded to server after adding IP information for it, and cache the recommendation business from server, specifically include: User IP pipe
Module is managed, for distributing IP address for institute accessing user;Wireless transport module, for receiving the number of users of each user terminal
According to, and corresponding IP information is added in receiving data;Wired interface module, the object being connected for providing gateway with server
Interface is managed, realizes the transmission of user data and the reception for recommending business;Recommend service buffer module, comes from server for caching
Recommendation business, and recommendation information is periodically sent to intelligent terminal.
Server constructs prediction model to target user, according to prediction result for storing the user data from gateway
Carry out business recommended, and recommendation results be sent to gateway caches, specifically included: wired interface module, for provide gateway and
The physical interface that server is connected realizes the reception of user data and the transmission for recommending business;Database module, for storing
Represent the feature vector and items physiological data of user browsing behavior;Prediction model building and update module, for constructing Web industry
Business is prediction model, and carries out real-time update to model by user data;Business recommended module is calculated for executing traffic forecast
Method, and the high web resource of similarity is found from Web library according to prediction result and is recommended;Web library, according to degree
Web characteristic storage rule, carries out the storage of web resource.
A kind of business recommended method flow diagram of mobile Web based on user behavior analysis as shown in Figure 1 specifically executes step
It is rapid as follows:
Step S101 initializes various parameters, as service period T, user-association user collect Ψ, measurement period coefficient N1、
N2, long-term habits important factor w1And short-term emotional important factor w2Deng.
Step S102 designs Web feature vector, extracts webpage URL information, and establish Web according to extracted URL information
Mapping table between URL and Web feature vector.
Wherein, Web feature vector includes three dimensions, and first dimension represents enterprise or organization, second dimension
Degree represents Web service type, third dimension represents the specific form of service under type of service;And it is extracted to a webpage
URL information process are as follows: the information that feature vector the first two dimension can be obtained by domain-name information, it can by the parsing of resource path name
The information of feature vector third dimension is obtained, such as network address http://sports.sina.com.cn/nba/ is parsed,
One group of feature vector (Sina, sport, nba) can be obtained, can be obtained between the network address and feature vector (Sina, sport, nba)
Mapping relations, and mapping table is established with this.
Step S103 is recorded according to target user's history web browsing, constructs Web service prediction model.
Step S104 constructs optimal model by user's long-term habits similarity and short-term emotional similarity, find with
The most like association user u of target user's m long-term habits*And the association user v that short-term emotional is most like*, detailed process is as follows:
The long-term habits similarity of different user is measured by the web browsing behavior similarity of user, it is described searching with
The most like association user of target user's long-term habits method particularly includes: in a cycle T, there is the association user collection of business generation
For Λ, the N before the period is counted1In × T time section, the Web service that target user m and association user u are generated, which browses, to be gathered,
Each element in set represents a kind of type of service, whereinU ∈ Λ counts two for the timing of business
User browses business transfer pair in record, i.e., previous business jumps to the latter business, number of hops;Firstly, according to target
The statistical result of user m is ranked up, and the rank of highest a pair of of the type of service of business transition frequency is denoted as 1, and and so on;
Secondly, statistical result of the user u according to oneself, the sortord corresponding to target user m are ranked up, obtaining one group of rank is
Out-of-order sequence calculates the long-term habits phase of user u and target user m using Kendall rank correlation coefficient according to the sequence
Like degree R(m,u):
P is after user u sorts according to target user's m sortord, to occur after each rank with the cumulative of ordered pair quantity
With;N is target user m in N1The business occurred in browsing record in × T time section is shifted to quantity.Calculate association user collection
In Λ, the similarity of all users and target user m, and the similar user's optimal model of long-term habits is established, searching and mesh
Mark the most like association user u of user m long-term habits*, model is as follows:
The short-term emotional similarity of different user is measured by the physiological parameter similarity of user, searching and target user
The most like association user of short-term emotional method particularly includes: in a cycle T, the association user for having business to generate integrates as Λ, system
