CN107180078A - A kind of method for vertical search based on user profile learning - Google Patents
A kind of method for vertical search based on user profile learning Download PDFInfo
- Publication number
- CN107180078A CN107180078A CN201710263913.6A CN201710263913A CN107180078A CN 107180078 A CN107180078 A CN 107180078A CN 201710263913 A CN201710263913 A CN 201710263913A CN 107180078 A CN107180078 A CN 107180078A
- Authority
- CN
- China
- Prior art keywords
- user
- attribute
- interest
- time
- actual resource
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of method for vertical search based on user profile learning, it is specially：Each property content for individual subscriber interest demand filled in during according to user's registration, determines the interest preference of user, sets up initial user interest model；If user carries out screening inquiry by attribute classification, attribute Query record is increased into individual subscriber interest model and completes model modification；User is calculated to the interest level of all actual resources, is then ranked up and stored from high to low by interest level；Personalized search then according to user select according to keywords or attribute carries out screening inquiry, generate initial Query Result list；For the actual resource in initial Query Result list, with reference to interest level of the user to these actual resources, the actual resource in list is ranked up from high to low by interest level, final Query Result list is generated.The present invention is by the study to user interest, the result for being actively supplied to user's current interest consistent.
Description
Technical field
The present invention relates to a kind of method for vertical search based on user profile learning, more particularly in vertical search website
Middle storage user browses the method for vertical search of information or User behavior, belongs to vertical search technical field.
Background technology
It is the epoch of an information explosion now, information has not been scarce resource, the situation that various information emerge in an endless stream
Under, the notice of user becomes rare on the contrary.The further development of search engine technique, gives user convenience.According to
The newest issues of CNNIC CNNIC《39th China Internet network state of development statistical report》Data,
By in December, 2016, Chinese search engine user scale is up to 6.02 hundred million, and utilization rate is 82.4%, search engine depth integration people
Work intelligence, vertical specialty chemical conversion development trend.On the one hand, the species of search information is more enriched；On the other hand, search engine pin
To user in the search need of different field, more intelligent, comprehensively, professional searching entities are released.Thus search has been triggered to draw
Hold up industry and new vertical, professional development trend occur.
The search channel of user mainly includes free word full-text search, keyword retrieval, systematic searching and other special letters
The retrieval of breath.Compared with traditional search engine, vertical search engine is needed according to content rather than by analyzing between webpage
Linking relationship carry out retrieval result sequence.And personalized vertical search needs to analyze user behavior, and set up use
Family personalized model, the ranking results met individual requirements.At present, most of vertical search engine websites both provide entity
The Advanced Search of attribute selection, but the sort result of screening is often the ascending order or descending according to certain condition.
In existing individuation search method, main flow way of search is to be based on socialization annotation search, is received by user
Hide, mark and the behavior such as shared and the social labeled data produced, directly reflect user to the quality evaluation of webpage and interior
Hold and understand.Web search performance is improved to webpage sorting using social labeled data.
But the information retrieval service that vertical search engine is provided at this stage still suffers from following deficiency.First, vertical search is drawn
Passive query matching can only be carried out by holding up.After only user have submitted query word, search engine can just make a response.Which imply
Two large problems, are on the one hand that user is many times difficult to clearly sum up the demand of oneself, even if on the other hand query word can
To embody user's current demand, but most of vertical search engine websites are not considered to extract the interest of user, and
Personalized search is carried out, or actively makes information recommendation service；Secondly, socialization mark is carried out to resource in offer user to hang down
In straight search engine, generally, easily being marked by rubbish is influenceed：Malicious user may largely use what is occurred extensively
Label carrys out wrongful its influence power of lifting, is even more to bring huge challenge to the validity of search result.
Therefore, need to pursue a kind of personalized vertical search of higher precision in vertical search engine.
The content of the invention
The technical problems to be solved by the invention are：A kind of method for vertical search based on user profile learning is provided, led to
The study to user's search behavior is crossed, the interest of user, the result for being actively supplied to user's current interest consistent is excavated.
The present invention uses following technical scheme to solve above-mentioned technical problem：
A kind of method for vertical search based on user profile learning, comprises the following steps：
Step 1, in the vertical search engine website comprising attribute selection Advanced Search, according to during initialization and user
Interaction, determines the interest preference of user；
Step 2, according to the interest preference of user, initialising subscriber interest model, initialising subscriber interest model table are set up
It is shown as：xi=(attribute value：The number of times that attribute value is paid close attention to by user-attribute value it is the last by user pay close attention to when
Between), xiRepresent ith attribute；
Step 3, if user is inquired about according to attribute, the attribute value selected according to user, to user interest model
Carry out real-time update；If user is inquired about according to keyword, user interest model is not updated；
Step 4, if navigation patterns occur for user, the browsing time is set as threshold value, if browsing the time of certain actual resource
More than threshold value, then the property value of the actual resource is obtained, and real-time update is carried out to user interest model；
Step 5, using multiple attribute utility theory, with reference to the preference heterogeneity and forgetting factor of user, to all actual resources
Carry out interest level calculating and be ranked up from high to low by interest level；
Step 6, according to user select according to keywords or attribute carries out screening inquiry, generate initial Query Result row
Table；
Step 7, for the actual resource in initial Query Result list, with reference to interest of the user to these actual resources
Angle value, the actual resource in initial Query Result list is ranked up from high to low by interest level, and generation is final to look into
Ask the results list.
As a preferred embodiment of the present invention, detailed process described in step 1 is：First entered into user comprising attribute sieve
When selecting the vertical search engine website of Advanced Search, interacted for each attribute with user, obtain user on each attribute
Interest preference.
As a preferred embodiment of the present invention, if user is inquired about according to attribute described in step 3, selected according to user
The attribute value selected, carrying out real-time update detailed process to user interest model is：If the attribute value of user's selection exists
In user interest model, then the attribute value plus 1 by the number of times that user pays close attention to, and update the attribute value it is the last by with
The time of family concern, otherwise, the number of times that the attribute value, the attribute value are paid close attention to by user and the attribute value are the last
The time paid close attention to by user is increased in user interest model.
As a preferred embodiment of the present invention, the detailed process of the step 5 is：
Step 51, preference heterogeneity is calculated using AF-IDF algorithms, calculation formula is：
ki=AF (X/T) * loga(N/DF(ki))
Wherein, kiPreference heterogeneity of the user to j-th of actual resource ith attribute is represented, AF (X/T) is user in the cycle
To the average time of j-th of actual resource ith attribute concern in T, X is pays close attention to the total degree of the attribute, and N draws for vertical search
Hold up the sum of actual resource in website, DF (ki) it is total with j-th actual resource ith attribute value identical actual resource
Number, a is hyper parameter；
Step 52, forgetting factor f is introduced in preference heterogeneityi, then the preference heterogeneity k introduced after forgetting factori' be：
k′i=ki*fi
Step 53, set attribute that vertical search engine website includes as n, i.e. i=1,2 ..., n, then user is to jth
The interest level of individual actual resource is represented with multiplication utility function：
Uj=k '1+k′2+…+k′n+k′1k′2+k′1k′3+…+k′1k′n+k′2k′3+k′2k′4+…+k′2k′n+k′3k′4+
k′3k′5+…+k′3k′n+……+k′n-1k′n
Wherein, UjRepresent interest level of the user to j-th of actual resource；
Step 54, the interest level of all actual resources is ranked up from high to low, and exports ranking results.
As a preferred embodiment of the present invention, the forgetting factor fiCalculation formula is：
fi=exp (- logb(t)/f)
Wherein, t be j-th actual resource ith attribute value the last time time and current time that are concerned it
Difference, f is half-life period, and b is hyper parameter.
The present invention uses above technical scheme compared with prior art, with following technique effect：
1st, the order of magnitude of inventive algorithm complexity is smaller, if attribute number is n, each actual resource is only in n2Level is with regard to energy
Calculate the interest level of the actual resource.
2nd, under most of calculating of inventive algorithm is online, fraction is calculated in real time, even if the reality of the vertical search system
Body resource and registered user are more, and the operation that general enterprises level server can fully meet algorithm is realized.
3rd, inventive algorithm highly versatile, can be realized, and have good extendibility on various vertical search engines.
Brief description of the drawings
Fig. 1 is a kind of schematic flow sheet of the method for vertical search based on user profile learning of the present invention.
Embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the drawings.Below by
The embodiment being described with reference to the drawings is exemplary, is only used for explaining the present invention, and is not construed as limiting the claims.
As shown in figure 1, being a kind of schematic flow sheet of the method for vertical search based on user profile learning, specific steps are such as
Under：
Step 1：In the vertical search engine website comprising attribute selection Advanced Search, according to initialization when with user hand over
Mutually determine the interest preference of user, occur the problem of to avoid cold start-up.
Step 2：Initialising subscriber interest model is set up, and records and browse note by the attribute selection inquiry of user in real time
Record is adjusted to model.User is inquired about and navigation patterns record using attribute selection each time, by attribute query information and clear
The actual resource attribute information look at is increased in user individual model.
Step 3：Using multiple attribute utility theory, with reference to the preference heterogeneity of user, all actual resources are carried out on backstage
Personalized interest angle value is calculated, and generates the actual resource sorted lists of user personalized interest.Specially：
Step 301：If the n given attribute in vertical search engine, the personalization preferences the effective equation of user can be with
Described with multiplication utility function：Uj=k1+k2+……+kn+k1k2+k2k3+k1k3+……+kn-1kn.Wherein UjRepresent j-th
Entity is in the interest level of active user, kiRepresent preference heterogeneity of the user to the ith attribute value of entity.Formulae express is every
Individual entity property value preference heterogeneity independently calculate after results added, and each entity property value preference heterogeneity will with it is other every
Individual entity property value preference heterogeneity is added again after carrying out product.
Step 302：Calculate preference heterogeneity kiDifferent to the interest-degree of entity based on user, the present invention is using average
Average Frequency/Inverse Document Frequency (AF-IDF) algorithms characterize user to the i-th of entity
Individual property value fancy grade, ki=AF (X/T) * loga(N/DF(ki)), wherein AF (X/T) be user within the T cycles to entity
The average time (inquiry or number of visits) of the concern of ith attribute value, X is the property value total degree of concern, and n is that this is vertical
The sum of entity in search system, DF (ki) it is that attribute value is kiEntity number, calculates after each attribute weight, obtains successively
Interest preference vector of the user based on attributive classification.
Step 303：The interest of user is often to change, and interest forgetting factor f will be introduced in preference heterogeneity, i.e.,：k′i
=ki*fi, fi=exp (- logb(t)/f), wherein t is that property value last time is concerned about the current time difference, and f is partly to decline
Phase.
Step 304：Personalized interest angle value is calculated successively to each entity in vertical search system, until last
Entity is calculated and finished.
Step 305：Sorted from high to low according to personalized interest angle value, and export ranking results, deposit user
Under people's interest model.
Step 4：According to user's search condition, inquired about according to keyword or attribute selection, generate initial Query Result row
Table.
Step 5：For the actual resource in initial query the results list for having generated, with reference to the reality of user personalized interest
Body resource sorted lists, the actual resource in initial query the results list is resequenced from high to low according to interest level.
By taking the search of certain domestic full-sized car Vertical Website as an example (but being not limited only to this example), the website is searched except there is one
Rope frame provides user and according to keywords searches for outer, also provides one and precisely selects car module by classification, including by price, by vehicle,
By brand, by 4 attributes of discharge capacity.
S1, to realize personalized search, it is necessary to learn user interest.Firstly the need of user's registration information, individual character is set up
Change model.The content of 4 base attributes for individual subscriber interest demand is filled in during user's registration.According to initialization when with use
Family interaction determines the interest preference of user, occurs the problem of to avoid cold start-up.The initialization personalized interest model of user
It is made up of the vector of 4 base attributes.
S2, the initialization personalized interest model of hypothesis user 1 are：
Price=(8-10 ten thousand:1-t), vehicle=(SUV:1-t), brand=(Great Wall:1-t) discharge capacity=(1.1-1.6L:1-
t)。
Note：The value of price attribute vector is that " 8-10 ten thousand ", " 1 " is its value, is represented by user's concern once, and "-" is company
Symbol is connect, " t " is the last time paid close attention to by user.
S3, the inquiry mode of user have two kinds, if being screened according to attribute classification, show user pass other to the Attribute class
Note, embodies the preference of user, it is necessary to be updated individual subscriber interest model.If not screened according to attribute classification, that is,
Keyword query, then do not record the preference of user, and individual subscriber interest model is not updated yet.
Renewal process is to record the search of user or navigation patterns each time, is stored into customized information record, with complete
Into the renewal of customized information.Personal interest is not fixed, but dynamic change.With individual subscriber itself and it is outer
Influence, the preference of user can shift.It is probably that the user has just joined when user's initialization by taking scene as an example
Processing is made, and the requirement to car is to ride instead of walk, may be to the less demanding of price, but carrying with personal growth and quality of life
Height is, it is necessary to lift the requirement to car, and at this moment the personalized interest of user just changes, then the search of user and browsing content
Also it can change.
S4, selected attribute classification increased in user individual model, to the situation of existing classification, the increase of its value
1, and update the time finally paid close attention to.
Assuming that user 1, initiating the other screening inquiry of Attribute class is：
Price：8-10 ten thousand；Vehicle：Two box；Brand：Ford；Discharge capacity：1.1-1.6L.
Then corresponding personalized model be updated for：Price=(8-10 ten thousand:2-t), vehicle=(SUV:1-t, wing-rooms on either side of a one-story house sedan-chair
Car:1-t), brand=(Great Wall:1-t, Ford:1-t), discharge capacity=(1.1-1.6L:2-t).
S5, using multiple attribute utility theory, with reference to the preference heterogeneity of user, individual character is carried out to all actual resources on backstage
Change interest level to calculate, generate the actual resource sorted lists of user personalized interest.Specially：
If the n given attribute in vertical search engine, the personalization preferences the effective equation of user can be imitated with multiplication
Described with function：Uj=k1+k2+……+kn+k1k2+k2k3+k1k3+……+kn-1kn.Wherein UjRepresent that j-th of entity is being worked as
The interest level of preceding user, kiRepresent preference heterogeneity of the user to the ith attribute value of entity.Formulae express belongs to for each entity
Property value preference heterogeneity independently calculate after results added, and each entity property value preference heterogeneity will belong to other each entities
Property value preference heterogeneity carry out product after be added again.
Calculate preference heterogeneity kiDifferent to the interest-degree of entity based on user, the present invention is using Average Frequency/
Inverse Document Frequency (AF-IDF) algorithms characterize ith attribute value fancy grade of the user to entity,
ki=AF (X/T) * loga(N/DF(ki)), wherein AF (X/T) is concern of the user to the ith attribute value of entity within the T cycles
Average time (inquiry or number of visits), X is the property value total degree of concern, and n is entity in the vertical search system
Sum, DF (ki) it is that attribute value is kiEntity number, is calculated after each attribute weight successively, is obtained user and is based on attributive classification
Interest preference vector.
The interest of user is often to change, and interest forgetting factor f will be introduced in preference heterogeneityi, i.e.,：k′i=ki*fi,
fi=exp (- logb(t)/f), wherein t is the time to current time difference that property value last time is concerned, and f is half-life period.
Personalized interest angle value is calculated successively to each entity in vertical search system, until last entity is calculated
Finish.
Sorted from high to low according to personalized interest angle value, and export ranking results, be stored in individual subscriber interest mould
Under type.Sort result be according to algorithm in itself, calculate the score of each actual resource, and be ranked up.By taking scene as an example,
Backstage is the score that each user calculates all actual resources., can be with when user, which opens site home page, does not initiate inquiry
Entity is included recommending column according to the descending arrangement of score.If user initiates according to keywords or attribute selection is inquired about,
Candidate's actual resource is first obtained, is then resequenced according to the score of actual resource, is shown to search result column.
In order to preferably describe algorithm, make following simplify and assume：
The interest model of user 1 is precisely that S4 steps update, and the vector of 4 attributes is：
Price=(8-10 ten thousand:2-t),
Vehicle=(SUV:1-t, two box:1-t),
Brand=(Great Wall:1-t, Ford:1-t),
Discharge capacity=(1.1-1.6L:2-t).
Certain entity U1The exactly similar vehicle of S4 steps concern, its attribute is：Price：8-10 ten thousand；Vehicle：Wing-rooms on either side of a one-story house sedan-chair
Car；Brand：Ford；Discharge capacity：1.1-1.6L.
Vertical Website automobile entity sum n=1000.
Price is 500 in 8-10 ten thousand quantity,
Vehicle is that the quantity of two box is 250,
Brand is that the quantity of Ford is 25,
The quantity that discharge capacity is 1.1-1.6L is 250,
Cycle T is temporarily set to 1.
Due to concern time t with inquiry on the same day, no time difference, then forgetting factor is 1.
Calculate U1=k1+k2+k3+k4+k1k2+k1k3+k1k4+k2k3+k2k4+k3k4Value, show user to entity U1It is emerging
Interesting degree.The value that a is taken in this example is 2.
k1=2*log2(1000/500)=2
k2=2*log2(1000/250)=4
k3=2*log2(1000/25)=10.64
k4=2*log2(1000/250)=4
U1=2+4+10.64+4+2*4+2*10.64+2*4+4*10.64+4*4+10.64*4=159.04
Entity U1Weights in the personalized model of user 1 are 159.04.
Illustrate weights of the entity in user individual model, all data above by a simple embodiment
Setting be artificial it is assumed that needing to be defined for the data in specific vertical entity website when implementing.Cycle T, half-life period f
Need to obtain more efficiently setting by testing etc. parameter.
If navigation patterns occur for S6, user, the browsing time can be set here as threshold value, some resource is browsed and exceedes
During threshold value, then show that user is paid close attention to the attribute classification of this actual resource, embody interest preference, be then updated according to S4 steps
Individual subscriber interest model.
The technological thought of above example only to illustrate the invention, it is impossible to which protection scope of the present invention is limited with this, it is every
According to technological thought proposed by the present invention, any change done on the basis of technical scheme each falls within the scope of the present invention
Within.
Claims (5)
1. a kind of method for vertical search based on user profile learning, it is characterised in that comprise the following steps：
Step 1, in the vertical search engine website comprising attribute selection Advanced Search, according to initialization when and user friendship
Mutually, the interest preference of user is determined；
Step 2, according to the interest preference of user, initialising subscriber interest model is set up, initialising subscriber interest model is expressed as：
xi=(attribute value：The time that the number of times that attribute value is paid close attention to by user-attribute value the last time is paid close attention to by user), xiTable
Show ith attribute；
Step 3, if user is inquired about according to attribute, the attribute value selected according to user is carried out to user interest model
Real-time update；If user is inquired about according to keyword, user interest model is not updated；
Step 4, if navigation patterns occur for user, the browsing time is set as threshold value, if the time for browsing certain actual resource exceedes
Threshold value, then obtain the property value of the actual resource, and carry out real-time update to user interest model；
Step 5, using multiple attribute utility theory, with reference to the preference heterogeneity and forgetting factor of user, all actual resources are carried out
Interest level is calculated and is ranked up from high to low by interest level；
Step 6, according to user select according to keywords or attribute carries out screening inquiry, generate initial Query Result list；
Step 7, for the actual resource in initial Query Result list, with reference to interest-degree of the user to these actual resources
Value, the actual resource in initial Query Result list is ranked up from high to low by interest level, generates final inquiry
The results list.
2. the method for vertical search based on user profile learning according to claim 1, it is characterised in that have described in step 1
Body process is：When user first enters into the vertical search engine website comprising attribute selection Advanced Search, for each attribute
Interacted with user, obtain interest preference of the user on each attribute.
3. the method for vertical search based on user profile learning according to claim 1, it is characterised in that if described in step 3
User is inquired about according to attribute, then the attribute value selected according to user, and it is specific to carry out real-time update to user interest model
Process is：If the attribute value of user's selection is in user interest model, time that the attribute value is paid close attention to by user
Number Jia 1, and updates the attribute value the last time paid close attention to by user, otherwise, by the attribute value, the attribute value quilt
The time that the number of times of user's concern and attribute value the last time are paid close attention to by user is increased in user interest model.
4. the method for vertical search based on user profile learning according to claim 1, it is characterised in that the step 5
Detailed process is：
Step 51, preference heterogeneity is calculated using AF-IDF algorithms, calculation formula is：
ki=AF (X/T) * loga(N/DF(ki))
Wherein, kiPreference heterogeneity of the user to j-th of actual resource ith attribute is represented, AF (X/T) is that user is right in cycle T
The average time of j-th of actual resource ith attribute concern, X is pays close attention to the total degree of the attribute, and N is vertical search engine net
The sum of actual resource in standing, DF (ki) for sum with j-th of actual resource ith attribute value identical actual resource, a
For hyper parameter；
Step 52, forgetting factor f is introduced in preference heterogeneityi, then the preference heterogeneity k introduced after forgetting factori' be：
k′i=ki*fi
Step 53, set attribute that vertical search engine website includes as n, i.e. i=1,2 ..., n, then user is real to j-th
The interest level of body resource is represented with multiplication utility function：
Uj=k '1+k′2+…+k′n+k′1k′2+k′1k′3+…+k′1k′n+k′2k′3+k′2k′4+…+k′2k′n+
k′3k′4+k′3k′5+…+k′3k′n+……+k′n-1k′n
Wherein, UjRepresent interest level of the user to j-th of actual resource；
Step 54, the interest level of all actual resources is ranked up from high to low, and exports ranking results.
5. the method for vertical search based on user profile learning according to claim 4, it is characterised in that the forgetting factor
fiCalculation formula is：
fi=exp (- logb(t)/f)
Wherein, t is concerned for the value last time of j-th of actual resource ith attribute time and the difference of current time, f
For half-life period, b is hyper parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710263913.6A CN107180078A (en) | 2017-04-21 | 2017-04-21 | A kind of method for vertical search based on user profile learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710263913.6A CN107180078A (en) | 2017-04-21 | 2017-04-21 | A kind of method for vertical search based on user profile learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107180078A true CN107180078A (en) | 2017-09-19 |
Family
ID=59832063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710263913.6A Pending CN107180078A (en) | 2017-04-21 | 2017-04-21 | A kind of method for vertical search based on user profile learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107180078A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862004A (en) * | 2017-10-24 | 2018-03-30 | 科大讯飞股份有限公司 | Intelligent sorting method and device, storage medium and electronic equipment |
| CN108537596A (en) * | 2018-04-16 | 2018-09-14 | 北京车慧互动广告有限公司 | Method, apparatus, system and the memory that vehicle is recommended in search box |
| CN108920649A (en) * | 2018-06-29 | 2018-11-30 | 北京百度网讯科技有限公司 | A kind of information recommendation method, device, equipment and medium |
| CN109376297A (en) * | 2018-08-31 | 2019-02-22 | 中国联合网络通信集团有限公司 | User characteristic data treating method and apparatus |
| CN109559208A (en) * | 2019-01-04 | 2019-04-02 | 平安科技（深圳）有限公司 | A kind of information recommendation method, server and computer-readable medium |
| CN111625726A (en) * | 2020-06-02 | 2020-09-04 | 小红书科技有限公司 | User portrait processing method and device |
| CN113159184A (en) * | 2021-04-23 | 2021-07-23 | 武汉陌青春创意科技有限公司 | User interest feature identification method and device |
Citations (3)
| 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 |
| CN102081604A (en) * | 2009-11-27 | 2011-06-01 | 上海电机学院 | Search method for meta search engine and device thereof |
| CN103984721A (en) * | 2014-05-13 | 2014-08-13 | 中国矿业大学 | Personalized book searching method based on interactive evolutionary optimization |
-
2017
- 2017-04-21 CN CN201710263913.6A patent/CN107180078A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102081604A (en) * | 2009-11-27 | 2011-06-01 | 上海电机学院 | Search method for meta search engine and device thereof |
| CN101770520A (en) * | 2010-03-05 | 2010-07-07 | 南京邮电大学 | User interest modeling method based on user browsing behavior |
| CN103984721A (en) * | 2014-05-13 | 2014-08-13 | 中国矿业大学 | Personalized book searching method based on interactive evolutionary optimization |
Non-Patent Citations (2)
| Title |
|---|
| 勾智楠: ""垂直搜索引擎的研究与实现"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 勾智楠: ""基于用户兴趣的多属性效用算法的研究"", 《河北工业科技》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862004A (en) * | 2017-10-24 | 2018-03-30 | 科大讯飞股份有限公司 | Intelligent sorting method and device, storage medium and electronic equipment |
| CN108537596A (en) * | 2018-04-16 | 2018-09-14 | 北京车慧互动广告有限公司 | Method, apparatus, system and the memory that vehicle is recommended in search box |
| CN108920649A (en) * | 2018-06-29 | 2018-11-30 | 北京百度网讯科技有限公司 | A kind of information recommendation method, device, equipment and medium |
| CN108920649B (en) * | 2018-06-29 | 2021-09-17 | 北京百度网讯科技有限公司 | Information recommendation method, device, equipment and medium |
| CN109376297A (en) * | 2018-08-31 | 2019-02-22 | 中国联合网络通信集团有限公司 | User characteristic data treating method and apparatus |
| CN109559208A (en) * | 2019-01-04 | 2019-04-02 | 平安科技（深圳）有限公司 | A kind of information recommendation method, server and computer-readable medium |
| CN111625726A (en) * | 2020-06-02 | 2020-09-04 | 小红书科技有限公司 | User portrait processing method and device |
| CN113159184A (en) * | 2021-04-23 | 2021-07-23 | 武汉陌青春创意科技有限公司 | User interest feature identification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107180078A (en) | A kind of method for vertical search based on user profile learning | |
| Kumar et al. | A movie recommender system: Movrec | |
| CN102982042B (en) | A kind of personalization content recommendation method, platform and system | |
| Zhou et al. | Userrec: A user recommendation framework in social tagging systems | |
| Chen et al. | Collabseer: a search engine for collaboration discovery | |
| CN108647276B (en) | Searching method | |
| KR101565339B1 (en) | Recommendation system using collective intelligence and method thereof | |
| JP5386663B1 (en) | Information processing apparatus, information processing method, information processing program, and recording medium | |
| CN110020128B (en) | Search result ordering method and device | |
| CN101408897B (en) | Personalized enquiry expanding method based on collaboration filtering | |
| CN107944035B (en) | Image recommendation method integrating visual features and user scores | |
| CN108363804A (en) | Local model weighted fusion Top-N movie recommendation method based on user clustering | |
| CN105095433A (en) | Recommendation method and device for entities | |
| CN101140588A (en) | Method and apparatus for ordering incidence relation search result | |
| CN103646092A (en) | SE (search engine) ordering method based on user participation | |
| CN102968465A (en) | Network information service platform and search service method based on network information service platform | |
| CN104462383A (en) | Movie recommendation method based on feedback of users' various behaviors | |
| CN103294692A (en) | Information recommendation method and system | |
| CN113806630A (en) | Attention-based multi-view feature fusion cross-domain recommendation method and device | |
| CN106156372A (en) | The sorting technique of a kind of internet site and device | |
| CN106960044A (en) | A kind of Time Perception personalization POI based on tensor resolution and Weighted H ITS recommends method | |
| CN109992674A (en) | Recommendation method fusing automatic encoder and knowledge graph semantic information | |
| Saranya et al. | A personalized online news recommendation system | |
| CN105894310A (en) | Personalized recommendation method | |
| CN102214227B (en) | Automatic public opinion monitoring method based on internet hierarchical structure storage |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170919 |
|
| RJ01 | Rejection of invention patent application after publication |