CN108256119A - A kind of construction method of resource recommendation model and the resource recommendation method based on the model - Google Patents
A kind of construction method of resource recommendation model and the resource recommendation method based on the model Download PDFInfo
- Publication number
- CN108256119A CN108256119A CN201810151601.0A CN201810151601A CN108256119A CN 108256119 A CN108256119 A CN 108256119A CN 201810151601 A CN201810151601 A CN 201810151601A CN 108256119 A CN108256119 A CN 108256119A
- Authority
- CN
- China
- Prior art keywords
- resource
- user
- label
- weight
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000010276 construction Methods 0.000 title claims abstract description 10
- 230000003542 behavioural effect Effects 0.000 claims abstract description 25
- 239000000463 material Substances 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 241001269238 Data Species 0.000 claims description 2
- 239000000284 extract Substances 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 11
- 230000006399 behavior Effects 0.000 description 14
- 238000010586 diagram Methods 0.000 description 9
- 238000013499 data model Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000009329 sexual behaviour Effects 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004883 computer application Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 239000000686 essence Substances 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 238000005406 washing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/955—Retrieval from the web using information identifiers, e.g. uniform resource locators [URL]
- G06F16/9562—Bookmark management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Abstract
The present invention provides a kind of construction method of resource recommendation model and the resource recommendation method based on the model, including：A, each resource information is obtained, and obtains each keyword label of each resource information and its word frequency weight；B, the behavioral data of each user is obtained, and obtains attention rate weight of each user to each resource information；C, chart database model is established, by the storage corresponding with each resource and the attention rate weight of each resource that it is paid close attention to of the weight and each user of the corresponding keyword label of each resource and each keyword label in the chart database model, to build the association between the resource, label and user.By upper, the application does not need to pretreatment cluster process, can flexibly build label, is conducive to carry out resource efficiently in real time to recommend.
Description
Technical field
The invention belongs to computer application technologies, and in particular to a kind of construction method of resource recommendation model and be based on
The resource recommendation method of the model.
Background technology
Database technology has become core and the basis of information system.But as the expansion of data scale and data are complicated
The increase of property, relational model can not meet field needs, by taking social networks as an example, will lead to data using relational database
Redundancy, and the dynamic of social data is not adapted to, multilayer as similar " good friend of good friend " can not be supported well
Complex query.For internal relation between data the problem of complicated and dynamic change, sight is turned to graphic data base by people again,
Graphic data base can effectively be stored, be managed, updating the data and its internal relation, and can efficiently perform multilayer complex operations.
Common graphic data base includes Neo4j, FlockDB, Jena, GraphDB etc..Neo4j is a high performance NOSQL figure
Graphic data library, structural data is stored on network rather than in table by it.It is one it is Embedded, based on disk, tool
The Java persistence engines of standby complete transactional attribute, but structural data is stored in network and (is called from mathematical angle by it
Figure) on rather than table in.Neo4j can also be counted as a high performance figure engine, which has ripe database
All characteristics.
The common way of recommendation has the way of recommendation and content-based recommendation mode based on collaborative filtering.Based on content
Recommendation only considered the nature of object, and object is formed set by label, is pushed away if one in consumption set to you
Recommend other objects in set；Proposed algorithm based on collaborative filtering makes full use of group wisdom, the i.e. row in a large amount of crowd
For with collect answer in data, with help, we obtain entire crowd the conclusion in statistical significance, the personalization level of recommendation
It is high.Cluster similarity analysis mainly was carried out to data by establishing resource model and user model when doing commending system in the past,
Each model scale model corresponding with its is obtained, and is ranked up to obtain recommendation results.These usual data are all stored in figure
In database, but this mode there are one it is apparent the shortcomings that be exactly that recommendation results can not calculate in real time, be typically necessary pretreatment
Cluster process can not flexibly build label, can not complete efficiently to calculate, and be unfavorable for efficiently recommending resource.
Therefore, at present there is an urgent need for a kind of resource recommendation method, in order to efficiently be recommended resource.
Invention content
In view of this, the application provides a kind of construction method of resource recommendation model and the resource recommendation side based on the model
Method when carrying out resource recommendation, does not need to pretreatment cluster process, can flexibly build label, be conducive to resource high-efficiency
Recommended in real time.
The application provides a kind of construction method of resource recommendation model, including：
A, each resource information is obtained, and obtains each keyword label of each resource information and its word frequency weight；
B, the behavioral data of each user is obtained, and obtains each user and the attention rate of each resource information is weighed
Weight；
C, chart database model is established, by the corresponding keyword label of each resource and the power of each keyword label
The figure is arrived in weight and each user storage corresponding with each resource and the attention rate weight of each resource that it is paid close attention to
In database model, to build the association between the resource, label and user.
By upper, the resource recommendation model of the application structure does not need to pretreatment cluster process, can flexibly build label,
Be conducive to carry out resource efficiently in real time to recommend.
Preferably, the step of each keyword label and its word frequency weight that each resource information is obtained described in step A, wraps
It includes：
For each resource information, each information that the resource information is included merges generation long text information, so
Word segmentation processing is carried out to the Wen Changben information of generation afterwards, and is preserved and to form language material；
For the language material that each resource is formed, the language material is analyzed by TF-IDF algorithms, extracts each money
Each keyword label of source information and its word frequency weight.
By upper, be conducive to obtain each keyword label of each resource information and its word frequency weight.
Preferably, the step of attention rate weight of each user to each resource information is obtained described in step B includes：
For each resource, according to user to the behavioral data of the resource, different behavioral datas has different
Weight, and the time corresponding time decay factor of the behavioral data generation with reference to the user, calculate the user to institute
State the attention rate weight of resource.
By upper, by considering above-mentioned each condition, be conducive to accurately acquire attention rate of the user to resource.
Preferably, the user is to the calculation formula of the attention rate weight of the resource：
W=WBehavior 1+WBehavior 2+…WBehavior i…+WBehavior n……
Wherein, WBehavior iIt is to represent the weight that i-th kind of behavioral data is calculated；
Wherein, WBehavior i=ai* (c0*bday0+c1*bday1+…cj*bdayj+…+cm*bdaym……)
Wherein, ai is the behavior weight of corresponding behavioral data i；B is time decay factor, and value is less than 1；Dayj is described
Day number of days j so far or preceding jth day occur for behavioral data；Whether cj values occur the behavioral data for corresponding preceding jth day
I, its value is 1 if occurring, and does not occur to be 0.
By upper, by above-mentioned formula, be conducive to accurately acquire concern of the user to resource.
Preferably, the behavioral data includes at least one：It clicks, read, purchase, collecting.
Preferably, the step B is further included：
Obtain the user characteristics label of each user；
The step D is further included：User characteristics label is corresponded in user's storage to chart database model.
Preferably, the user characteristics label includes at least one：Personal basic attribute label, self or others
The corresponding keyword label of resource of evaluation label, user preferences to oneself；
Wherein, the content of the personal basic attribute label includes at least one：Gender, age, education degree, duty
Industry, hobby, residence, company, contact method；
Wherein, the content of the evaluation label includes at least one：Impression label, credit appraisal label.
The application also provides a kind of resource recommendation method of resource recommendation model built based on the above method, including：
A8, by each keyword mark under target user's each resource corresponding to the attention rate weight of each resource
The word frequency multiplied by weight of label obtains each product value, and each product value is ranked up from high to low；
B8, the keyword label that will be greater than corresponding to the product value of specified threshold recommend target user；
C8, to the corresponding resource of keyword label described in target user's recommendation step B8.
By upper, the application is conducive to carry out efficiently accurate resource recommendation.
The application also provides a kind of resource recommendation method of resource recommendation model built based on the above method, including：
The each user for the resource that A9, acquisition and target user pay close attention to jointly is as user to be recommended；It obtains and respectively treats respectively
The quantitative value of resource that recommended user pays close attention to jointly with target user；And each quantitative value is ranked up from high to low；
B9, user to be recommended corresponding to the quantitative value of specified threshold is will be greater than as similar users, by similar users
User information recommends target user；The user information includes at least one：User's name, the resource of user's concern.
By upper, the application is conducive to efficiently accurately to target user recommend the user information of similar users.
The application also provides a kind of resource recommendation method of resource recommendation model built based on the above method, including：
A10, by each keyword mark under each user each resource corresponding to the attention rate weight of each resource
Word frequency label multiplied by weight obtain each product value, and each product value is ranked up from high to low；
B10, keyword label corresponding to the product value of specified threshold is will be greater than as one in user characteristics label
Point；
C10, acquisition and target user have each user of common user characteristics label as user to be recommended；Respectively
Obtain the quantitative value of each user to be recommended and the common user characteristics label of target user；And by each quantitative value from high to low into
Row sequence；
D10, using the quantity be more than specified threshold similar users to be recommended as similar users, by the use of similar users
Family information recommendation is to target user；The user information includes at least one：User's name, the resource of user's concern.
By upper, the application also helps the user information for efficiently accurately recommending its similar users to target user.
In conclusion resource recommendation model and resource recommendation method that the application provides, are carrying out resource and similar users
When information (similar users information is also one kind of resource) is recommended, pretreatment cluster process is not needed to, can flexibly build mark
Label complete efficient calculating, are conducive to carry out resource efficiently in real time to recommend.
Description of the drawings
Fig. 1 is that a kind of construction method of resource recommendation model provided by the embodiments of the present application and the resource based on the model push away
Recommend the flow diagram of method；
Fig. 2 a and Fig. 2 b are the schematic diagram provided by the embodiments of the present application for providing a user keyword label calculating process；
For the schematic diagram provided by the embodiments of the present application for providing a user keyword label calculating process；
Fig. 3 is the structure diagram of resource recommendation model provided by the embodiments of the present application.
Specific embodiment
The application is illustrated below in conjunction with the attached drawing in the embodiment of the present application.
A kind of resource recommendation method provided by the embodiments of the present application, wherein, including the data based on Neo4j diagram data models
The construction step in library and the resource recommendation step based on the database.
Wherein, as shown in Figure 1, the construction step of the database based on Neo4j diagram data models includes the following steps：
S101 obtains each resource information to be recommended, and obtain each resource information each keyword label and its
Word frequency weight；Specifically, including：
S101.1, it obtains need to do the resource information recommended first, such as can be news, article, e-book etc.,
Can be information in webpage etc., the best quantity of resource is more, wide variety, wherein the information of each resource includes：Text, first number
It is believed that breath, evaluation of the user to resource and the label beaten to resource.
S101.2, for each resource information, each information which is included merges generation long text information, so
This information is grown to this article of generation afterwards and carries out word segmentation processing, and preserved and to form language material；
S101.3, the language material formed for each resource analyze the language material by TF-IDF algorithms, determine
Go out the weight of keyword that each resource includes and each keyword, and be more than the keyword of threshold value as its corresponding money for weight
The keyword label in source.These keywords under the resource are the keyword label for constituting the resource.
Wherein, the main thought of TF-IDF is：If the frequency TF high that some word or phrase occur in an article, and
And seldom occur in other articles, then it is assumed that this word or phrase have good class discrimination ability, are adapted to classify.
In the present embodiment, using TF-IDF algorithms calculate respectively each word in language material TF (word frequency, TF, Term Frequency),
IDF (reverse document-frequency, IDF, Inverse Document Frequency), then by TF and IDF, the two values are multiplied, i.e.,
It can obtain the TF-IDF values as the word weight.Wherein, some word is higher to the importance of article, its TF-IDF values will
Bigger namely weight is bigger.Wherein：
TF refers to word frequency, and in the given file of portion, word frequency refers to that some given word goes out in this document
Existing frequency, therefore the TF of certain word can be by all word numbers of number divided by this document that the word occurs in this document
It obtains.This frequency values would generally be normalized, it to be prevented to be biased to long file.IDF refers to reverse document-frequency, is one
The measurement of word general importance.The IDF of a certain particular words, can file by general act number divided by comprising the word
Number, then obtained quotient is taken the logarithm to obtain.
Wherein, the above process can also use TextRank algorithm or artificial notation methods to carry out keyword label to resource
Setting.
S102 obtains the user behavior data of target user, and calculates user and each resource information to be recommended accordingly
Attention rate, which can be represented (that is, attention rate weight) with weight.
Specifically, according to user to the behavioral data of certain resource, when different weights is given to different behaviors, and being added in
Between decay factor calculate final behavior weight, as the user to the attention rate of certain resource；Wherein, the behavioral data is at least
Including but not limited to following one：Click, read, collection, forward, buy etc.；
Wherein, the calculation of attention rate weight is exemplified below, user's read resource behavior weight is denoted as 0.3, uses
Family collection resource weight is denoted as 0.6, and time decay factor can use 0.9/ day, and a user read resource, today yesterday
Resource is collected, then user is to the attention rate of the resource：0.3*0.91+0.6*0.90=0.27+0.6=0.87.If the use
Yesterday was read resource at family, had also read resource today, and collected resource today, then user is to the attention rate weight of the resource
For：0.3*0.91+0.3*0.90+0.6*0.90=0.27+0.3+0.6=1.17.
The attention rate weight equation can arrange as follows：
User is to the attention rate weight W=W of certain resourceBehavior 1+WBehavior 2+…WBehavior i…+WBehavior n... [formula 1].
Wherein, WBehavior iIt is to represent the attention rate weight that i-th kind of behavioral data is calculated, following formula is used to calculate：
WBehavior i=ai* (c0*bday0+c1*bday1+…cj*bdayj+…+cm*bdaym) ... [formula 2].
Wherein, ai is the behavior weight of corresponding behavioral data i；B is time decay factor, and value is less than 1；Dayj is the row
For day number of days j so far or preceding jth day occurs, what today occurred was considered as to 0 day, i.e. day0=0 today, what yesterday occurred
For to 1 day today, i.e. day1=1；Whether cj values occur the behavioral data i for corresponding preceding jth day, its value is if occurring
1, do not occur to be then 0, be not counted in the calculating of this day attention rate, such as first 3rd day does not have behavior data i, then c3=0,
The behavior data i attention rate of first 3rd day so as to calculate is 0, that is, represents that this day attention rate is not included in.In addition, daym
It can limit as needed and such as be up to day5, i.e., only consider the behavioral data in 5 days.
In addition, recommendation in order to more be met or more accurately recommending, the present invention is also to the calculating of attention rate
Formula is further improved as follows：For non-general consumption class resource, behavior is marked off and terminates sexual behaviour classification, i.e. category row
To occur, attention rate can decline rapidly, such as large-scale resource, as household appliances (TV, washing machine, computer), furniture (bed,
Bookcase) etc. for non-common consumer product, once completing buying behavior, the possibility that user buys similar products again is relatively low, because
This improves the attention rate weight calculation formula of the correspondence formula 1 of this kind of resource as follows：
User is to attention rate weight=e of non-general consumption class resourcez* (attention rateBehavior 1+ attention rateBehavior 2+ ... concern
DegreeBehavior i...+attention rateBehavior n) ... [formula 3].
Wherein, e represents the decay factor when behavior of the termination sexual behaviour classification occurs, and the class behavior is represented less than 1, z
Whether occur, z=1 or certain natural number when occurring, be 0 when not occurring.In this way, when the TV as resource as above completes purchase row
To be rear, decay factor e causes the user calculated to meet regular situation to the attention rate meeting rapid decay of the TV, and not
When there is TV buying behavior, then e0=1, the formula 3 is identical with formula 1, i.e., does not influence the calculating of former attention rate.Wherein, compared with
Good, the probability that z values can again occur in a short time with the behavior of the termination sexual behaviour classification is inversely proportional.
Wherein, S102 is further included：
Obtain the corresponding behavioral data of each similar users to be recommended；And the money of similar users concern to be recommended is obtained accordingly
Source and resource attention rate weight；
Obtain the user characteristics label of target user and similar users to be recommended.
The application, which further includes, obtains target user and the corresponding each user tag of similar users to be recommended, Yong Hubiao
Label mainly include：Personal basic attribute label, others' evaluation label to oneself, the corresponding key of resource of user preferences browsing
Word label etc..The content of personal basic attribute label may include：Gender, the age, education degree, occupation, hobby, residence,
Company, contact method etc..Evaluation label may include：Impression label (such as Duo Shou parties, Xue Shengdang, small pure and fresh, angry youth), credit
Evaluate label (such as credit, public praise, professional skill) etc..
S103, establish between each resource that target user is paid close attention to it to be associated with and establish each resource right with it
Association between the keyword label answered；
Wherein, S103 is further included：
Establish the association between the resource of each corresponding concern of similar users to be recommended.
S104 establishes chart database model (for example, Neo4j database models), by each steps of above-mentioned S101 to S103
Obtained data, such as the weight of the corresponding keyword label of each resource and each keyword label, each resource
Association between corresponding keyword label；And the concern of each resource and each resource that target user pays close attention to it
It spends in the associated storage to the chart database model between weight and target user and its each resource paid close attention to.
Wherein, S104 is further included：
It is each to treat by the resource of the corresponding concern of similar users to be recommended and the attention rate weight of each resource
Recommend the incidence relation between the resource of the corresponding concern of similar users；And the user tag of each similar users to be recommended is deposited
It stores up in the chart database model.
Wherein, S104 is further included：
Target user and the corresponding each user information of similar users to be recommended and user characteristics label storage are arrived
In chart database model.
The above process can use the SDK of the various language of Neo4j offers or directly be carried out using restful API
Data import.
As shown in figure 3, illustrate the model.The figure (Graph) that Neo4j is created is based on attribute graph model, in the model,
Each entity has ID (Identity) unique mark (that is, User ID), and each node is grouped by label (Lable), Mei Geguan
All there are one unique type, the basic conception of attribute graph model has for system：Node (Node), relationship (Relationship and category
Property (Property).That is, there is relationship between node and other nodes, node itself has attribute.For example, between resource and label
Relevant, relevant between user and resource, relevant between user and label, each node also has the attribute of its own.
Compared to relevant database, Neo4j database sharings can be carried out after completing by Cypher sentences real-time query operation (or
Person directly carries out digital independent using Restful API).This process does not need to that data progress cluster calculation is obtained recommending locating in advance
Manage result.
So far step completes the structure of the database based on Neo4j diagram data models.Then, you can be based on
The resource recommendation of the database, it is specific as follows：
S105, according to the target user stored in the Neo4j databases to the attention rate weight of each resource and each
The word frequency weight of the corresponding keyword label of resource and each keyword label recommends the key of each resource to target user
Word label；And the quantity according to the resource paid close attention to jointly with target user, recommend the phase similar to its interest to target user
Like user.
S105.1, wherein, about " according to target user to the attention rate of each resource, to each money of target user's recommendation
The keyword label in source ", specifically includes：
A1, obtain the target user that is stored in the Neo4j databases of aforementioned structure to the attention rate weight of each resource with
Each keyword label weight under its corresponding each resource；
A2, by each keyword mark under target user's each resource corresponding to the attention rate weight of each resource
It signs multiplied by weight and obtains each product value, and each product value is ranked up from high to low；
A3, the keyword label that will be greater than corresponding to the product value of the first specified threshold (value can be set) recommend mesh
User is marked, using the keyword label as the part in the user characteristics label of target user；
A4, recommend the corresponding resource of the keyword label to target user.
It is illustrated below：
As shown in Figure 2 a, it is assumed that target user U, resource to be recommended are A (its attention rate weight paid close attention to by user U
0.8), B (its attention rate weight paid close attention to by user U is 0.7).Wherein, corresponding keyword label a1 (the word frequency weights of resource A
0.8), a2 (word frequency weight 0.6), a3 (word frequency weight 0.7)；The corresponding keyword label b1 of resource B (word frequency weight 0.5), b2
(word frequency weight 0.7), b3 (word frequency weight 0.9).
As shown in Figure 2 b, by user U to A resources, the attention rate weight of B resources and each key under A resources, B resources
Word label multiplied by weight obtains each product value.
Each product value is ranked up from high to low：a1(0.64)；b3(0.63)；a3(0.56)；b2(0.49)；a2
(0.48)；b1(0.35).
Will be greater than the first specified threshold, (value can be set, such as the threshold value is set as corresponding to product value 0.5)
Keyword label recommends target user：Such as keyword label a1, b3, a3 are recommended into user.And by the keyword label
Corresponding resource recommendation to target user, wherein, a keyword label may correspond to multiple resources herein, can all push away it
It recommends to user.
S105.2, wherein, about " according to the quantity for the resource paid close attention to jointly with target user, to target user recommend with
The similar similar users of its interest ".It specifically includes：
The resource of the corresponding concern of each user stored in B1, the Neo4j databases according to aforementioned structure, obtains it
The quantitative value of resource that his user's (similar users to be recommended) pays close attention to jointly with target user respectively；And by each quantitative value by height
It is ranked up to low；
B2, the use that will be greater than similar users to be recommended corresponding to the quantitative value of the second specified threshold (value can be set)
Family information, such as the resource recommendation that user's name (User ID can be used to represent) and corresponding similar users are paid close attention to is to target
User.
Wherein, it about " recommending the similar users similar to its interest to target user ", further includes：
A ', by each pass under similar users to be recommended each resource corresponding to the attention rate weight of each resource
Keyword label multiplied by weight obtains each product value, and each product value is ranked up from high to low；
B ', the keyword label that will be greater than corresponding to the product value of specified threshold are special as the user of similar users to be recommended
Levy the part in label；
In the user characteristics label of C ', the user characteristics label for obtaining each similar users to be recommended respectively and target user
Same label quantity；And the quantity is more than to the user information of the similar users to be recommended of specified threshold, such as user
The resource recommendation that title (User ID can be used to represent) and corresponding similar users are paid close attention to is to target user.
Wherein, the Neo4j diagram data model databases based on aforementioned structure determine target user and are browsing certain resource,
Such as during certain e-book, when recommending the corresponding keyword label of related resource to target user, following Cypher may be used and hold
Line statement：
MATCH
(user:USER)-[favour:FAVOUR]->(book:BOOK)-[mark:MARK]->(tag:TAG)WHERE
User.userid={ userid } RETURN user.gender as gender, tag.name AS tag, SUM
(favour.score*mark.score)AS score ORDER BY score DESC。
Wherein, the Neo4j diagram data model databases based on aforementioned structure when judging that user browses certain resource, recommend phase
During like user to target user, following Cypher may be used and perform sentence：
MATCH(user:USER)-[f2:FAVOUR]->(book:BOOK)<-[f1:FAVOUR]-(users:USER)
WHERE user.userid={ userid } RETURN users.name, count (*) as Strength order by
Strength DESC。
In conclusion the model of the application can directly recommend the pass to it by the keyword label that target user pays close attention to
The corresponding resource information of keyword label, the application do not need to pretreatment cluster process, can flexibly build label, complete efficient
Calculating, be conducive to efficiently carry out real-time recommendation.Also, the application pays close attention to money jointly also according to similar users and target user
The quantity of source information recommends the user information of similar users to user；Or according to target user and the common use of similar users
Family feature tag quantity to target user recommend similar users user information (including by the resource recommendation that similar users are paid close attention to
User), therefore the application can carry out efficient resource recommendation.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
With within principle, any modification, equivalent replacement, improvement and so on should all be included in the protection scope of the present invention god.
Claims (10)
1. a kind of construction method of resource recommendation model, which is characterized in that including：
A, each resource information is obtained, and obtains each keyword label of each resource information and its word frequency weight；
B, the behavioral data of each user is obtained, and obtains attention rate weight of each user to each resource information；
C, establish chart database model, by the weight of the corresponding keyword label of each resource and each keyword label,
And the figure number is arrived in each user storage corresponding with each resource and the attention rate weight of each resource that it is paid close attention to
According in the model of library, to build the association between the resource, label and user.
2. according to the method described in claim 1, it is characterized in that, each key of each resource information is obtained described in step A
The step of word label and its word frequency weight, includes：
For each resource information, each information that the resource information is included merges generation long text information, then right
The Wen Changben information of generation carries out word segmentation processing, and is preserved and to form language material；
For the language material that each resource is formed, the language material is analyzed by TF-IDF algorithms, extracts each resource letter
Each keyword label of breath and its word frequency weight.
3. method according to claim 1 or 2, which is characterized in that each user is obtained described in step B to each money
The step of attention rate weight of source information, includes：
For each resource, according to user to the behavioral data of the resource, different weights that different behavioral datas has,
And the time corresponding time decay factor of the behavioral data generation with reference to the user, the user is calculated to the resource
Attention rate weight.
4. according to the method described in claim 3, it is characterized in that, calculating of the user to the attention rate weight of the resource
Formula is：
W=WBehavior 1+WBehavior 2+…WBehavior i…+WBehavior n……
Wherein, WBehavior iIt is to represent the weight that i-th kind of behavioral data is calculated；
Wherein, WBehavior i=ai* (c0*bday0+c1*bday1+…cj*bdayj+…+cm*bdaym……)
Wherein, ai is the behavior weight of corresponding behavioral data i；B is time decay factor, and value is less than 1；Dayj is the behavior
Day number of days j so far or preceding jth day occur for data；Whether cj values occur the behavioral data i for corresponding preceding jth day, if
It is 1 that then its value, which occurs, does not occur to be then 0.
5. according to the method described in claim 4, it is characterized in that, the behavioral data includes at least one：It clicks, read
It reads, purchase, collect.
6. according to the method described in claim 1, it is characterized in that, the step B is further included：
Obtain the user characteristics label of each user；
The step D is further included：User characteristics label is corresponded in user's storage to chart database model.
7. according to the method described in claim 6, it is characterized in that, the user characteristics label includes at least one：It is a
People's essential attribute label, self or others are to the evaluation label of oneself, the corresponding keyword label of resource of user preferences；
Wherein, the content of the personal basic attribute label includes at least one：Gender, the age, education degree, occupation,
Hobby, residence, company, contact method；
Wherein, the content of the evaluation label includes at least one：Impression label, credit appraisal label.
8. a kind of resource recommendation method of the resource recommendation model based on claim 1-7 either method structure, which is characterized in that
Including：
A8, by each keyword label under target user's each resource corresponding to the attention rate weight of each resource
Word frequency multiplied by weight obtains each product value, and each product value is ranked up from high to low；
B8, the keyword label that will be greater than corresponding to the product value of specified threshold recommend target user；
C8, to the corresponding resource of keyword label described in target user's recommendation step B8.
9. a kind of resource recommendation method of the resource recommendation model based on claim 1-7 either method structure, which is characterized in that
Including：
The each user for the resource that A9, acquisition and target user pay close attention to jointly is as user to be recommended；It obtains respectively each to be recommended
The quantitative value of resource that user pays close attention to jointly with target user；And each quantitative value is ranked up from high to low；
B9, user to be recommended corresponding to the quantitative value of specified threshold is will be greater than as similar users, by the user of similar users
Information recommendation is to target user；The user information includes at least one：User's name, the resource of user's concern.
10. a kind of resource recommendation method of the resource recommendation model based on claim 1-7 either method structure, feature exist
In, including：
A10, by each keyword target word under each user each resource corresponding to the attention rate weight of each resource
Frequency label multiplied by weight obtains each product value, and each product value is ranked up from high to low；
B10, keyword label corresponding to the product value of specified threshold is will be greater than as the part in user characteristics label；
C10, acquisition and target user have each user of common user characteristics label as user to be recommended；It obtains respectively
Each user to be recommended and the quantitative value of the common user characteristics label of target user；And each quantitative value is arranged from high to low
Sequence；
D10, the similar users to be recommended that the quantity is more than to specified threshold believe the user of similar users as similar users
Breath recommends target user；The user information includes at least one：User's name, the resource of user's concern.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810151601.0A CN108256119B (en) | 2018-02-14 | 2018-02-14 | Resource recommendation model construction method and resource recommendation method based on model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810151601.0A CN108256119B (en) | 2018-02-14 | 2018-02-14 | Resource recommendation model construction method and resource recommendation method based on model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108256119A true CN108256119A (en) | 2018-07-06 |
| CN108256119B CN108256119B (en) | 2021-12-28 |
Family
ID=62745255
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810151601.0A Expired - Fee Related CN108256119B (en) | 2018-02-14 | 2018-02-14 | Resource recommendation model construction method and resource recommendation method based on model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108256119B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109559152A (en) * | 2018-10-24 | 2019-04-02 | 深圳市万屏时代科技有限公司 | A kind of network marketing method, system and computer storage medium |
| CN109857761A (en) * | 2018-12-28 | 2019-06-07 | 珍岛信息技术（上海）股份有限公司 | A kind of database optimizing method and its system |
| CN110263161A (en) * | 2019-05-29 | 2019-09-20 | 阿里巴巴集团控股有限公司 | A kind of processing method of information, device and equipment |
| CN110727784A (en) * | 2019-09-05 | 2020-01-24 | 上海异势信息科技有限公司 | Article recommendation method and system based on content |
| CN110990698A (en) * | 2019-11-29 | 2020-04-10 | 珠海大横琴科技发展有限公司 | Recommendation model construction method and device |
| CN111125086A (en) * | 2018-10-31 | 2020-05-08 | 北京国双科技有限公司 | Method, device, storage medium and processor for acquiring data resources |
| CN111177129A (en) * | 2019-12-16 | 2020-05-19 | 中国平安财产保险股份有限公司 | Label system construction method, device, equipment and storage medium |
| CN111488453A (en) * | 2019-01-25 | 2020-08-04 | 北京猎户星空科技有限公司 | Resource grading method, device, equipment and storage medium |
| CN111737588A (en) * | 2020-08-24 | 2020-10-02 | 南京国睿信维软件有限公司 | User portrait knowledge similarity calculation method |
| CN112069388A (en) * | 2020-09-02 | 2020-12-11 | 上海风秩科技有限公司 | Entity recommendation method, system, computer device and computer-readable storage medium |
| CN112115368A (en) * | 2020-09-29 | 2020-12-22 | 安徽访得信息科技有限公司 | Method for content information distribution engine based on big data |
| CN112287179A (en) * | 2020-06-30 | 2021-01-29 | 浙江好络维医疗技术有限公司 | Patient identity matching method combining connection priority algorithm and graph database |
| CN112380452A (en) * | 2021-01-14 | 2021-02-19 | 北京崔玉涛儿童健康管理中心有限公司 | User interest collection method and device in infant content recommendation |
| CN112650948A (en) * | 2020-12-30 | 2021-04-13 | 华中师范大学 | Information network construction method, system and application for education informatization evaluation |
| CN113297457A (en) * | 2021-05-24 | 2021-08-24 | 陕西合友网络科技有限公司 | High-precision intelligent information resource pushing system and pushing method |
| CN113780415A (en) * | 2021-09-10 | 2021-12-10 | 平安科技（深圳）有限公司 | User portrait generation method, device, equipment and medium based on small program game |
| CN116244496A (en) * | 2022-12-06 | 2023-06-09 | 山东紫菜云数字科技有限公司 | Resource recommendation method based on industrial chain |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102867016A (en) * | 2012-07-18 | 2013-01-09 | 北京开心人信息技术有限公司 | Label-based social network user interest mining method and device |
| CN103995839A (en) * | 2014-04-30 | 2014-08-20 | 兴天通讯技术（天津）有限公司 | Commodity recommendation optimizing method and system based on collaborative filtering |
| CN104636448A (en) * | 2015-01-23 | 2015-05-20 | 广东欧珀移动通信有限公司 | Music recommendation method and device |
| CN106095949A (en) * | 2016-06-14 | 2016-11-09 | 东北师范大学 | A kind of digital library's resource individuation recommendation method recommended based on mixing and system |
| CN106447066A (en) * | 2016-06-01 | 2017-02-22 | 上海坤士合生信息科技有限公司 | Big data feature extraction method and device |
-
2018
- 2018-02-14 CN CN201810151601.0A patent/CN108256119B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102867016A (en) * | 2012-07-18 | 2013-01-09 | 北京开心人信息技术有限公司 | Label-based social network user interest mining method and device |
| CN103995839A (en) * | 2014-04-30 | 2014-08-20 | 兴天通讯技术（天津）有限公司 | Commodity recommendation optimizing method and system based on collaborative filtering |
| CN104636448A (en) * | 2015-01-23 | 2015-05-20 | 广东欧珀移动通信有限公司 | Music recommendation method and device |
| CN106447066A (en) * | 2016-06-01 | 2017-02-22 | 上海坤士合生信息科技有限公司 | Big data feature extraction method and device |
| CN106095949A (en) * | 2016-06-14 | 2016-11-09 | 东北师范大学 | A kind of digital library's resource individuation recommendation method recommended based on mixing and system |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109559152A (en) * | 2018-10-24 | 2019-04-02 | 深圳市万屏时代科技有限公司 | A kind of network marketing method, system and computer storage medium |
| CN111125086A (en) * | 2018-10-31 | 2020-05-08 | 北京国双科技有限公司 | Method, device, storage medium and processor for acquiring data resources |
| CN111125086B (en) * | 2018-10-31 | 2023-02-24 | 北京国双科技有限公司 | Method, device, storage medium and processor for acquiring data resources |
| CN109857761A (en) * | 2018-12-28 | 2019-06-07 | 珍岛信息技术（上海）股份有限公司 | A kind of database optimizing method and its system |
| CN109857761B (en) * | 2018-12-28 | 2022-11-11 | 珍岛信息技术（上海）股份有限公司 | Database optimization method and system |
| CN111488453B (en) * | 2019-01-25 | 2024-02-23 | 北京猎户星空科技有限公司 | Resource grading method, device, equipment and storage medium |
| CN111488453A (en) * | 2019-01-25 | 2020-08-04 | 北京猎户星空科技有限公司 | Resource grading method, device, equipment and storage medium |
| CN110263161A (en) * | 2019-05-29 | 2019-09-20 | 阿里巴巴集团控股有限公司 | A kind of processing method of information, device and equipment |
| CN110263161B (en) * | 2019-05-29 | 2023-09-26 | 创新先进技术有限公司 | Information processing method, device and equipment |
| CN110727784A (en) * | 2019-09-05 | 2020-01-24 | 上海异势信息科技有限公司 | Article recommendation method and system based on content |
| CN110727784B (en) * | 2019-09-05 | 2023-11-10 | 上海异势信息科技有限公司 | Article recommendation method and system based on content |
| CN110990698B (en) * | 2019-11-29 | 2021-01-08 | 珠海大横琴科技发展有限公司 | Recommendation model construction method and device |
| CN110990698A (en) * | 2019-11-29 | 2020-04-10 | 珠海大横琴科技发展有限公司 | Recommendation model construction method and device |
| CN111177129A (en) * | 2019-12-16 | 2020-05-19 | 中国平安财产保险股份有限公司 | Label system construction method, device, equipment and storage medium |
| CN111177129B (en) * | 2019-12-16 | 2023-08-08 | 中国平安财产保险股份有限公司 | Method, device, equipment and storage medium for constructing label system |
| CN112287179A (en) * | 2020-06-30 | 2021-01-29 | 浙江好络维医疗技术有限公司 | Patient identity matching method combining connection priority algorithm and graph database |
| CN112287179B (en) * | 2020-06-30 | 2024-02-23 | 浙江好络维医疗技术有限公司 | Patient identity matching method combining connection priority algorithm with graph database |
| CN111737588A (en) * | 2020-08-24 | 2020-10-02 | 南京国睿信维软件有限公司 | User portrait knowledge similarity calculation method |
| CN112069388B (en) * | 2020-09-02 | 2023-07-21 | 上海风秩科技有限公司 | Entity recommendation method, system, computer device and computer readable storage medium |
| CN112069388A (en) * | 2020-09-02 | 2020-12-11 | 上海风秩科技有限公司 | Entity recommendation method, system, computer device and computer-readable storage medium |
| CN112115368A (en) * | 2020-09-29 | 2020-12-22 | 安徽访得信息科技有限公司 | Method for content information distribution engine based on big data |
| CN112650948B (en) * | 2020-12-30 | 2022-04-29 | 华中师范大学 | Information network construction method, system and application for education informatization evaluation |
| CN112650948A (en) * | 2020-12-30 | 2021-04-13 | 华中师范大学 | Information network construction method, system and application for education informatization evaluation |
| CN112380452A (en) * | 2021-01-14 | 2021-02-19 | 北京崔玉涛儿童健康管理中心有限公司 | User interest collection method and device in infant content recommendation |
| CN113297457A (en) * | 2021-05-24 | 2021-08-24 | 陕西合友网络科技有限公司 | High-precision intelligent information resource pushing system and pushing method |
| CN113780415A (en) * | 2021-09-10 | 2021-12-10 | 平安科技（深圳）有限公司 | User portrait generation method, device, equipment and medium based on small program game |
| CN113780415B (en) * | 2021-09-10 | 2023-08-15 | 平安科技（深圳）有限公司 | User portrait generating method, device, equipment and medium based on applet game |
| CN116244496A (en) * | 2022-12-06 | 2023-06-09 | 山东紫菜云数字科技有限公司 | Resource recommendation method based on industrial chain |
| CN116244496B (en) * | 2022-12-06 | 2023-12-01 | 山东紫菜云数字科技有限公司 | Resource recommendation method based on industrial chain |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108256119B (en) | 2021-12-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108256119A (en) | A kind of construction method of resource recommendation model and the resource recommendation method based on the model | |
| US9703877B2 (en) | Computer-based evaluation tool for selecting personalized content for users | |
| Perugini et al. | Recommender systems research: A connection-centric survey | |
| Hill et al. | Network-based marketing: Identifying likely adopters via consumer networks | |
| Liu et al. | Analyzing changes in hotel customers’ expectations by trip mode | |
| Wei et al. | A survey of recommendation systems in electronic commerce | |
| CN101410864B (en) | Behavior sighting system | |
| CN102279851B (en) | Intelligent navigation method, device and system | |
| Dembczynski et al. | Predicting ads clickthrough rate with decision rules | |
| Hsu et al. | Mining skewed and sparse transaction data for personalized shopping recommendation | |
| CN102138140A (en) | Information processing with integrated semantic contexts | |
| JP2009099145A (en) | Method for performing discovery of digital information in subject area | |
| CN102160329A (en) | Facilitating collaborative searching using semantic contexts associated with information | |
| Gursoy et al. | Influence maximization in social networks under deterministic linear threshold model | |
| EP2827294A1 (en) | Systems and method for determining influence of entities with respect to contexts | |
| Van Gysel et al. | Reply with: Proactive recommendation of email attachments | |
| CN108664515A (en) | A kind of searching method and device, electronic equipment | |
| Niu et al. | Product hierarchy-based customer profiles for electronic commerce recommendation | |
| CN109559152A (en) | A kind of network marketing method, system and computer storage medium | |
| US8478702B1 (en) | Tools and methods for determining semantic relationship indexes | |
| Lin | Association rule mining for collaborative recommender systems. | |
| Cengiz et al. | Analysis of pre-weighted and post-weighted association rule mining | |
| Velàsquez et al. | Building a knowledge base for implementing a web-based computerized recommendation system | |
| GB2517358A (en) | Recommendation creation system | |
| CN115048503A (en) | User preference label design method based on content analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211228 |