Count the N before the period2In × T time section, every physiological data set of target user m and association user v, physiological data is such as
Body temperature, blood pressure, pulse, heartbeat etc..The jth kind physiological data of target user m and association user v, is denoted as S respectivelyj1={ aj1,
aj2,..aji..,ajNAnd Sj2={ bj1,bj2,..bji..,bjN, wherein N is target user m and association user v in N2When × T
Between in section, the minimum value for the number for thering is business to generate, and N≤N2, 1≤i≤N, 1≤j≤L, L is are surveyed physiological data type;
ajiAnd bjiRespectively target user m and association user v are in time N2In × T, the jth kind physiological data of i-th acquisition;To two
Element in set calculates the jth kind physiological data similarity of user u and target user v using Pearson correlation coefficient
ρj(m,v):
Respectively set Sj1And Sj2Mean value;a'ji、b'jiRespectively ajiAnd bjiNormalized form,
RespectivelyWithNormalized form,
The then short-term emotional similarity ρ of target user m and user v(m,v):
It calculates in association user collection Λ, the short-term emotional similarity of all users and target user m, and establishes short-term emotional
Similar user's optimal model is found and the most like association user v of target user m short-term emotional*, model is as follows:
Step S105 is recorded using the browsing of target user m and two optimal users, periodically predicts mould to Web service
Type is updated.Detailed process is as follows:
First, it is determined that whether target user m has browsing behavior generation in current period;If so, then carrying out following two behaviour
Make: first, utilize target user browsing record update Web service prediction model: according to target user m the period browsing
Record, inquiry Web URL and Web maps feature vectors table generate feature vector, and using the feature vector generated to Web industry
Business prediction model is updated, and corresponding transfer number adds 1;Second, update long-term habits important factor w1And short-term emotional is important
Factor w2: the similarity of comparison user m and two optimal users the generated Web feature vector within the period, if user m and use
Family u*Feature vector similarity it is higher, then long-term habits important factor w1Increase by 0.1, short-term emotional important factor w2Reduce
0.1;If user m and user v*Similarity is higher, then w1Reduce 0.1, w2Increase by 0.1;Wherein, w1、w2Initial value be all 0.5,
And 0≤w1≤ 1,0≤w2≤1。
If no, updating Web service prediction model using the browsing record of two optimal users: firstly, according to two
Browsing record of the optimal user within the period, inquires the mapping table between Web URL and Web feature vector, generates corresponding special
Levy vector;Finally, the feature vector using generation is updated Web service prediction model, corresponding transfer number adds w1、w2。
Step S106 executes prediction algorithm using Web service prediction model, and according to prediction result, realizes mobile Web
It is business recommended.
Wherein, process is predicted are as follows: in Web service prediction model, each storage node represents a kind of service feature vector,
And its storage is transfer case of the current signature vector to another feature vectors, i.e., a kind of type of service is transferred to another kind
Type of service.When executing prediction, signature analysis is carried out to the browsing record of target user first, obtains corresponding multiple groups
Feature vector, and according to node corresponding in Web service prediction model, the transfer case of storage, according to each group feature vector timing
Property is successively searched, and whether each node contains corresponding transfer characteristic vector.If so, then with last group of feature vector institute
In corresponding node, the highest feature vector of transition frequency is as prediction result, if not having, browsing record is only used as updating letter
Breath, does not provide prediction.
Wherein, recommendation process are as follows: find and respectively tieed up with predicted characteristics vector from prestoring in web resource library according to prediction result
The most matched resource path of information, selects last updated web resource under the store path to be pushed to target user.
A kind of Web service prediction model more new strategy schematic diagram as shown in Figure 2, specific renewal process are as follows:
First determine whether target user m has browsing behavior generation in some period.If so, then directly will be in browsing record
Feature vector transfer case be stored in Web service prediction model, and count frequency add 1.In turn, long-term habits important factor is updated
w1And short-term emotional important factor w2, method particularly includes: it finds in the association user for thering is business to generate in the period, is used with target
The family highest user u of m long-term habits similarity*And the highest user v of short-term emotional similarity*, judge that their type of service is
It is no identical as user's m type of service.If exactly the same or Quan Butong, w1、w2It is constant;If target user m is only similar to long-term habits
Spend highest user u*Type of service is identical, then w1Increase by 0.1, w2Reduce 0.1;Otherwise w1Reduce 0.1, w2Increase by 0.1.
If judging whether there is new browsing in the period in association user without browsing behavior in the target user m period
Behavior generates.If no, Web service prediction model does not update;It is practised for a long time if so, then finding in the period with target user m
The used highest user u of similarity*And the highest user v of short-term emotional similarity*, target, which is represented, using their business conduct uses
Business conduct caused by family m business conduct as caused by long-term habits factor and short-term emotional, the corresponding frequency that counts add w1、w2。
A kind of Web service prediction model schematic diagram as described in Figure 3, specific building process are as follows:
Each storage node represents a kind of type of service in prediction model, and node stores current business feature vector to separately
A kind of transfer case of service feature vector generates Web feature vector firstly, recording according to the web browsing of user;Secondly, will
The Web feature vector of generation finds respective stored node in a model, if node exists, updates the node storing data;If
Node is not present, then creates it, and updates node storing data.
The above, the only specific embodiment in the present invention, but scope of protection of the present invention is not limited thereto, appoints
What is familiar with the people of the technology within the technical scope disclosed by the invention, it will be appreciated that expects transforms or replaces, and should all cover
Within scope of the invention, therefore, the scope of protection of the invention shall be subject to the scope of protection specified in the patent claim.
Claims (10)
1. the business recommended method of mobile Web based on user behavior analysis periodically carries out mobile Web business to target user and pushes away
It recommends, which is characterized in that comprising the following specific steps
Step 1, Web feature vector is designed, the mapping table between Web URL and Web feature vector is established, specifically:
1.1, the domain name of Web URL is parsed, extracts the first and second characteristic values of the corresponding Web, wherein fisrt feature
Value is enterprise or organization's title, and Second Eigenvalue is type of service;The resource path name of Web URL is parsed, is mentioned
The third feature value of the corresponding Web is taken, third feature value is the specific form of service under type of service;To obtain Web feature
Vector, element are first to third feature value;
1.2, according to the Web URL information of extraction, establish the mapping table between Web URL and Web feature vector;
Step 2, the web resource library based on Web feature vector is established, specifically:
2.1, all URL informations for prestoring web resource are extracted, the mapping table between Web URL and Web feature vector is inquired, it is raw
At corresponding Web feature vector;
2.2, according to the characteristic value in Web feature vector, classification storage, i.e., same storage road are carried out to all web resources that prestore
The same type of service of same company or organization is corresponded under diameter, to establish web resource library;
Step 3, it is recorded according to the history web browsing of target user, constructs Web service prediction model, specifically:
3.1, it is recorded according to the history web browsing of target user, inquires the mapping table between Web URL and Web feature vector, obtain
History web browsing to target user records corresponding several Web feature vectors;
3.2, the tree structure of characterization target user's web browsing record is established, which is not present root node, and any two
A different node can subsequent node each other, wherein each node characterizes a feature vector, and is stored in current node current
The number that node is shifted to other nodes;The tree structure is Web service prediction model;
Step 4, it is recorded according to the browsing in target user's current period, carries out Web service recommendation to target user, specifically:
4.1, extract the browsing record at target user's current period start time to current time, inquiry Web URL and Web feature
Mapping table between vector generates corresponding Web feature vector;
4.2, according to browsing timing, the Web feature vector that 4.1 are generated and the Web service prediction model progress in step 3
Match, if all Web feature vectors and mutual transfer relationship can be found in Web service prediction model corresponding node with
And transfer path, then with the Web feature vector at target user's nearest moment in corresponding node the highest Web feature of transfer number
Vector executes 4.3 as prediction result;Otherwise, prediction is not executed, next period is waited;
4.3, according to prediction result, inquire web resource library, find with the most matched store path of prediction result, by the storage road
Newest web resource under diameter recommends target user.
2. the business recommended method of the mobile Web according to claim 1 based on user behavior analysis, which is characterized in that week
Phase property updates web resource library, and records renewal time.
3. the business recommended method of the mobile Web according to claim 1 based on user behavior analysis, which is characterized in that should
Method further includes periodically being updated to Web service prediction model, specifically:
If in current period, target user has browsing to record, then record, inquiry are browsed according to target user in current period
Mapping table between WebURL and Web feature vector generates corresponding Web feature vector, utilizes generation according to step 3.2
Web feature vector and mutual transfer relationship are updated Web service prediction model;Meanwhile current period is compared respectively
The phase of browsing record between interior target user and the most like association user of its long-term habits, the most like association user of short-term emotional
Like degree, if between target user and the most like association user of long-term habits browsing record similarity be higher than target user with it is short
The similarity recorded is browsed between the most like association user of phase mood, then long-term habits important factor w1Add 0.1, short-term emotional
Important factor w2Subtract 0.1, otherwise short-term emotional important factor w2Add 0.1, long-term habits important factor w1Subtract 0.1;
If in current period, target user does not browse record, then according to the long-term habits of target user in current period most phase
It is recorded like the browsing of association user and the most like association user of short-term emotional, inquires reflecting between Web URL and Web feature vector
Firing table generates corresponding Web feature vector, utilizes the Web feature vector and mutual transfer relationship generated according to step 3.2
Web service prediction model is updated, wherein between the Web feature vector of the corresponding most like association user of long-term habits
Transfer number adds w1, the transfer number between the Web feature vector of the corresponding most like association user of short-term emotional adds w2。
4. the business recommended method of the mobile Web according to claim 3 based on user behavior analysis, which is characterized in that w1、
w2Initial value be all 0.5, and 0≤w1≤ 1,0≤w2≤1。
5. the business recommended method of the mobile Web according to claim 3 based on user behavior analysis, which is characterized in that should
Method further includes periodically acquiring the most like association user of long-term habits and the most like association user of short-term emotional of target user,
Specifically:
1) the association user collection Ψ of target user is determined, wherein association user is the connection of the mobile social circle of target user
People;
2) the association user u most like with the long-term habits of target user m is periodically found*, specifically:
In a cycle, the association user for having browsing to record integrates as Λ, whereinFirstly, counting the N before the period1
The browsing of target user m and any association user u records in × T time section, and generates corresponding feature vector set, wherein u
∈ Λ, N1>=100, T are the period;Secondly, counting the feature vector set of target user m and association user u according to browsing timing
The frequency of occurrences of middle feature vector transfer pair, and statistical result is ranked up;Again, according to the statistical result after sequence, divide
The long-term habits similarity R of any association user u Yu target user m are not calculated(m,u);Finally, the long-term habits with target user m
The highest association user of similarity is the most like association user u of long-term habits*;
3) it periodically finds and the most like association user v of target user m short-term emotional*, specifically:
In a cycle, the association user collection for having browsing to record isWherein,Firstly, counting the N before the period2
Every physiological data set of target user m and any association user v in × T time section, whereinN2≤10;Its
It is secondary, to the physiological data in the jth kind physiological data set of target user m and association user v, calculate separately target user m with
The similarity ρ of jth kind physiological data between any association user vj(m,v);Again, target user m and any association user v
Short-term emotional similarity ρ(m,v)ForWherein, 1≤j≤L, L is are counted physiological data type;Most
It afterwards, is the most like association user v of short-term emotional with the highest association user of short-term emotional similarity of target user m*。
6. the business recommended method of the mobile Web according to claim 5 based on user behavior analysis, which is characterized in that adopt
The long-term of any association user u and target user m is calculated separately with Kendall rank correlation coefficient or Spearman rank correlation coefficient
It is accustomed to similarity R(m,u)。
7. the business recommended method of the mobile Web according to claim 5 based on user behavior analysis, which is characterized in that adopt
The jth kind between target user m and any association user v is calculated separately with Pearson correlation coefficient or cosine similarity function
The similarity ρ of physiological dataj(m,v)。
8. the mobile Web business cooperation recommender system based on user behavior analysis characterized by comprising wearable terminal, use
Family terminal, gateway, server, wherein
Wearable terminal includes physiological data collection module and the first wireless transport module, wherein physiological data collection module is real
When acquire user every physiological data, user terminal is transmitted to by the first wireless transport module;
User terminal includes the second wireless transport module, browser module, Web feature vector generation module, local data cache
Module, recommending module, third wireless transport module, wherein the second wireless transport module be used for the first wireless transport module into
Row wireless communication;Browser module is used to collect the browsing record of user, shows recommendation business;Web feature vector generation module
Feature extraction is carried out for the browsing record to user, generates Web feature vector;Local data cache module is used for caching
The physiological data of family browsing record and its corresponding Web feature vector, user;Recommending module is for receiving the recommendation from gateway
Business, and shown in browser module in the form of pop-up after the format conversion;Third wireless transport module is used for user terminal
With the wireless communication of gateway;
Gateway includes: the 4th wireless transport module, User IP management module, recommends service buffer module, wherein the 4th wireless biography
Defeated module is used for the wireless communication of user terminal and gateway;User IP management module, for distributing IP address for institute accessing user;
Recommend service buffer module, recommendation business is sent to user's end for caching the recommendation business from server, and periodically
End;
Server includes database module, prediction model building and update module, business recommended module, web resource library, wherein
Database module, for storing the feature vector for representing user browsing behavior and every physiological data;Prediction model building and more
New module is periodically updated for constructing Web service prediction model, and to model;Business recommended module is for executing Web
Traffic forecast, and the high web resource of similarity is found from web resource library according to prediction result and is recommended;Web resource library,
For being stored according to Web feature vector to web resource.
9. the mobile Web business cooperation recommender system according to claim 8 based on user behavior analysis, feature exist
In, between gateway and server pass through wired interface module connect.
10. the mobile Web business cooperation recommender system according to claim 8 based on user behavior analysis, feature exist
In, wearable terminal further includes data preprocessing module, for the physiological data of acquisition to be cleaned and is filtered, removal acquisition
Mistake and error.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610688083.7A CN106375369B (en) | 2016-08-18 | 2016-08-18 | The business recommended method of mobile Web and Collaborative Recommendation system based on user behavior analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610688083.7A CN106375369B (en) | 2016-08-18 | 2016-08-18 | The business recommended method of mobile Web and Collaborative Recommendation system based on user behavior analysis |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106375369A CN106375369A (en) | 2017-02-01 |
CN106375369B true CN106375369B (en) | 2019-05-28 |
Family
ID=57878030
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610688083.7A Active CN106375369B (en) | 2016-08-18 | 2016-08-18 | The business recommended method of mobile Web and Collaborative Recommendation system based on user behavior analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106375369B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108665513B (en) * | 2017-03-27 | 2021-04-06 | 腾讯科技(深圳)有限公司 | Drawing method and device based on user behavior data |
CN106991496B (en) * | 2017-03-29 | 2020-06-30 | 南京邮电大学 | User behavior hierarchical association prediction method oriented to mobile social environment |
CN109117444A (en) * | 2017-06-23 | 2019-01-01 | 宏碁股份有限公司 | The method and correlation data processing system of recommendation service |
CN108629362B (en) * | 2018-05-30 | 2019-11-01 | 华南师范大学 | A kind of learning behavior habit discovery quantization system and method towards mobile environment |
CN109151515B (en) * | 2018-09-12 | 2021-11-12 | 广东乐心医疗电子股份有限公司 | Interaction system and method in performance scene |
CN110008974A (en) * | 2018-11-23 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Behavioral data prediction technique, device, electronic equipment and computer storage medium |
CN109886823A (en) * | 2019-02-25 | 2019-06-14 | 北京奇艺世纪科技有限公司 | A kind of recommended method and device of social circle |
CN110245293B (en) * | 2019-05-28 | 2023-07-11 | 深圳市雅阅科技有限公司 | Network content recall method and device |
CN111914160B (en) * | 2019-10-11 | 2022-06-21 | 宁波大学 | Method for pushing service content of mobile intelligent terminal |
CN111563190B (en) * | 2020-04-07 | 2023-03-14 | 中国电子科技集团公司第二十九研究所 | Multi-dimensional analysis and supervision method and system for user behaviors of regional network |
CN111935231A (en) * | 2020-07-13 | 2020-11-13 | 支付宝(杭州)信息技术有限公司 | Information processing method and device |
CN113536108A (en) * | 2021-05-31 | 2021-10-22 | 珠海大横琴科技发展有限公司 | Information recommendation method and device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101770520A (en) * | 2010-03-05 | 2010-07-07 | 南京邮电大学 | User interest modeling method based on user browsing behavior |
CN101923545A (en) * | 2009-06-15 | 2010-12-22 | 北京百分通联传媒技术有限公司 | Method for recommending personalized information |
CN102215300A (en) * | 2011-05-24 | 2011-10-12 | 中国联合网络通信集团有限公司 | Telecommunication service recommendation method and system |
CN103235824A (en) * | 2013-05-06 | 2013-08-07 | 上海河广信息科技有限公司 | Method and system for determining web page texts users interested in according to browsed web pages |
CN103235826A (en) * | 2013-05-06 | 2013-08-07 | 上海河广信息科技有限公司 | Time window adjustment method |
CN103246725A (en) * | 2013-05-06 | 2013-08-14 | 上海河广信息科技有限公司 | Wireless network based data traffic pushing system and method |
-
2016
- 2016-08-18 CN CN201610688083.7A patent/CN106375369B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101923545A (en) * | 2009-06-15 | 2010-12-22 | 北京百分通联传媒技术有限公司 | Method for recommending personalized information |
CN101770520A (en) * | 2010-03-05 | 2010-07-07 | 南京邮电大学 | User interest modeling method based on user browsing behavior |
CN102215300A (en) * | 2011-05-24 | 2011-10-12 | 中国联合网络通信集团有限公司 | Telecommunication service recommendation method and system |
CN103235824A (en) * | 2013-05-06 | 2013-08-07 | 上海河广信息科技有限公司 | Method and system for determining web page texts users interested in according to browsed web pages |
CN103235826A (en) * | 2013-05-06 | 2013-08-07 | 上海河广信息科技有限公司 | Time window adjustment method |
CN103246725A (en) * | 2013-05-06 | 2013-08-14 | 上海河广信息科技有限公司 | Wireless network based data traffic pushing system and method |
Also Published As
Publication number | Publication date |
---|---|
CN106375369A (en) | 2017-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106375369B (en) | The business recommended method of mobile Web and Collaborative Recommendation system based on user behavior analysis | |
CN105701216B (en) | A kind of information-pushing method and device | |
CN101414296B (en) | Self-adapting service recommendation equipment and method, self-adapting service recommendation system and method | |
CN100401292C (en) | Systems and methods for search query processing using trend analysis | |
CN102859516B (en) | Generating improved document classification data using historical search results | |
CN102622417B (en) | The method and apparatus that information record is ranked up | |
CN107451861B (en) | Method for identifying user internet access characteristics under big data | |
CN102667761B (en) | Scalable cluster database | |
CN101025737B (en) | Attention degree based same source information search engine aggregation display method | |
CN104408043B (en) | A kind of information processing method and server | |
CN105045932B (en) | A kind of data page querying method based on descending storage | |
CN102760151B (en) | Implementation method of open source software acquisition and searching system | |
CN103049440A (en) | Recommendation processing method and processing system for related articles | |
CN103942221B (en) | Search method and equipment | |
CN105760469A (en) | High-dimensional approximate image retrieval method based on inverted LSH in cloud computing environment | |
US20110184815A1 (en) | System and method for sharing profits with one or more content providers | |
CN105760468A (en) | Large-scale image querying system based on inverted position-sensitive Hash indexing in mobile environment | |
CN108282508A (en) | Determination method and device, information-pushing method and the device in geographical location | |
CN109685128A (en) | A kind of MB-kmeans++ clustering method and the user conversation clustering method based on it | |
CN104077293A (en) | Webpage acquisition method and device | |
CN115408618B (en) | Point-of-interest recommendation method based on social relation fusion position dynamic popularity and geographic features | |
CN102053960B (en) | Method and system for constructing quick and accurate Internet of things and Internet search engine according to group requirement characteristics | |
Rosaci et al. | MASHA: A multi-agent system handling user and device adaptivity of Web sites | |
Badis et al. | P2PCF: A collaborative filtering based recommender system for peer to peer social networks | |
Li | Design and implementation of intelligent travel recommendation system based on internet of things. |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |