CN104572982B - Personalized recommendation method and system based on problem guiding - Google Patents
Personalized recommendation method and system based on problem guiding Download PDFInfo
- Publication number
- CN104572982B CN104572982B CN201410856318.XA CN201410856318A CN104572982B CN 104572982 B CN104572982 B CN 104572982B CN 201410856318 A CN201410856318 A CN 201410856318A CN 104572982 B CN104572982 B CN 104572982B
- Authority
- CN
- China
- Prior art keywords
- article
- user
- theme
- tree
- guiding
- 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
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
- G06F16/337—Profile generation, learning or modification
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The present invention provides a kind of personalized recommendation method and system based on problem guiding, and method therein includes：First step：Obtain semantic topic or behavior theme and build multiple problem guiding trees；Wherein, the problem of being inputted according to user obtains semantic topic；Behavior theme is obtained according to the browsing content of user；Data filtering is carried out to user behavior data, user's score data and article metadata, multiple article themes are obtained, is built and the acquired one-to-one problem guiding tree of multiple article themes based on genetic algorithm；Second step：Matched by theme, determine problem guiding tree, hobby theme is obtained according to semantic topic or behavior theme；Hobby theme is matched with problem guiding tree, the problem of matching with hobby theme homing tree is selected, problem output is carried out to user according to homing tree the problem of selecting, obtains user preference data to carry out personalized recommendation.Using the present invention, user's cold start-up problem is can solve the problem that, Consumer's Experience is improved.
Description
Technical field
The present invention relates to article recommended technology field, more specifically, it is related to a kind of personalization based on problem guiding and pushes away
Recommend method and system.
Background technology
At present, personalized recommendation system is widely used in a variety of applications in the Internet, applications, wherein, personalized recommendation algorithm
The click traffic data produced in user interaction process and the content of text data of generation are often all relied on, according to these seas
Amount data mining goes out the potential hobby and demand of user.
Due to this feature of personalized recommendation algorithm, user's cold start-up can be all faced when being recommended for new user
This industry common problem, because new user there is no or only seldom interaction data, in actual application, often all
It is auxiliary of the new user with greater need for commending system, old user becomes more apparent upon to application not to be recommended too dependent on machine on the contrary.
For the cold start-up problem of new user, generally using traditional collective strategy, the method supplied using ranking is most
Hot topic scoring highest article is as recommendations, the characteristics of this recommendation method using global ranking have lost personalization,
Owner is caused to see what content was just as, it is difficult to meet the demand of long-tail user.
Another is for the cold start-up way to solve the problem of new user：By way of booting problem, in user's note
The firsthand data of user is gathered by way of forcing to answer a question after volume, so as to carry out personalization based on these data
Recommend.The data processing method to booting problem is to use information content (Entropy) algorithm at present so that by minimal number of
Problem obtains maximum information content.It is by being divided the evaluation data set of article known users that it, which implements thinking,
Analysis, such as be used as the suitable article of selection come Construct question homing tree by two layers of problem guiding tree shown in Fig. 1.
As shown in figure 1, the corresponding article A- articles M of each node is the Bian representatives in the article most preferably putd question to, homing tree
The feedback (like, do not like and do not know) to upper strata article of user.For example problem guiding tree can inquiry user be first
It is no to like article A, do not know if user answers, homing tree can then put question to whether user likes article C, if user answers
Do not like, then homing tree can then inquire article J, by that analogy.
The process that homing tree is built is actually the process of point group's packet to customer group, it is assumed that known users are to article
Evaluating data, which is concentrated, n user, the marking according to user to article A, is divided into three groups of users and is saved as candidate's of decision tree
Point.By analyzing the similitude (variance) that user evaluates in every group of group, article A is assessed as quality (every group of group of packet problem
Inner evaluation is more similar better), by that analogy, for each child node, it is necessary to be iterated for the customer group in current group
Packet, until homing tree level reaches certain threshold values.Node according to homing tree putd question to can just quickly find with
The common hobby of the similar customer group of active user's behavior, so as to solve the problems, such as the cold start-up in commending system.Fig. 2 is shown
Problem guiding flow, as shown in Fig. 2 its specific problem guiding process is：
After commending system structure, according to user's score data in station and the outer evaluating data in station to each article according to
Evaluation score carries out user grouping, then user's scoring variance in calculating group, and then draw multigroup variance sum；Choose minimum side
The article of poor sum builds its child node as current decision tree node, and according to scoring.During decision tree builds, judge
Whether decision tree depth exceedes pre-set threshold value, if the depth of constructed decision tree exceedes pre-set threshold value, completes decision tree
Structure；Otherwise further to each child node data set repeat the processing procedure of foregoing " packet~structure child node ",
Until completing the structure of decision tree.Decision tree build finish, it is possible to according to the decision tree of the structure User logs in website/
Inquire evaluation of the user to article afterwards using (step S202).
According to the description of above-mentioned flow, bootstrap technique is primarily present lower three problems the problem of current：
1. bootmode is dumb.Such bootstrap technique is bundled in the register flow path of user, and user needs
Each problem is answered according to flow, user is on stream without any dominant right, it is impossible to for a certain class problem
Selection is carried out to answer.
2. problem lacks relevance.Global article is employed in problem selection, is most talked about due to pursuing excessive information amount,
Lack relevance between problem, this mode causes similar algorithm to can be used only on initial user, it is impossible to persistently carried out with user
Interaction.For example, if user merely desires to buy suitable clothes in store, such algorithm can not only put question to related on clothes ask
Topic carries out classification problem guiding.
3. sparse matrix support is low.It is all based on carrying out a point group for the classification of an article marking in problem selection
It is grouped (such as " liking ", " not liking " and " not knowing " three classifications), is answered because user evaluates rating matrix most of
With being all extremely sparse in scene, the article that user evaluates only accounts for the very little part of total article number, so as group refines
Data are reduced, and the number of users of " not knowing " classification can cause the information of all optional articles considerably beyond other categorical measures
Amount is all smaller to be putd question to from without suitable problem, and algorithm is terminated in advance.
In order to solve problems described above, it is desirable to provide a kind of method of effective problem guiding, so that it is cold to solve user
Starting problem and raising Consumer's Experience.
The content of the invention
In view of the above problems, it is an object of the invention to provide a kind of personalized recommendation method based on problem guiding and it is
System, to solve the problems, such as user's cold start-up, improves Consumer's Experience.
According to an aspect of the present invention there is provided a kind of personalized recommendation method based on problem guiding, including two steps
Suddenly；
First step：Obtain semantic topic or behavior theme and build multiple problem guiding trees；Wherein,
The problem of being inputted according to user obtains semantic topic；
Behavior theme is obtained according to the browsing content of user；
Data filtering is carried out to user behavior data, user's score data and article metadata, user, article and happiness is obtained
Multiple article themes of good degree ternary relation, and built and acquired multiple article themes one-to-one corresponding based on genetic algorithm
The problem of homing tree；
Second step：Problem guiding tree is determined by theme matching；Wherein,
Hobby theme is obtained according to semantic topic or behavior theme；
Hobby theme is matched with problem guiding tree, it is determined that the problem of matching with hobby theme homing tree, according to
The problem of determining homing tree carries out problem output to user, obtains user preference data to carry out personalized recommendation.
Furthermore it is preferred that scheme be, the problem of being inputted according to user obtain semantic topic during,
The problem of user is inputted carries out Chinese word segmentation and semantic analysis, obtains semantic topic；Wherein,
During Chinese word segmentation, the paragraph sentence in natural language description text is split using Chinese Word Automatic Segmentation
For word；Chinese Word Automatic Segmentation includes maximum matching algorithm, most long equal word algorithm and minimum variation algorithm；
During semantic analysis, each word in the paragraph sentence of counting user input is similar to article theme
Degree；
During semantic topic is obtained, the semantic topic that maximum similarity theme is the sentence is chosen.
Furthermore it is preferred that scheme be, according to the browsing content of user obtain behavior theme during,
User behavior is obtained according to the browsing content of user, hobby journey of the user to article theme is obtained according to user behavior
Degree, user preferences degree is maximum and article theme of more than pre-set threshold value is used as behavior theme.
Furthermore it is preferred that scheme be, the selection according to user to each article in the list containing multiple articles, obtain user
The article most liked.
Furthermore it is preferred that scheme be to be built based on genetic algorithm and acquired multiple article themes are one-to-one
During problem guiding tree,
User's evaluation data set corresponding to each article theme of acquisition is carried out population Selecting operation, crossing operation,
Mutation operator generates problem guiding tree；Wherein,
During population Selecting operation, winning individual is selected from colony, worst individual is eliminated；Wherein, planting
Initial stage of group's Selecting operation carries out crossing operation using N number of article collection is generated at random, complete every time crossing operation with
After the mutation operator, N number of article collection is randomly selected from candidate item collection, and calculate the article collection population in population Selecting operation
Average fitness, wherein, N be more than 1；
During crossing operation, any two article collection is randomly selected every time, and any two article is handed over
One new article collection of generation is changed, and meets newly-generated article and is concentrated without the article repeated；
During mutation operator, any article that any article generated after crossing operation is concentrated is selected at random
Replacing is taken, and meets the article after variation and is concentrated without the article repeated, and obtains being averaged for the article collection population after variation
Fitness.
Furthermore it is preferred that scheme be, the article collection population generated according to crossing operation and the mutation operator it is average suitable
The average fitness and threshold values of article collection population in response, population Selecting operation, sub- section is carried out according to current item theme
Point groups of users distribution, generates problem guiding tree.
According to another aspect of the present invention there is provided a kind of personalized recommendation system based on problem guiding, including：
Semantic topic acquiring unit, obtains semantic topic the problem of for being inputted according to user；
Behavior theme acquiring unit, for obtaining behavior theme according to the browsing content of user；
Theme acquiring unit, for carrying out data filtering to user behavior data, user's score data and article metadata,
And obtain multiple article themes of user, article and fancy grade ternary relation,
Problem guiding tree generation unit, is corresponded for being built based on genetic algorithm with acquired multiple article themes
The problem of homing tree；
Theme acquiring unit is liked, for according to semantic topic or behavior theme, obtaining hobby theme；
Theme matching unit, for hobby theme to be matched with problem guiding tree, it is determined that matching with hobby theme
The problem of homing tree；
Problem output unit, for carrying out problem output to user according to homing tree the problem of determination, obtains user preferences
Data are so as to carry out personalized recommendation.
Furthermore it is preferred that scheme be, the selection according to user to each article in the list containing multiple articles, obtain user
The article most liked.
Furthermore it is preferred that scheme be that problem guiding tree generation unit includes：
Population Selecting operation module, for selecting winning individual from colony, eliminates worst individual；Wherein, in population
The initial stage of Selecting operation carries out crossing operation using N number of article collection is generated at random, completes crossing operation with becoming every time
After xor, N number of article collection will be randomly selected from candidate item collection, and calculate the flat of article collection population in population Selecting operation
Equal fitness, wherein, N is more than 1；
Crossing operation module, is swapped for randomly selecting any two article collection, and to internal any two article
A new article collection is produced, and meets newly-generated article and is concentrated without the article repeated；
Mutation operator module, for being randomly selected to any article that any article generated after crossing operation is concentrated
Change, and meet the article after variation and concentrate without the article repeated, and obtain the average suitable of the article collection population after variation
Response.
Furthermore it is preferred that scheme be that problem guiding tree generation unit also includes：
Groups of users distributes module, for the average adaptation of the article collection population generated according to crossing operation and mutation operator
Degree, the average fitness of article collection population in population Selecting operation and pre-set threshold value, son is carried out according to current item theme
Node users group allocation, generates problem guiding tree.
It was found from technical scheme above, personalized recommendation method and system of the invention based on problem guiding pass through
Semantic topic is obtained, the problem of guiding can not be flexibly matched with conventional method is solved, by increasing capacitance it is possible to increase the flexibility of user preferences is obtained；
By obtaining behavior theme, the problem of traditional problem boot flow can only be fixed using in register flow path is solved, can be real-time
Aid in user's purchase commodity；Forest is guided by using genetic algorithm Construct question, is grouped during solution Sparse of low quality
The problem of, it is possible to increase the flexibility of user grouping and the acquisition of Global Information amount.
In order to realize above-mentioned and related purpose, one or more aspects of the invention include will be explained in below and
The feature particularly pointed out in claim.Some illustrative aspects of the present invention are described in detail in following explanation and accompanying drawing.
However, some modes in the various modes for the principle that the present invention only can be used that these aspects are indicated.In addition, of the invention
It is intended to include all these aspects and their equivalent.
Brief description of the drawings
By reference to the explanation and the content of claims below in conjunction with accompanying drawing, and with to the present invention more comprehensively
Understand, other purposes of the invention and result will be more apparent and should be readily appreciated that.In the accompanying drawings：
Two layers of problem guiding tree schematic diagram that Fig. 1 is；
Fig. 2 is problem guiding method flow schematic diagram；
Fig. 3 is the personalized recommendation method schematic flow sheet based on problem guiding according to the embodiment of the present invention；
Fig. 4 is that the example flow of the method for the personalized recommendation based on problem guiding according to the embodiment of the present invention is illustrated
Figure；
Fig. 5 is the personalized recommendation system structured flowchart based on problem guiding according to the embodiment of the present invention；
Fig. 6 is the structured flowchart of the semantic topic acquiring unit according to the embodiment of the present invention；
Fig. 7 is the structured flowchart of the behavior theme acquiring unit according to the embodiment of the present invention；
Fig. 8 is the homing tree generation module structured flowchart according to the problem of the embodiment of the present invention.
Identical label indicates similar or corresponding feature or function in all of the figs.
Embodiment
In the following description, for purposes of illustration, in order to provide the comprehensive understanding to one or more embodiments, explain
Many details are stated.It may be evident, however, that these embodiments can also be realized in the case of these no details.
From the point of view of foregoing prior art, problem guiding is only accounted for from the angle of maximum fault information, in user's body
Test without any optimization, user can lose patience and terminate question and answer in advance.Simultaneously in the selection of problem, commented using fixed
Divide threshold values to be classified, be due to that article Evaluations matrix has seriously although being effectively controlled on computation complexity
Loose line, can cause the uniformity in group very low, and influence follow-up recommendation results based on such user grouping.
For these reasons, the present invention proposes using acquisition semantic topic, obtains behavior theme and based on heredity calculation
Three aspects of method Construct question homing tree come improve bootmode flexibility and relevance and solving matrix it is openness.
The specific embodiment of the present invention is described in detail below with reference to accompanying drawing.
Fig. 3 shows the personalized recommendation method flow according to embodiments of the present invention based on problem guiding.
As shown in figure 3, the personalized recommendation method based on problem guiding that the present invention is provided mainly includes two steps.
First step：S310：Obtain semantic topic or behavior theme and build multiple problem guiding trees.
Specifically, the problem of being inputted according to user, obtains semantic topic；According to the browsing content of user, behavior master is obtained
Topic；Data filtering is carried out to user behavior data, user's score data and article metadata, and obtains user, article and hobby
Multiple article themes of degree ternary relation, are built based on genetic algorithm and are asked correspondingly with acquired multiple article themes
Inscribe homing tree.
Wherein, relation constitutes multiple article themes between user, article and fancy grade ternary, for example, some users
Hobby action movie, some user preferences horror films, some user preferences comedies etc., these action movies, horror film, comedy are all
It is article theme, the slice, thin piece of different types constitutes multiple article themes.
Second step：S320：Matched by theme, determine problem guiding tree.
Specifically, hobby theme is obtained according to semantic topic or behavior theme, hobby theme is carried out with problem guiding tree
Matching, it is determined that the problem of matching with hobby theme homing tree, problem output is carried out according to homing tree the problem of determination to user,
User preference data is obtained to carry out personalized recommendation.
Wherein, in step S310, the problem of user is inputted carries out Chinese word segmentation and semantic analysis, obtains semantic main
Topic；User behavior is obtained according to the browsing content of user, fancy grade of the user to article theme is obtained according to user behavior, and
Judge whether its fancy grade exceedes pre-set threshold value, choose user preferences degree maximum and more than the article theme of pre-set threshold value
For behavior theme；Population Selecting operation, crossing operation, mutation operator generation problem guiding are carried out to each article theme of acquisition
Tree.
It is to obtain semantic topic and behavior theme and build multiple problem guiding trees in the core of the present invention, below will be detailed
It is thin to illustrate how to obtain semantic topic, obtain behavior theme and build multiple problem guiding trees.
Firstth, semantic topic is obtained
The problem of user is inputted carries out Chinese word segmentation and semantic analysis, obtains semantic topic.
Wherein, Chinese word segmentation refers to be split as the paragraph sentence in natural language description text using Chinese Word Automatic Segmentation
Word.Conventional Chinese Word Automatic Segmentation includes maximum matching algorithm, most long equal word algorithm and minimum variation algorithm.
Semantic analysis refers to each word journey similar to article theme in the paragraph sentence of statistical summaries user input
Degree.If analyzing paragraph by Chinese word segmentation in the paragraph of user's input includes n word, while setting the article of pre- prefinishing
Theme (such as action movie, horror film etc.) has m, is respectively [T1,T2,…Tm], word i and article theme j phase are set in addition
Closing property is PIj,The similitude of word and article theme can be by calculating with article theme in language material training data (for example, film
Metadata story of a play or opera description information, the film review text data of user etc.) in co-occurrence probabilities and obtain, its formula is expressed as：
Wherein, CiRepresent the number of times that word i occurs in language material；CjRepresent the number of times that article theme j occurs in language material；
CijThe number of times occurred jointly for word i and article theme.
Finally paragraph sentence s and article theme j similarity degree are：
Also, the semantic topic that maximum similarity theme t is the sentence is chosen, its formula is expressed as：T=argmaxjPsj
Therefore,, can by using the pattern and semantic analysis mode of dialogue chat during semantic topic is obtained
To extract realm information from user session content, so as to carry out recommendation guiding for specific field, conventional method is solved
The defect of booting problem can not be flexibly matched with, increase obtains the flexibility of user preferences.
Secondth, behavior theme is obtained
According to the browsing content of user, user behavior is obtained, and whether theme is liked according to the user behavior of acquisition statistics
More than threshold values, finally using user preferences degree is maximum and fancy grade exceedes the theme of pre-set threshold value and is used as behavior theme.
Wherein, obtain user behavior refer to collection storage user browse web sites, the operation behavior of Mobile solution, such as video
Broadcasting, download, comment, collection in website etc. are operated.
Whether statistics hobby theme, which exceedes threshold values, refers to calculate happiness of the user to article theme by the operation behavior of user
Good degree, and judge whether fancy grade exceedes pre-set threshold value.
The calculating of fancy grade can carry out matching primitives by business rule, for example：Play video and add 1 point, download and add
1.5 points, collection Jia 2 and graded.Fancy grade of the user to an article can be counted by collecting multiple vaild acts, simultaneously
Collect a user to multiple article fancy grades under an article theme, and then obtain a user to an article theme
Hobby value.
During behavior theme is obtained, obtained by gathering user behavior data and liking digging technology using behavior
User's article theme interested, actively carries out real-time interactive by way of pop-up with user, solves traditional problem guiding stream
The problem of journey can only be fixed using in register flow path, can aid in user to buy commodity in real time.
Therefore, obtaining semantic topic can support to link up with user's direct dialogue text, and obtaining behavior theme can support
Boot flow is actively intervened during user browses web sites, both modes can improve the flexibility of guiding.
3rd, Construct question homing tree
During Construct question homing tree, first, to user behavior data, user's score data and article metadata
Data filtering is carried out, and obtains multiple article themes of user, article and fancy grade ternary relation；Then, based on heredity calculation
Method is built guides gloomy with the acquired one-to-one problem guiding tree of multiple article themes, the problem of forming multiple article themes
Woods.
, wherein it is desired to explanation, in the present invention, the choosing according to user to each article in the list containing multiple articles
Select, obtain the article that user most likes.
Traditional problem bootmode (is liked, do not like and not known to the hobby opinion of single article for inquiry user
Road), the bootmode that the present invention is provided is for inquiry user to the fancy grade of multiple articles (for example, liking in following 4 films
Which portion), therefore, the problem of of the invention bootmode have bigger level of coverage.
Specifically, use is sorted out according to information filterings such as user behavior data, user's score data and article metadata
Family to the fancy grade of article (for example：1-5 points, 1 point is disagreeable, and 5 points are to like), and according to the relation of article and article theme
Consumer articles score data is classified according to article theme；User under multiple article themes, article, happiness can be obtained
The ternary relation matrix of good degree.If user i is R to article j evaluating dataij∈ R, separately set the article collection under article theme t
For St, the evaluation data set under article theme t is expressed as：
R (t)={ Rij∈R|j∈St}
Genetic algorithm is based on to the data set under each article theme and distinguishes Construct question homing tree, so as to construct multiple
The problem of article theme, guides forest to solve the problems, such as the semantic association during problem guiding.
That is, by using data filtering techniques, category filter is carried out to data, for the number of different article themes
Different article problem of subject homing trees are built according to collection, and the semantic topic and behavior theme that are gone out by semantic analysis are selected
Theme, and then homing tree the problem of select most close with user preferences theme are liked, so as to provide the guiding of relevance
Problem.
During being built based on genetic algorithm with the acquired one-to-one problem guiding tree of multiple article themes,
Population Selecting operation, crossing operation, mutation operator generation problem guiding tree are carried out to each article theme of acquisition, so as to build
The problem of going out multiple article themes guides forest.
Specifically, population Selecting operation refers to select winning individual from colony in genetic algorithm, eliminates inferior
Body.The structure of whole homing tree needs to calculate each node using genetic algorithm, chooses optimal division article collection
(for example：Film《Red sorghum》,《Farewell My Concubine》,《Live》).
And it is possible to obtain the fitness for assessing division article collection by the following method：If article theme includes N number of article
W1,W2,...Wn, then N+1 group (child node) G can be separated the users into according to known users score data collection1,G2,...Gn,
Gn+1If scorings of the user i to article theme j is Rij(setting number range as 1-5), the definition of grouping set 1 to n is：
Gj=i ∈ U | Rij≥3}
Wherein, U is the user complete or collected works of present node, GjWhat set was represented is to article WjThe user group liked；It is any one
Individual user can belong to multiple packets (child node) simultaneously, and packet N+1 definition is：
(child node) G in one groupiFitness analysis standard GRiFor：
Wherein, μikRepresent group GiIn all users to article K average score, if gathering the user that article K gives a mark
For R (k), then：
So, article collection W comprehensive fitness degree evaluation criteria GR is split for any one componentwFor the fitness of all groups
Sum, its formula is expressed as：
, wherein it is desired to which explanation, the population Selecting operation initial stage is next using N number of article collection progress is generated at random
The crossing operation of step, after Population breeding is completed every time (i.e.：After crossing operation and mutation operator), will from candidate item collection with
Machine chooses N number of article collection, and the selection probability of each article collection is directly proportional to fitness and can be expressed as：Pw∝GRw, wherein, N is big
In 1；Also, the fitness of the article collection population in population Selecting operation obtains its average fitness.
During crossing operation, any two article collection is randomly selected every time, and internal any two article is entered
Row, which is exchanged, produces a new article collection, and meets newly-generated article concentration without the article repeated；This is to ensure
Stability of the population quantity in genetic iteration calculating, newly-generated offspring's article collection quantity is greater than the quantity of parent article collection
(such as twice quantity), enough candidate's clusters are provided for population Selecting operation module.
During mutation operator, any article that any article generated after crossing operation is concentrated is selected at random
Replacing is taken, and meets the article after variation and is concentrated without the article repeated, and obtains being averaged for the article collection population after variation
Fitness.
Then according in crossing operation and the average fitness of the article collection population of mutation operator generation, population Selecting operation
Article collection population average fitness and threshold values, according to current topic carry out children User group allocation, generate problem
Homing tree.
That is, completing average fitness and the father of article collection population newly-generated after crossing operation and mutation operator
Generation population is (i.e.：Population after Selecting operation) the difference of average fitness whether be less than threshold values, or whether iterations exceed
Pre-set threshold value.If it is, the article theme for completing present node is chosen, children User group is carried out according to current item theme
The distribution of group, the article collection for then completing whole all nodes of tree according to default depth is chosen, and generates problem guiding tree.If
It is not above, then returns to population selection computing and repeat to calculate.
Above-mentioned is the detailed process of homing tree the problem of generation, and the problem of having ultimately produced multiple article themes guides gloomy
Woods.The support of many problem guiding trees is needed because theme is matched, the number of multiple different article themes is exported by data filtering
Problem guiding tree is set up respectively according to collection.In order to which homing tree can use distribution simultaneously the problem of improving efficiency, different article themes
The computing simultaneously on many physical machines of row Computational frame.
It is the guiding that many different themes are built by topic distillation data based on genetic algorithm Construct question homing tree
Tree, so that the problem of constructing multi-threaded guides forest, therefore, it is possible to match the problem of choosing different article themes by theme
Homing tree is guided, and improves the correlation of bootmode.The problem of problem guiding tree classification using multiple different articles without
It is that different scorings are interval, so as to reduce the influence that Deta sparseness is brought, improves the uniformity in different user group.
In step s 320, first, the hobby theme of user is determined according to the semantic topic of acquisition and behavior theme；Then
The hobby theme of the user of determination and problem guiding forest are carried out into theme to match, it is determined that the problem of with hobby theme matching draws
Tree is led, problem output is carried out to user according to homing tree the problem of determination.
Specifically, choose the problem of matching homing tree with user to enter according to the semantic topic of acquisition or behavior theme
Row is interactive, that is to say, that guide the article collection in tree node to be putd question to user according to the problem of selecting.For example：You like with
Which lower articleOption one：Article A, option 2：Article B, option 3：Article C, option 4：All do not like or do not know
The genetic algorithm Construct question homing tree of the present invention, chooses a scoring interval (for example：Scoring is more than 3 and represents happiness
Joyous scoring is interval) different articles as division option (i.e.：The user of identical items is liked to be divided into one group), rather than a thing
The different scorings of product are interval alternatively (article according to liking, disagreeable be divided into three groups of users with not knowing), this
The problem of mode of kind causes to be grouped of low quality when can solve the problem that Sparse is (it is, solve most of users without scoring
One group is divided into, the problem of different groups of customer group volume deviation is very big is caused), improve the flexibility of user grouping and whole
The acquisition of body information content.
Method in order to which the personalized recommendation based on problem guiding is further described, Fig. 4 is shown according to the present invention
The example flow of the method for the personalized recommendation based on problem guiding of embodiment.
As shown in figure 4, the example flow of the method for the personalized recommendation based on problem guiding that the present invention is provided includes：
S401：Start；
S402：User logs in website application；
Step S403-S406, step S407-S410 are performed respectively；
S403：User directly inputs problem；
S404：Chinese word segmentation；
S405：Semantic analysis；
S406：Obtain semantic topic.
S407：User directly inputs problem；
S408：Record user behavior；
S409：Whether statistics hobby theme exceedes threshold values；If so, step S410 is performed, if it is not, performing step S408；
S410：Acquisition behavior theme.And,
S411：User's score data；
S412：User behavior data；
S413：Article metadata；
S414：Data conversion is carried out according to theme；
That is, carrying out data filtering to user's score data, user behavior data and article metadata, get many
The ternary relation of individual article theme.
Then, the problem of genetic algorithm builds different themes homing tree is carried out according to different themes, forms problem guiding
Forest, i.e.,：Step S415-S419 is performed,
S415：Population Selecting operation；
S416：Crossing operation；
S417：Mutation operator；
S418：Whether optimal result exceeds threshold values；If so, step S419 is performed, if it is not, performing step S415；
S419：Generate problem guiding tree；
The tree-like guiding forest of being a problem of multiple article problem guidings.
S420：Theme is matched；
It is, determining the hobby theme of user according to the semantic topic of acquisition or behavior theme；Then by the use of determination
The hobby theme at family carries out theme with problem guiding forest and matched, and selects the problem of matching with hobby theme homing tree.
S421：Output problem；
It is, guiding the article collection in tree node to be putd question to user according to the problem of selecting.
In the embodiment shown in fig. 4, it can be supported and user's direct dialogue text ditch by semantic topic extraction assembly
It is logical, it can support actively to intervene boot flow during user browses web sites by behavior subject distillation, both mode energy
Enough flexibilities for improving guiding；Based on genetic algorithm Construct question homing tree, the mode of conventional construction problem guiding tree is optimized,
With multiple articles as division item, the number of candidate's division item can increase with number of articles exponentially, using traditional traversal
Contrast can not choose optimal division item, i.e.,：By Selecting operation, crossing operation and mutation operator, solve multiple articles and choose
Multiple problems, it is ensured that the article option information amount of selection is maximum.
Corresponding with the above method, the present invention also provides a kind of personalized recommendation system based on problem guiding, and Fig. 5 is shown
Personalized recommendation system logical construction based on problem guiding according to embodiments of the present invention.
As shown in figure 5, the personalized recommendation system 500 based on problem guiding that the present invention is provided includes：Semantic topic is obtained
Take unit 510, behavior theme acquiring unit 520, theme acquiring unit 530, problem guiding tree generation unit 540, hobby theme
Acquiring unit 550, theme matching unit 560 and problem output unit 570.
Wherein, semantic topic acquiring unit 510 is used to obtain semantic topic the problem of input according to user.
Behavior theme acquiring unit 520 is used to obtain behavior theme according to the browsing content of user.
Theme acquiring unit 530 is used to carry out data mistake to user behavior data, user's score data and article metadata
Filter, and obtain multiple article themes of user, article and fancy grade ternary relation.
Problem guiding tree generation unit 540 is used to build and a pair of acquired multiple article themes 1 based on genetic algorithm
The problem of answering homing tree.
Liking theme acquiring unit 550 is used for according to semantic topic or behavior theme, obtains hobby theme.
Theme matching unit 560 is used to be matched hobby theme with problem guiding tree, it is determined that with liking theme phase
With the problem of homing tree.
Problem output unit 570 is used to carry out problem output to user according to homing tree the problem of determination, obtains user's happiness
Good data are so as to carry out personalized recommendation.
Fig. 6 shows the structure of semantic topic acquiring unit according to embodiments of the present invention, as shown in fig. 6, semantic topic
Acquiring unit 510 includes：Chinese word segmentation module 511, semantic module 512 and semantic topic acquisition module 513.
Wherein, Chinese word segmentation module 511 is used to utilize Chinese Word Automatic Segmentation by the paragraph sentence in natural language description text
Son is split as word；Chinese Word Automatic Segmentation includes maximum matching algorithm, most long equal word algorithm and minimum variation algorithm；
Each word that semantic module 512 is used in the paragraph sentence that counting user is inputted is similar to article theme
Degree；
Semantic topic acquisition module 513 is used to choose the semantic topic that maximum similarity theme is the sentence.
Fig. 7 shows the structure of behavior theme acquiring unit according to embodiments of the present invention, as shown in fig. 7, behavior theme
Acquiring unit 520 includes：User behavior acquisition module 521, fancy grade acquisition module 522 and behavior theme acquisition module 523.
Wherein, user behavior acquisition module 521 is used to obtain user behavior according to the browsing content of user；
Fancy grade acquisition module 522 is used to obtain fancy grade of the user to article theme according to user behavior；
Behavior theme acquisition module 523 is used to choose user preferences degree maximum and more than the article theme of pre-set threshold value
It is used as behavior theme.
In addition, selection of the theme acquiring unit 530 according to user to each article in the list containing multiple articles, obtains and uses
The article that family is most liked.
In addition, the problem of Fig. 8 shows according to embodiments of the present invention homing tree generation module structure, as shown in figure 8, problem
Homing tree generation unit 540 includes：Population Selecting operation module 541, crossing operation module 542, mutation operator module 543 and use
Family group allocation module 544.
Wherein, population Selecting operation module 541 is used to select winning individual from colony, eliminates worst individual；Wherein,
In the initial stage of population Selecting operation using N number of article collection progress crossing operation is generated at random, complete to intersect fortune each
Calculate with after mutation operator, randomly selecting N number of article collection from candidate item collection, and calculate the article collection population in population Selecting operation
Average fitness, wherein, N be more than 1.
Crossing operation module 542 is used to randomly select any two article collection, and swaps production to any two article
A raw new article collection, and newly-generated article concentration is met without the article repeated.
Mutation operator module 543 is used to select any article that any article generated after crossing operation is concentrated at random
Replacing is taken, and meets the article after variation and is concentrated without the article repeated, and obtains being averaged for the article collection population after variation
Fitness.
Groups of users distribution module 544 is used for the average suitable of the article collection population generated according to crossing operation and mutation operator
The average fitness and pre-set threshold value of article collection population in response, population Selecting operation, are carried out according to current item theme
Children User group allocation, generates problem guiding tree.
The more specific interaction of above-mentioned each module or unit, may refer to the description in method flow, no longer go to live in the household of one's in-laws on getting married herein
State.
Personalized recommendation method proposed by the present invention based on problem guiding can be seen that by above-mentioned embodiment and be
System, by obtaining semantic topic, can solve the problem that the problem of guiding can not be flexibly matched with conventional method, increase obtains user preferences
Flexibility；By obtaining behavior theme, it can solve the problem that traditional problem boot flow can only be fixed and use asking in register flow path
Topic, aids in user's purchase commodity in real time；Forest is guided by using genetic algorithm Construct question, is led when can solve the problem that Sparse
The problem of packet is of low quality is caused, the flexibility of user grouping and the acquisition of Global Information amount is improved.
Describe and pushed away according to the personalization proposed by the present invention based on problem guiding in an illustrative manner above with reference to accompanying drawing
Recommend method and system.It will be understood by those skilled in the art, however, that for the invention described above proposed based on problem guiding
Personalized recommendation method and system, can also make various improvement on the basis of present invention is not departed from.Therefore, it is of the invention
Protection domain should be determined by the content of appended claims.
Claims (10)
1. a kind of personalized recommendation method based on problem guiding, including two steps；
First step：Obtain semantic topic or behavior theme and build multiple problem guiding trees；Wherein,
The problem of being inputted according to user obtains semantic topic,
Behavior theme is obtained according to the browsing content of user；
Data filtering is carried out to user behavior data, user's score data and article metadata, user, article and hobby journey is obtained
Multiple article themes of ternary relation are spent, and are asked correspondingly with acquired multiple article themes based on genetic algorithm structure
Inscribe homing tree；
Second step：Problem guiding tree is determined by theme matching；Wherein,
Hobby theme is obtained according to institute's semantic topic or the behavior theme；
The hobby theme is matched with described problem homing tree, it is determined that the problem of matching with the hobby theme is guided
Tree, problem output is carried out according to identified problem guiding tree to user, is obtained user preference data and is pushed away so as to carry out personalization
Recommend.
2. the personalized recommendation method as claimed in claim 1 based on problem guiding, wherein, the problem of being inputted according to user
During obtaining semantic topic,
The problem of user is inputted carries out Chinese word segmentation and semantic analysis, obtains semantic topic；Wherein,
During the Chinese word segmentation, the paragraph sentence in natural language description text is split using Chinese Word Automatic Segmentation
For word；The Chinese Word Automatic Segmentation includes maximum matching algorithm, most long equal word algorithm and minimum variation algorithm；
During the semantic analysis, each word and article theme in the paragraph sentence for the problem of counting user is inputted
Similarity degree；
During the acquisition semantic topic, it is acquired semantic topic to choose maximum similarity theme.
3. the personalized recommendation method as claimed in claim 1 based on problem guiding, wherein, in the browsing content according to user
During acquisition behavior theme,
User behavior is obtained according to the browsing content of user；
Fancy grade of the user to article theme is obtained according to the user behavior, and by user preferences degree maximum and exceeded
The article theme of pre-set threshold value is used as behavior theme.
4. the personalized recommendation method as claimed in claim 1 based on problem guiding, wherein,
Selection according to user to each article in the list containing multiple articles, obtains the article that user most likes.
5. the personalized recommendation method based on problem guiding as claimed in claim 4, wherein, based on genetic algorithm build with
During the acquired one-to-one problem guiding tree of multiple article themes,
Population Selecting operation, crossing operation, variation are carried out to user's evaluation data set corresponding to each article theme of acquisition
Computing generates problem guiding tree；Wherein,
During the population Selecting operation, winning individual is selected from colony, worst individual is eliminated；Wherein, in institute
The initial stage for stating population Selecting operation carries out the crossing operation using N number of article collection is generated at random, completes institute every time
Crossing operation is stated with after the mutation operator, randomly selecting N number of article collection from candidate item collection, and calculate the population selection fortune
The average fitness of article collection population in calculation, wherein, N is more than 1；
During the crossing operation, any two article collection is randomly selected every time, and any two article is handed over
One new article collection of generation is changed, and meets newly-generated article and is concentrated without the article repeated；
During the mutation operator, any article that any article for being generated after the crossing operation is concentrated is carried out with
Machine, which is chosen, to be changed, and meets the article concentration after variation without the article repeated, and obtains the article collection population after variation
Average fitness.
6. the personalized recommendation method as claimed in claim 5 based on problem guiding, wherein,
The average fitness of the article collection population generated according to the crossing operation and the mutation operator, population selection are transported
The average fitness and threshold values of article collection population in calculation, children User group allocation is carried out according to current item theme,
Generate problem guiding tree.
7. a kind of personalized recommendation system based on problem guiding, including：
Semantic topic acquiring unit, obtains semantic topic the problem of for being inputted according to user；
Behavior theme acquiring unit, for obtaining behavior theme according to the browsing content of user；
Theme acquiring unit, for carrying out data filtering to user behavior data, user's score data and article metadata, and is obtained
Multiple article themes of family, article and fancy grade ternary relation are taken,
Problem guiding tree generation unit, is asked correspondingly for being built based on genetic algorithm with acquired multiple article themes
Inscribe homing tree；
Theme acquiring unit is liked, for according to institute's semantic topic or the behavior theme, obtaining hobby theme；
Theme matching unit, for guiding forest to be matched hobby theme and the described problem, it is determined that with the hobby
The problem of theme matches homing tree；
Problem output unit, for carrying out problem output to user according to homing tree the problem of determination, obtains user preference data
So as to carry out personalized recommendation.
8. the personalized recommendation system as claimed in claim 7 based on problem guiding, wherein,
Selection of the theme acquiring unit according to user to each article in the list containing multiple articles, obtains user and most likes
Article.
9. the personalized recommendation system as claimed in claim 8 based on problem guiding, wherein, the generation of described problem homing tree is single
Member based on genetic algorithm during being built with the acquired one-to-one problem guiding tree of multiple article themes, to obtaining
Each article theme corresponding to user's evaluation data set carry out population Selecting operation, crossing operation, mutation operator；Also,
Described problem homing tree generation unit includes：
Population Selecting operation module, for selecting winning individual from colony, eliminates worst individual；Wherein, in the population
The initial stage of Selecting operation carries out the crossing operation using N number of article collection is generated at random, and the intersection is completed every time
Computing from candidate item collection with after the mutation operator, randomly selecting N number of article collection, and calculate in the population Selecting operation
The average fitness of article collection population, wherein, wherein, N is more than 1；
Crossing operation module, for randomly selecting any two article collection, and swaps generation one to any two article
New article collection, and newly-generated article concentration is met without the article repeated；
Mutation operator module, for being randomly selected to any article that any article generated after the crossing operation is concentrated
Change, and meet the article after variation and concentrate without the article repeated, and obtain the average suitable of the article collection population after variation
Response.
10. the personalized recommendation system as claimed in claim 9 based on problem guiding, wherein, the generation of described problem homing tree
Unit also includes：
Groups of users distributes module, for being averaged for the article collection population that is generated according to the crossing operation with the mutation operator
The average fitness and pre-set threshold value of article collection population in fitness, the population Selecting operation, according to current item master
Topic carries out children User group allocation, generates problem guiding tree.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410856318.XA CN104572982B (en) | 2014-12-31 | 2014-12-31 | Personalized recommendation method and system based on problem guiding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410856318.XA CN104572982B (en) | 2014-12-31 | 2014-12-31 | Personalized recommendation method and system based on problem guiding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104572982A CN104572982A (en) | 2015-04-29 |
| CN104572982B true CN104572982B (en) | 2017-10-31 |
Family
ID=53089044
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410856318.XA Active CN104572982B (en) | 2014-12-31 | 2014-12-31 | Personalized recommendation method and system based on problem guiding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104572982B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106126541B (en) * | 2016-06-14 | 2019-07-19 | 华南师范大学 | Personalized solution recommended method and system based on big data knowledge base |
| CN106776701B (en) * | 2016-11-14 | 2020-04-07 | 东软集团股份有限公司 | Problem determination method and device for item recommendation |
| CN106649842A (en) * | 2016-12-30 | 2017-05-10 | 上海博泰悦臻电子设备制造有限公司 | Cross recommendation method based on fusion data, system and vehicle machine |
| CN107992459A (en) * | 2017-10-12 | 2018-05-04 | 如是科技（大连）有限公司 | Resume generation method and device |
| CN107992610A (en) * | 2017-12-15 | 2018-05-04 | 泰康保险集团股份有限公司 | A kind of automatic question-answering method, device, equipment and storage medium |
| CN108228910B (en) * | 2018-02-09 | 2023-05-12 | 艾凯克斯（嘉兴）信息科技有限公司 | Method for applying cyclic neural network to association selection problem |
| CN110399471A (en) * | 2018-04-25 | 2019-11-01 | 北京快乐智慧科技有限责任公司 | A kind of guiding situational dialogues method and system |
| CN110930969B (en) * | 2019-10-14 | 2024-02-13 | 科大讯飞股份有限公司 | Background music determining method and related equipment |
| CN112288456A (en) * | 2020-01-22 | 2021-01-29 | 北京沃东天骏信息技术有限公司 | Article selection method and apparatus |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006004441A (en) * | 2005-06-24 | 2006-01-05 | Yokogawa Electric Corp | Information service system |
| CN101296154A (en) * | 2007-04-28 | 2008-10-29 | 上海盛大网络发展有限公司 | Automatic answering system |
| CN101354714A (en) * | 2008-09-09 | 2009-01-28 | 浙江大学 | Method for recommending problem based on probability latent semantic analysis |
| CN101409690A (en) * | 2008-11-26 | 2009-04-15 | 北京学之途网络科技有限公司 | Method and system for obtaining internet user behaviors |
| CN103455638A (en) * | 2013-09-26 | 2013-12-18 | 中国科学院自动化研究所 | Behavior knowledge extracting method and device combining reasoning and semi-automatic learning |
| CN103995858A (en) * | 2014-05-15 | 2014-08-20 | 北京航空航天大学 | Individualized knowledge active pushing method based on task decomposition |
| JP2014526747A (en) * | 2011-09-13 | 2014-10-06 | ▲騰▼▲訊▼科技（深▲セン▼）有限公司 | Data alignment method and apparatus |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101339551B (en) * | 2007-07-05 | 2013-01-30 | 日电（中国）有限公司 | Natural language query demand extension equipment and its method |
| BR112013030777A2 (en) * | 2011-06-03 | 2016-12-06 | Uc Group Ltd | systems and methods for user registration, validation and monitoring on multiple websites |
-
2014
- 2014-12-31 CN CN201410856318.XA patent/CN104572982B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006004441A (en) * | 2005-06-24 | 2006-01-05 | Yokogawa Electric Corp | Information service system |
| CN101296154A (en) * | 2007-04-28 | 2008-10-29 | 上海盛大网络发展有限公司 | Automatic answering system |
| CN101354714A (en) * | 2008-09-09 | 2009-01-28 | 浙江大学 | Method for recommending problem based on probability latent semantic analysis |
| CN101409690A (en) * | 2008-11-26 | 2009-04-15 | 北京学之途网络科技有限公司 | Method and system for obtaining internet user behaviors |
| JP2014526747A (en) * | 2011-09-13 | 2014-10-06 | ▲騰▼▲訊▼科技（深▲セン▼）有限公司 | Data alignment method and apparatus |
| CN103455638A (en) * | 2013-09-26 | 2013-12-18 | 中国科学院自动化研究所 | Behavior knowledge extracting method and device combining reasoning and semi-automatic learning |
| CN103995858A (en) * | 2014-05-15 | 2014-08-20 | 北京航空航天大学 | Individualized knowledge active pushing method based on task decomposition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104572982A (en) | 2015-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104572982B (en) | Personalized recommendation method and system based on problem guiding | |
| CN105095219B (en) | Micro-blog recommendation method and terminal | |
| US20150081725A1 (en) | System and method for actively obtaining social data | |
| US10223438B1 (en) | System and method for digital-content-grouping, playlist-creation, and collaborator-recommendation | |
| CN103338403B (en) | Individual character program commending method in radio data system and this system | |
| CN101482884A (en) | Cooperation recommending system based on user predilection grade distribution | |
| CN102411754A (en) | Personalized recommendation method based on commodity property entropy | |
| CN102332025A (en) | Intelligent vertical search method and system | |
| CN108920577A (en) | Television set intelligently recommended method | |
| US20170193531A1 (en) | Intelligent Digital Media Content Creator Influence Assessment | |
| CN103559622A (en) | Characteristic-based collaborative filtering recommendation method | |
| Okazaki et al. | How to mine brand Tweets: Procedural guidelines and pretest | |
| CN104268130A (en) | Social advertising facing Twitter feasibility analysis method | |
| CN107423335A (en) | A kind of negative sample system of selection for single class collaborative filtering problem | |
| CN105809275A (en) | Item scoring prediction method and apparatus | |
| Chaturvedi et al. | Recommender system for news articles using supervised learning | |
| CN109508407A (en) | The tv product recommended method of time of fusion and Interest Similarity | |
| CN106021428A (en) | KNN and three-way decision-based movie recommendation method | |
| Bogaert et al. | Predicting Self‐declared movie watching behavior using Facebook data and Information‐Fusion sensitivity analysis | |
| Ashraf et al. | Personalized news recommendation based on multi-agent framework using social media preferences | |
| US20140101064A1 (en) | Systems and Methods for Automated Reprogramming of Displayed Content | |
| CN117474631A (en) | Recommendation algorithm for data mining and emotion analysis based on e-commerce platform comments | |
| CN109299368A (en) | A kind of method and system for the intelligent personalized recommendation of environmental information resource AI | |
| CN113657766A (en) | Tourist attraction joy index metering method based on tourist multi-metadata | |
| Adilah et al. | An augmented method of selecting fashion talent by adding social media characteristic |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |