CN107229666B - A kind of interest heuristic approach and device based on recommender system - Google Patents
A kind of interest heuristic approach and device based on recommender system Download PDFInfo
- Publication number
- CN107229666B CN107229666B CN201611253179.7A CN201611253179A CN107229666B CN 107229666 B CN107229666 B CN 107229666B CN 201611253179 A CN201611253179 A CN 201611253179A CN 107229666 B CN107229666 B CN 107229666B
- Authority
- CN
- China
- Prior art keywords
- term
- interest
- user
- word
- operation behavior
- 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
-
- 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 interest heuristic approach and device based on recommender system, comprising: based on the interest word for recommending page presentation to construct in advance；The corresponding related data information of the interest word is obtained according to the operation behavior to the interest word；Recommended models are updated according to the operation behavior to the related data information, and provide recommending data by the updated recommended models；So that for obtaining recommendation using function is explored independent of active.
Description
Technical field
The present invention relates to field of intelligent control, and in particular to a kind of interest heuristic approach and device based on recommender system.
Background technique
In recent years, with the explosive development of mobile Internet and big data technology, intelligent recommendation technology is more and more
Internet product on be used widely, for example, news recommend (typical products such as today's tops), film books recommend (beans
Valve), video recommendations (youku.com, iqiyi.com etc.), commercial product recommending (Taobao, pocket shopping).The main hardly possible that recommended technology faces
Point is how comprehensively to excavate the interest and demand of user as far as possible.Specifically, a user is given, is dug based on some data
Pick means (for example excavate user and click history, Web browsing history, social network information), our available certain customers are emerging
Interest, thus solve recommend cold start-up (so-called cold start-up refers to: how in the case where no a large number of users data design individuality
Change recommender system and make user satisfied to recommendation results to be ready exactly to be cold-started problem using recommender system) problem.But
It is to meet known user interest simply recommended range is easy to cause to narrow and user satisfaction decline.
The recommender system of mainstream generally uses following several ways at present:
1, the content of some interest extensions is inserted into the content recommended: such as, it is assumed that recommended models know user for EXO
It is interested with deer break, the news of other South Korea stars can be inserted into when recommending related news.If user also clicks these
The label of " South Korea star " will be added the interest model of user in content, system.
2, interest extension tag being stamped to the content of recommendation: continuing to use above example, system can recommend at each
EXO related article behind stamp the label of " South Korea star ", user can click the content checked under the label and confirmation " happiness
Vigorously ".After obtaining user's confirmation, " South Korea star " label will be added the interest model of user in system.
3, provide an entrance for interest exploration: system provides an entrance to user, facilitates user to existing interior
Exploration interest when appearance is fed up with.Above example is continued to use, when user wants to look at that there are also can when what interesting content
To enter the interest tags that " exploration " portal view is recommended, if to recommending " South Korea star " label out interested, so that it may
Confirmation is clicked, at this moment " South Korea star " label will be added the interest model of user in system.
Above-mentioned several common recommended methods are had the drawback that
For method 1) for, because being the content for being inserted directly into interest exploration in recommendation, if interest is explored
Content selection it is bad, will lead to explore efficiency it is relatively low, cause influence user experience.
For method 2) for, although small to user experience injury, that explores is limited in scope.Such as EXO's
Article, system can stamp the label of " South Korea star ", but if user crosses " liking " tag expression, bigger model
The interest exploration enclosed will will receive limitation.Such as if system is wanted to confirm whether user likes " Japanese star ", you can't get
It supports.
For method 3) for, relative to method 1) and method 2) compare most flexibly, it both can directly recommend " interest mark
Label " improve the efficiency that interest is explored；It can recommend not occur in the article that user has seen out without being bound by the limitation of article again
The interest tags crossed.But method 3) needing user, actively function is explored in use, and most of user for recommender system
Recommendation, therefore, method 3 are obtained in the case of passively) in exploration function utilization rate it is lower.
How to provide one kind can be independent of user actively using exploration function, and avoids directly using recommendation as grain
The user interest heuristic approach explored on a large scale and user is caused to dislike spent becomes the skill that those skilled in the art's needs solve
Art problem.
Summary of the invention
The present invention provides a kind of interest heuristic approach and device based on recommender system, to solve to depend in the prior art
User actively uses the problem of exploring function and poor user experience.
The present invention provides a kind of interest heuristic approach based on recommender system, comprising:
Based on the interest word for recommending page presentation to construct in advance；
The corresponding related data information of the interest word is obtained according to the operation behavior to the interest word；
Recommended models are updated according to the operation behavior to the related data information, and pass through the updated recommendation mould
Type provides recommending data.
Preferably, the interest word constructed in advance includes:
The highest term of search rate under each classification is searched in the search log of the recommender system, forms retrieval
Set of words；
The term in the retrieval set of words is screened for each user；
It is stored in the term after screening as interest word in the interest word data of the user.
Preferably, the retrieval set of words includes: the term and the corresponding tag along sort of the term.
Preferably, the term in the retrieval set of words, which screen, includes:
By it is described retrieval set of words in the term corresponding to tag along sort respectively with the interest mark of each user
Label are matched, if matching, chooses the corresponding term of the tag along sort, and enter it is described will be described in after screening
Term is stored in the step in the interest word data of the user as interest word, if mismatching, deletes the term.
Preferably, the term using after screening is stored in the interest word data of the user as interest word
Include:
The term after the screening is ranked up.
Preferably, the term to after the screening is ranked up, comprising:
Calculate the retrieval score value that the term corresponds to each user；
Descending arrangement is carried out according to the term of the retrieval score value to each user.
Preferably, the retrieval score value that the calculating term corresponds to each user is obtained using following formula:
Score (u, v)=w1 × hot (v)+w2 × relevance (u, v)+w3 × fresh (u, v)
Wherein, the u indicates user；The v indicates term；W1, w2 and w3 representative function weight；Hot (v) indicates meter
Calculate the function of v temperature；The degree of correlation of relevance (u, v) expression u and v；Fresh (u, v) indicates v for the freshness of u.
Preferably, the term to after the screening, which is ranked up, includes:
Each user is recorded to the number of the operation behavior of the term；
The number is matched with the data in the order models pre-established, obtains the term operation behavior
Probability value；
Descending arrangement is carried out to the term according to the probability value.
Preferably, the term to after the screening is ranked up, using the term after sequence as
The interest word is stored in user interest word database, comprising:
The term after the sequence is filtered.
Preferably, described to include: based on the interest word for recommending page presentation to construct in advance
The form of interest word composition interest set of words is shown in the recommendation column for recommending the page.
Preferably, described to include: based on the interest word for recommending page presentation to construct in advance
According to preset carousel rule, the interest word is individually taken turns in the search box for recommending the page
It broadcasts.
Preferably, the basis updates recommended models to the operation behavior of the related data information, and by updating after
The recommended models provide recommending data include:
Corresponding sample data is generated according to the operation behavior to the related data information, and the sample data is added
Enter into the training data of recommended models；
The related data information that the recommended models are generated according to the operation behavior, adds in the recommended models
Add the feature weight of the operation behavior.
The present invention also provides a kind of interest exploration device based on recommender system, comprising:
Display unit, for based on the interest word for recommending page presentation to construct in advance；
Acquiring unit, for according to obtaining the corresponding dependency number of the interest word to the operation behavior of the interest word it is believed that
Breath；
Updating unit for updating recommended models according to the operation behavior to the related data information, and passes through update
The recommended models afterwards provide recommending data.
Preferably, the display unit includes: construction unit, and the construction unit includes:
Searching unit, for searching the highest inspection of search rate under each classification in the search log of the recommender system
Rope word forms retrieval set of words；
Screening unit, for being screened for each user to the term in the retrieval set of words；
Storage unit, for being stored in the interest word data of the user using the term after screening as interest word
In.
Preferably, the retrieval set of words includes: the term and the corresponding tag along sort of the term.
Preferably, the screening unit includes:
Matching unit, for by it is described retrieval set of words in the term corresponding to tag along sort respectively with it is each
The interest tags of user match, if matching, choose the corresponding term of the tag along sort, and the entrance general
The term after screening is stored in the step in the interest word data of the user as interest word, if mismatching, deletes
Except the term.
Preferably, the screening unit includes:
Sequencing unit, for being ranked up to the term after the screening.
Preferably, the sequencing unit includes:
Computing unit corresponds to the retrieval score value of each user for calculating the term；
Arrangement units, for carrying out descending arrangement according to the term of the retrieval score value to each user.
Preferably, the sequencing unit includes:
Recording unit, for recording each user to the number of the operation behavior of the term；
Matching unit, for matching the number with the data in the order models pre-established, described in acquisition
The probability value of term operation behavior；
Arrangement units, for carrying out descending arrangement to the term according to the probability value.
Preferably, the sequencing unit includes: filter element, for being filtered to the term after sequence.
Preferably, the updating unit includes:
Sample data generation unit, for generating corresponding sample number according to the operation behavior to the related data information
According to, and the sample data is added into the training data of recommended models；
Feature weight adding unit, the dependency number generated for the recommended models according to the operation behavior it is believed that
Breath, adds the feature weight of the operation behavior in the recommended models.
A kind of interest heuristic approach based on recommender system provided by the invention, by the recommendation page of recommender system
The interest word constructed in advance is directly shown, so that for obtaining recommendation using function is explored independent of active.Separately
Outside, the present invention is to obtain search result based on to the operation behavior for recommending interest word on the page, and according to search result
It operates click behavior and updates recommended models, so that subsequent recommendation efficiency is more.
Detailed description of the invention
It, below will be to embodiment in order to illustrate more clearly of the utility model embodiment and technical solution in the prior art
It is briefly described with attached drawing needed to be used in the description of the prior art, it should be apparent that, the accompanying drawings in the following description is only
It is some embodiments of the utility model, for those of ordinary skill in the art, in the premise not made the creative labor
Under, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of flow chart of interest heuristic approach based on recommender system provided by the invention；
Fig. 2 is that interest word exhibition method one shows in a kind of interest heuristic approach based on recommender system provided by the invention
It is intended to；
Fig. 3 is that interest word exhibition method two shows in a kind of interest heuristic approach based on recommender system provided by the invention
It is intended to；
Fig. 4 is a kind of structural schematic diagram of interest exploration device based on recommender system provided by the invention.
Specific embodiment
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention.But the present invention can be with
Much it is different from other way described herein to implement, those skilled in the art can be without prejudice to intension of the present invention the case where
Under do similar popularization, therefore the present invention is not limited to the specific embodiments disclosed below.
It please refers to shown in Fig. 1, Fig. 1 is a kind of process of interest heuristic approach based on recommender system provided by the invention
Figure, this method comprises:
Step S101: based on the interest word for recommending page presentation to construct in advance.
It is described to show the interest word constructed in advance in this embodiment on the recommendation page of recommender system and have two
Kind mode:
One is, be shown in the form of interest set of words it is described recommend the page recommendation column in (as shown in Figure 2), when with
Family brushes out interest word card when refreshing and obtaining recommendation at random, in a period of time after displaying those interest words will not again by
Recommend, that is, those interest are not being shown in the recommendation column for recommending the page in a period of time after displaying.
It is for second, according to preset carousel rule, by the interest word described in the form of single interest word
Recommend independent carousel (as shown in Figure 3) in the search box of the page.
The preset carousel rule, can be the carousel time for setting the interest word and/or the interest word
The sequence of carousel, the sequence of the carousel, which can be, successively carries out carousel according to the collating sequence of the interest word constructed in advance.Wheel
It can be variation when user refreshes the new recommendation of acquisition every time between sowing time once, all to update after one day, in a period of time not
Can recommend again etc..
Certain exhibition method may also is that the interest word that will be constructed in advance on the recommendation page with interest word set
While the form of conjunction (may also be referred to as are as follows: interest word card) is shown in the recommendation column for recommending the page, pushed away described
It recommends in the search box of the page according to preset carousel rule, the interest word is subjected to carousel in described search frame.
In this embodiment, following form can be used for the interest word detailed process constructed in advance:
The highest term of search rate under each classification is searched in the search log of the recommender system, forms retrieval
Set of words；That is, in the recommender system, searches all users and search for search rate in logs under each classification most
High term.
Such as: if searching search journal displaying user under [world] and [finance and economics] two classification, search rate highest
Term be [India changes paper money], the classification of the term [India changes paper money] is exactly [world] [finance and economics], can also be had certainly
Multiple classification, such as: it searches and shows user in search log under [amusement] and [sport] two classification, search rate is highest
Term is [sports star makes a film], and the retrieval set of words can include: that term [India changes paper money] is corresponding [world]
[finance and economics] tag along sort；Term [sports star makes a film] corresponding [amusement] and [sport] tag along sort, above are only
It is bright and illustrate, it is not intended to limit the present invention.
The term is searched according to search rate, is because of the high term pair of search rate under normal conditions
What is answered is all the popular information of comparison, and then the related content of the corresponding term also can be relatively more, to guarantee user interest
The efficiency of exploration.
In this embodiment, the retrieval set of words includes: the term and the corresponding tag along sort of the term.
The term in the retrieval set of words is screened for each user；That is, to each use
Family filters out its corresponding term, and the term of each user can have multiple.The process specifically screened can be, will
The corresponding tag along sort of the term in the retrieval set of words is matched with the interest tags of each user respectively, if
Matching, then retain the term and the term using after screening be stored in the interest of the user as interest word
Step in word data deletes the term if mismatching.
Such as: for a certain user u1, the interest tags with [world] and [finance and economics] will be in the retrieval set of words
Tag along sort be compared with its interest tags, then will be in the tag along sort when interest tags are identical as tag along sort
The corresponding term [India changes paper money] is chosen to be stored as the term after screening, however tag along sort is [amusement]
[sport] is different from the interest tags [world] of the user u1 and [finance and economics], therefore the tag along sort [amusement] and [body
Educate] the corresponding term [sports star makes a film] will not then be selected.
Based on above content, the interest word data of the user are stored in using the term after screening as interest word
In library.
It, can also be to described in after the screening in order to further increase each user to the level of interest of the term
Term is ranked up.The process specifically to sort can carry out in the following way:
Calculate the retrieval score value that the term corresponds to each user；
Descending arrangement is carried out according to the term of the retrieval score value to each user.
The retrieval score value can be obtained by following equation:
Score (u, v)=w1 × hot (v)+w2 × relevance (u, v)+w3 × fresh (u, v)
Wherein, the u indicates user；The v indicates term；W1, w2 and w3 representative function weight；Hot (v) indicates meter
Calculate the function of v temperature；The degree of correlation of relevance (u, v) expression u and v；Fresh (u, v) indicates v for the freshness of u.
The temperature of the term v can be the searching times of the user u whithin a period of time to indicate；
The degree of correlation of the user u and the term v, which can be, is sorted in the user u's with the term v
Relative weighting in interest tags indicates；
The term v can be the freshness of the user u have been seen with user u described in nearest a period of time
Content quantity in article title not comprising the term v indicates that fresh (u, the v) value is higher, illustrates the user u
Freshness is stronger to be felt for the term v.
The mode being ranked up for the term after the screening can also be by real with the method for machine learning
It is existing, specifically it may is that
Each user is recorded to the number of the operation behavior of the term；The operation behavior can be click behavior.
The number is matched with the data in the order models pre-established, obtains the term operation behavior
Probability value；
Descending arrangement is carried out to the term according to the probability value.
To prevent certain interest words from repeating to show user, after completing sequence to the term after screening, for same
For one user, its most interested interest word may be fixed situation in a period of time, according to possible interested general
Rate is recommended, and will lead to repetition recommendation, therefore, in this embodiment can also be as needed showing in a period of time to user
Interest word filters out, and user is caused to dislike；The time of filtering can be set to one day.
Step S102: the corresponding related data information of the interest word is obtained according to the operation behavior to the interest word.
The specific implementation process of the step S102 is the operation that user carries out the interest word shown on the recommendation page
Behavior can be the operation behavior for clicking interest word, enters after user clicks the interest word and recommends the next of the page
The grade page, it may be assumed that the search results pages with the interest word related content that display generates the interest word clicking operation behavior
Face, described search results page can be obtained by search engine.
On the basis of the result of page searching of related content, user can the result of page searching to related content carry out a little
Hit operate so that enter want obtain particular content information in viewing, it may be assumed that obtain the corresponding dependency number of the interest word it is believed that
Breath.
Such as: user clicks recommender system interest word [India changes paper money] for recommending on recommending the page, and mutually inside the Pass
It is shown on the original list of appearance such as:
" someone can't buy what experience of paper money is changed in dish someone suicide suddenly? Indian tells you！"；
" print matchmaker: " it is dead to change paper money " is up to 48 "；
" India's " lightning changes paper money " cause confusion grow thickly market confusion cause social unrest "；
……
……
" India's " changing paper money greatly " swap out business opportunity ".
For based on the related content on the above-mentioned result of page searching provided, " someone to can't buy dish someone to user's click
What experience is of paper money changed in suicide suddenly? Indian tells you！" and " print matchmaker: " it is dead to change paper money " is up to 48 ", corresponding two news exhibitions
Relevant specific contents information is shown.
Step S103: recommended models are updated according to the operation behavior to the related data information, and by updated
The recommended models provide recommending data.
The specific implementation process of the step S103 may is that
Corresponding sample data is generated according to the operation behavior to the related data information, and the sample data is added
Enter into the training data of recommended models.
The related data information that the recommended models generate the operation behavior as sample data, such as: use
Family can be user to the clicking operation of this article, the clicking operation to the operation behavior of given article, the operation behavior
Generated related data information；
In this implementation, the feature weight of the operation behavior is added in the recommended models.The feature weight can be with
Be: { user is to the interested degree of [world] news }, { user is to [India] interested degree }, { user should to [changing paper money]
The degree of interest }, the other articles for meeting features described above weight also more likely are recommended by model in subsequent recommendation
Come, the article may be recommended out by model can be using the method for recommended models numerous in the prior art, therefore, herein
It repeats no more.
A kind of interest heuristic approach based on recommender system provided by the invention, by the recommendation page of recommender system
The interest word constructed in advance is directly shown, so that recommender system, which does not depend on user, actively obtains user more using exploration function
More interest expression.In addition, the present invention is to obtain search result, and root based on the operation behavior to interest word on the recommendation page
Behavior is clicked according to the operation to search result and updates recommended models, so that subsequent recommendation is more efficient, because user is to certain
A keyword it is interested but may not each content to this keyword it is all interested.
It disclosed above a kind of specific implementation process of the interest heuristic approach based on recommender system of the present invention, with the side
Method embodiment is corresponding, the interest exploration device embodiment based on propulsion system that the present invention also provides a kind of.
Fig. 4 is please referred to, for a kind of structural schematic diagram of the interest exploration device based on recommender system of the present invention.The device
Specific implementation process it is substantially similar to the specific implementation process of the method, so the description to described device is fairly simple,
Particularly relevant place can illustrate with reference to the specific descriptions of method part.Installation practice described below is only to illustrate
Property.
As shown in figure 4, the present invention provides a kind of interest exploration device based on recommender system, comprising:
Display unit 401, for based on the interest word for recommending page presentation to construct in advance；
The display unit 401 includes: construction unit, and the construction unit includes:
Searching unit forms retrieval for searching the highest term of search rate in search log under various classification
Set of words；
Screening unit, for being screened for each user to the term in the retrieval set of words；
The screening unit includes:
Matching unit, for by it is described retrieval set of words in the term corresponding to tag along sort respectively with it is each
The interest tags of user match, if matching, choose the corresponding term of the tag along sort, and the entrance general
The term after screening is stored in the step in the interest word data of the user as interest word, if mismatching, deletes
Except the term.
The screening unit further include:
Sequencing unit, for being ranked up to the term after the screening.
The sequencing unit includes:
Computing unit corresponds to the retrieval score value of each user for calculating the term；
Arrangement units, for carrying out descending arrangement according to the term of the retrieval score value to each user.
Storage unit, for being stored in the interest word data of the user using the term after screening as interest word
In.
Or the sequencing unit includes:
Recording unit, for recording each user to the number of the operation behavior of the term；
Matching unit, for matching the number with the data in the order models pre-established, described in acquisition
The probability value of term operation behavior；
Arrangement units, for carrying out descending arrangement to the term according to the probability value.
The sequencing unit can also include:
Filter element, for being filtered to the term after sequence.
The retrieval set of words includes: the term and the corresponding tag along sort of the term.
Acquiring unit 402, for obtaining the corresponding dependency number of the interest word according to the operation behavior to the interest word
It is believed that breath.
Updating unit 403, for updating recommended models according to the operation behavior to the related data information, and by more
The recommended models after new provide recommending data.
The updating unit includes: sample data generation unit and feature weight adding unit.
The sample data generation unit, for generating corresponding sample according to the operation behavior to the related data information
Notebook data, and the sample data is added into the training data of recommended models；
The feature weight adding unit, the dependency number generated for the recommended models according to the operation behavior
It is believed that breath, adds the feature weight of the operation behavior in the recommended models.
Although the present invention is disclosed as above with preferred embodiment, it is not for limiting the present invention, any this field skill
Art personnel without departing from the spirit and scope of the present invention, can make possible variation and modification, therefore guarantor of the invention
Shield range should be subject to the range that the claims in the present invention are defined.
Claims (14)
1. a kind of interest heuristic approach based on recommender system characterized by comprising
Based on the interest word for recommending page presentation to construct in advance, the interest word refers to the highest inspection of search rate under each classification
Rope word；
The corresponding related data information of the interest word is obtained according to the operation behavior to the interest word；
Recommended models are updated according to the operation behavior to the related data information, and are mentioned by the updated recommended models
For recommending data；
The highest term of search rate under each classification is searched in the search log of the recommender system, forms retrieval word set
It closes；
The term in the retrieval set of words is screened for each user；
It is stored in the term after screening as interest word in the interest word data of the user；
The term using after screening is stored in the interest word data of the user as interest word
The term after the screening is ranked up；
The term to after the screening is ranked up, comprising:
Calculate the retrieval score value that the term corresponds to each user；
Descending arrangement is carried out according to the term of the retrieval score value to each user；
The retrieval score value that the calculating term corresponds to each user is obtained using following formula:
Score (u, v)=w1 × hot (v)+w2 × relevance (u, v)+w3 × fresh (u, v)
Wherein, the u indicates user；The v indicates term；W1, w2 and w3 representative function weight；Hot (v) indicates to calculate v
The function of temperature；The degree of correlation of relevance (u, v) expression u and v；Fresh (u, v) indicates v for the freshness of u.
2. the interest heuristic approach according to claim 1 based on recommender system, which is characterized in that the retrieval set of words
It include: the term and the corresponding tag along sort of the term.
3. the interest heuristic approach according to claim 2 based on recommender system, which is characterized in that described to the retrieval
The term in set of words carries out screening
By it is described retrieval set of words in the term corresponding to tag along sort respectively with the interest tags of each user into
Row matching chooses the corresponding term of the tag along sort, and enter the retrieval by after screening if matching
Word is stored in the step in the interest word data of the user as interest word, if mismatching, deletes the term.
4. the interest heuristic approach according to claim 1 based on recommender system, which is characterized in that described to the screening
The term afterwards, which is ranked up, includes:
Each user is recorded to the number of the operation behavior of the term；
The number is matched with the data in the order models pre-established, obtains the general of the term operation behavior
Rate value；
Descending arrangement is carried out to the term according to the probability value.
5. the interest heuristic approach according to claim 1 based on recommender system, which is characterized in that described to the screening
The term afterwards is ranked up, and is stored in user interest word data for the term after sequence as the interest word
In library, comprising:
The term after the sequence is filtered.
6. the interest heuristic approach according to claim 1 based on recommender system, which is characterized in that described based on recommendation page
Face shows that the interest word constructed in advance includes:
The form of interest word composition interest set of words is shown in the recommendation column for recommending the page.
7. the interest heuristic approach according to claim 1 based on recommender system, which is characterized in that described based on recommendation page
Face shows that the interest word constructed in advance includes:
According to preset carousel rule, the interest word is subjected to independent carousel in the search box for recommending the page.
8. the interest heuristic approach according to claim 1 based on recommender system, which is characterized in that the basis is to described
The operation behavior of related data information updates recommended models, and provides recommending data packet by the updated recommended models
It includes:
Generate corresponding sample data according to the operation behavior to the related data information, and by the sample data be added to
In the training data of recommended models；
The related data information that the recommended models are generated according to the operation behavior, adds institute in the recommended models
State the feature weight of operation behavior.
9. a kind of interest exploration device based on recommender system characterized by comprising
Display unit, for based on the interest word for recommending page presentation to construct in advance, the interest word to refer to searching under each classification
The highest term of rope frequency；
Acquiring unit, for obtaining the corresponding related data information of the interest word according to the operation behavior to the interest word；
Updating unit, for updating recommended models according to the operation behavior to the related data information, and by updated
The recommended models provide recommending data；
The display unit includes: construction unit, and the construction unit includes:
Searching unit, for searching the highest retrieval of search rate under each classification in the search log of the recommender system
Word forms retrieval set of words；
Screening unit, for being screened for each user to the term in the retrieval set of words；
Storage unit, for the term after screening to be stored in the interest word data of the user as interest word；
The screening unit includes:
Sequencing unit, for being ranked up to the term after the screening；
The sequencing unit includes:
Computing unit corresponds to the retrieval score value of each user for calculating the term；
Arrangement units, for carrying out descending arrangement according to the term of the retrieval score value to each user；
The retrieval score value that the calculating term corresponds to each user is obtained using following formula:
Score (u, v)=w1 × hot (v)+w2 × relevance (u, v)+w3 × fresh (u, v)
Wherein, the u indicates user；The v indicates term；W1, w2 and w3 representative function weight；Hot (v) indicates to calculate v
The function of temperature；The degree of correlation of relevance (u, v) expression u and v；Fresh (u, v) indicates v for the freshness of u.
10. the interest exploration device according to claim 9 based on recommender system, which is characterized in that the retrieval word set
Conjunction includes: the term and the corresponding tag along sort of the term.
11. the interest exploration device according to claim 10 based on recommender system, which is characterized in that the screening unit
Include:
Matching unit, for by it is described retrieval set of words in the term corresponding to tag along sort respectively with each user
Interest tags matched, if matching, chooses the corresponding term of the tag along sort, and enter it is described will screening
The term afterwards is stored in the step in the interest word data of the user as interest word, if mismatching, deletes institute
State term.
12. the interest exploration device according to claim 9 based on recommender system, which is characterized in that the sequencing unit
Include:
Recording unit, for recording each user to the number of the operation behavior of the term；
Matching unit obtains the retrieval for matching the number with the data in the order models pre-established
The probability value of word operation behavior；
Arrangement units, for carrying out descending arrangement to the term according to the probability value.
13. the interest exploration device according to claim 9 based on recommender system, which is characterized in that the sequencing unit
It include: filter element, for being filtered to the term after sequence.
14. the interest exploration device according to claim 9 based on recommender system, which is characterized in that the updating unit
Include:
Sample data generation unit, for generating corresponding sample data according to the operation behavior to the related data information,
And the sample data is added into the training data of recommended models；
Feature weight adding unit, for the related data information that the recommended models are generated according to the operation behavior,
The feature weight of the operation behavior is added in the recommended models.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611253179.7A CN107229666B (en) | 2016-12-30 | 2016-12-30 | A kind of interest heuristic approach and device based on recommender system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611253179.7A CN107229666B (en) | 2016-12-30 | 2016-12-30 | A kind of interest heuristic approach and device based on recommender system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107229666A CN107229666A (en) | 2017-10-03 |
| CN107229666B true CN107229666B (en) | 2019-02-26 |
Family
ID=59932858
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611253179.7A Active CN107229666B (en) | 2016-12-30 | 2016-12-30 | A kind of interest heuristic approach and device based on recommender system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107229666B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107886357A (en) * | 2017-11-06 | 2018-04-06 | 北京希格斯科技发展有限公司 | The method and system of content value is judged based on user behavior data |
| CN109189954B (en) * | 2018-09-07 | 2021-04-16 | 网易传媒科技（北京）有限公司 | Content recommendation method and device |
| CN110619076B (en) * | 2018-12-25 | 2023-04-07 | 北京时光荏苒科技有限公司 | Search term recommendation method and device, computer and storage medium |
| CN109688458A (en) * | 2019-01-14 | 2019-04-26 | 四川长虹电器股份有限公司 | The implementation method of smart television cloud desktop operation system based on big data algorithm |
| CN110287420B (en) * | 2019-06-28 | 2020-12-18 | 深圳前海微众银行股份有限公司 | Training method and device of information recommendation system |
| CN110598766B (en) * | 2019-08-28 | 2022-05-10 | 第四范式（北京）技术有限公司 | Training method and device for commodity recommendation model and electronic equipment |
| CN111680219B (en) * | 2020-06-09 | 2023-10-20 | 深圳市雅阅科技有限公司 | Content recommendation method, device, equipment and readable storage medium |
| CN113077326B (en) * | 2021-06-04 | 2021-10-08 | 中建电子商务有限责任公司 | Building commodity recommendation method, device and equipment based on boundary interest heuristic |
| CN114036403B (en) * | 2022-01-07 | 2022-03-25 | 智者四海（北京）技术有限公司 | User interest detection method, device and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102360358A (en) * | 2011-09-28 | 2012-02-22 | 百度在线网络技术（北京）有限公司 | Keyword recommendation method and system |
| CN103761263A (en) * | 2013-12-31 | 2014-04-30 | 武汉传神信息技术有限公司 | Method for recommending information for users |
| CN104035957A (en) * | 2014-04-14 | 2014-09-10 | 百度在线网络技术（北京）有限公司 | Search method and device |
| CN104537115A (en) * | 2015-01-21 | 2015-04-22 | 北京字节跳动科技有限公司 | Method and device for exploring user interests |
| CN105574176A (en) * | 2015-12-21 | 2016-05-11 | 北京奇虎科技有限公司 | Hot word recommending method and device with combination of multiple data sources |
| CN105653572A (en) * | 2015-08-20 | 2016-06-08 | 乐视网信息技术（北京）股份有限公司 | Resource processing method and apparatus |
| CN105718184A (en) * | 2014-12-05 | 2016-06-29 | 北京搜狗科技发展有限公司 | Data processing method and apparatus |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103914536B (en) * | 2014-03-31 | 2017-11-07 | 北京百度网讯科技有限公司 | A kind of point of interest for electronic map recommends method and system |
| CN104866530A (en) * | 2015-04-27 | 2015-08-26 | 宁波网传媒有限公司 | Recommendation system and method based on slider scores |
-
2016
- 2016-12-30 CN CN201611253179.7A patent/CN107229666B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102360358A (en) * | 2011-09-28 | 2012-02-22 | 百度在线网络技术（北京）有限公司 | Keyword recommendation method and system |
| CN103761263A (en) * | 2013-12-31 | 2014-04-30 | 武汉传神信息技术有限公司 | Method for recommending information for users |
| CN104035957A (en) * | 2014-04-14 | 2014-09-10 | 百度在线网络技术（北京）有限公司 | Search method and device |
| CN105718184A (en) * | 2014-12-05 | 2016-06-29 | 北京搜狗科技发展有限公司 | Data processing method and apparatus |
| CN104537115A (en) * | 2015-01-21 | 2015-04-22 | 北京字节跳动科技有限公司 | Method and device for exploring user interests |
| CN105653572A (en) * | 2015-08-20 | 2016-06-08 | 乐视网信息技术（北京）股份有限公司 | Resource processing method and apparatus |
| CN105574176A (en) * | 2015-12-21 | 2016-05-11 | 北京奇虎科技有限公司 | Hot word recommending method and device with combination of multiple data sources |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107229666A (en) | 2017-10-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107229666B (en) | A kind of interest heuristic approach and device based on recommender system | |
| US9652524B2 (en) | System and method for creating topic neighborhood visualizations in a networked system | |
| CN102611785B (en) | Personalized active news recommending service system and method for mobile phone user | |
| WO2019019554A1 (en) | Method and apparatus for obtaining recommendation information, and electronic device | |
| US20160171044A1 (en) | Methods And Systems For Using Community Defined Facets Or Facet Values In Computer Networks | |
| US8170916B1 (en) | Related-item tag suggestions | |
| CN103106208B (en) | A kind of streaming medium content in mobile Internet recommends method and system | |
| CN108629665A (en) | A kind of individual commodity recommendation method and system | |
| US8086504B1 (en) | Tag suggestions based on item metadata | |
| CN106339502A (en) | Modeling recommendation method based on user behavior data fragmentation cluster | |
| CN110020094A (en) | A kind of methods of exhibiting and relevant apparatus of search result | |
| Cheng et al. | Video ecommerce++: Toward large scale online video advertising | |
| US20140280549A1 (en) | Method and System for Efficient Matching of User Profiles with Audience Segments | |
| CN103970850B (en) | Site information recommends method and system | |
| US11176574B2 (en) | Contextual secondary content for search results | |
| CN102446180A (en) | Commodity searching method and device adopting same | |
| Li et al. | A method of purchase prediction based on user behavior log | |
| CN103793419A (en) | Information push method and device | |
| CN106682925A (en) | Method and device for recommending advertisement content | |
| CN110348920A (en) | A kind of method and device of recommended products | |
| CN106776860A (en) | One kind search abstraction generating method and device | |
| US20180060427A1 (en) | Navigating a Taxonomy Using Search Queries | |
| CN106961636A (en) | Commodity information recommendation method and device and electronic equipment | |
| Krebs et al. | Non-journalistic competitors of news media brands on Google and YouTube: From solid competition to a liquid media market | |
| Timpson et al. | A student perspective on e-resource discovery: has the Google factor changed publisher platform searching forever? |
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 | ||
| CP03 | Change of name, title or address |
Address after: Room 222, Floor 2, Building 1, Yard A23, North Third Ring West Road, Haidian District, Beijing 100098 Patentee after: Beijing Douyin Information Service Co.,Ltd. Address before: 100098 Building 1, AVIC Plaza, No. 43, North Third Ring West Road, Haidian District, Beijing Patentee before: BEIJING BYTEDANCE TECHNOLOGY Co.,Ltd. |
|
| CP03 | Change of name, title or address |