CN104462373A - Personalized recommendation engine implementing method based on multiple Agents - Google Patents
Personalized recommendation engine implementing method based on multiple Agents Download PDFInfo
- Publication number
- CN104462373A CN104462373A CN201410749221.9A CN201410749221A CN104462373A CN 104462373 A CN104462373 A CN 104462373A CN 201410749221 A CN201410749221 A CN 201410749221A CN 104462373 A CN104462373 A CN 104462373A
- Authority
- CN
- China
- Prior art keywords
- agent
- user
- data
- data mining
- algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000007418 data mining Methods 0.000 claims abstract description 73
- 238000005516 engineering process Methods 0.000 claims abstract description 24
- 238000013473 artificial intelligence Methods 0.000 claims abstract description 13
- 238000001914 filtration Methods 0.000 claims description 27
- 238000013499 data model Methods 0.000 claims description 24
- 230000008569 process Effects 0.000 claims description 11
- 238000013461 design Methods 0.000 claims description 8
- 238000013500 data storage Methods 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 4
- 238000010801 machine learning Methods 0.000 claims description 4
- 235000013399 edible fruits Nutrition 0.000 claims description 3
- 239000012141 concentrate Substances 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 4
- 239000003795 chemical substances by application Substances 0.000 description 115
- 230000006399 behavior Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 4
- 239000000047 product Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000009412 basement excavation Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 230000011514 reflex Effects 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 239000013065 commercial product Substances 0.000 description 1
- 230000003750 conditioning effect Effects 0.000 description 1
- 239000012636 effector Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0207—Discounts or incentives, e.g. coupons or rebates
- G06Q30/0222—During e-commerce, i.e. online transactions
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a personalized recommendation engine implementing method based on multiple Agents. Personalized recommendation of a recommendation engine is realized by data mining, a personalized recommendation engine is implemented by a multi-Agent technology in the field of distributed artificial intelligence and is based on multi-Agent modeling, personated modeling is performed based on Agent, and a recommendation system is personalized. The personalized recommendation engine implementing method based on the multiple Agents is used for the two technical fields of data mining in distributed artificial intelligence and data mining in electronic commerce. A personalized recommendation effect of an electronic commerce recommendation system is achieved by intelligence, sociality, learnability and independence of the multi-Agent technology.
Description
Technical field
The invention belongs to computer data digging technology field, relate to distributed artificial intelligence technology, for personalized recommendation, for a kind of based on the personalized recommendation engine implementing method of multi-Agent.
Background technology
Multi-Agent technology belongs to distributed artificial intelligence field, adopts distributed Agent cooperative cooperating completion system target.Agent is a computational entity or program, can the change of sensing external environment, and the target according to oneself is made a response.Agent has intelligence, independence, autonomy, reactive, social ability and inferential capability.The Agent system majority of current structure is in experiment and academic research stage, and the related ends in personalized recommendations in E-business field is also little.
Jadex is an objectdistributed middleware development platform based on Java, adopts the framework (ServiceComponent Architecture is called for short SCA) based on assembly.Jadex provides BDI (Belief-Desire-Intention, conviction-hope-intent model) recommended engine framework, allow user easily design and development multi-Agent application.Its BDI V1 version achieves the inference engine of goal-oriented (object-oriented) based on Java and XML, BDI V2 version is substantially identical with V1 version, improve and be the rule-based reasoning again achieving BDI based on RETE regulation engine, Jadex BDI V3 Kernel is the latest edition of Jadex BDI inference engine, the V3 of latest edition introduces brand-new programmed method, conceal the more interior details of Jadex framework, transparent DLL (dynamic link library) is provided, the mode of XML is no longer adopted to configure BDI, but the mode adopting Java to explain realizes, the application relying on injection and reflex mechanism is a large feature.Be integrated with more Object--oriented method and theory in BDI V3 simultaneously, allow realization become more simple, support succession, POJO programming and rely on to inject.
Collaborative filtering (Collaborative Filtering, CF) proposed algorithm is the algorithm carrying out personalized recommendation based on the behavior of colony and group wisdom, extensively adopt in e-commerce website personalized recommendation system, may be used for the recommendation of commodity, books, film, blog and news etc.This algorithm mainly comprises two algorithms:
First, based on the recommendation (User CF) of user, namely recommend the commodity liked of similar users to user in brief, the calculating of similar users adopts Pearson correlation coefficient usually, and choosing of similar neighborhood generally adopts K-Neighborhoods (neighbours of fixed qty) or Threshold-based Neighborhoods (neighbours based on similarity threshold).
The second, based on the recommendation of article similarity (Item CF), according to all users, the preference of a certain article is calculated to the similarity of article, also can adopt Pearson correlation coefficient, recommend similar article to user.
Two kinds of algorithms respectively have excellent lacking, for unique user, the diversity that User CF algorithm is recommended is better than Item CF, and the article that User CF many recommendations similar users is liked, be generally popular commodity to user, Item CF can recommend the article of much similar but non-hot topic to user.Collaborative filtering does not point out how to fully utilize User CF and ItemCF algorithm.
The realization of collaborative filtering has Mahout in the community that increases income, it is the machine learning storehouse developed by Apache Software Foundation (Apache Software Foundation), it is one of project of the Hadoop ecosystem, realize the process of the Machine Learning of large data based on Hadoop, achieve User CF and Item CF algorithm, cluster algorithm and Method of The Classification Analysis etc.The Java that Mahout also provides a lot of linear algebra and statistics simultaneously realizes.The major applications of Mahout realizes based on the Map Reduce framework of Hadoop, and the Taste Library of Mahout also provides the realization of non-Hadoop simultaneously.
Summary of the invention
Technical matters to be solved by this invention is: the recommend method that existing e-commerce system uses is all sing on web daily record, and have no the report using artificial intelligence technology, the practical application of the multi-Agent technology in current distributed artificial intelligence field is not extensive, especially in the personalized recommendation of e-commerce field, there is no good achievement.Meanwhile, the analysis of current large data mining analysis, particularly personalized recommendation, adopts the application of multi-Agent technology little.
Technical scheme of the present invention is: a kind of personalized recommendation engine implementing method based on multi-Agent, recommended engine realizes personalized recommendation by data mining, the multi-Agent technology in distributed artificial intelligence field is adopted to realize personalized recommendation engine, described recommended engine is based on multi-agent modeling, design multiple Agent, comprise Agent as follows:
● User Agent: the input receiving user, and show final recommendation results;
● Recommendation Agent: the input according to user starts data mining, and control the process of whole data mining, the difference for user inputs to out the data of personalized recommendation;
● Data Mining Agent: perform data mining;
● Strategy Agent: execution algorithm decision-making, determines to adopt which kind of data mining algorithm;
● User CF Agent: perform User CF algorithm;
● Item CF Agent: perform Item CF algorithm;
● Search Agent: according to the result of data mining, searches for the details of the corresponding article of Result in data store;
● Recommendation Build Agent: the net result of data mining is formatted as the result that can show;
Adopt layer architecture in personalized recommendation engine, the superiors' application layer comprises above-mentioned Agent; Middle layer is data model layer, for encapsulating the operation to database, provides transparent database access operation; The bottom is data storage layer, stores the data needed for data mining.
The present invention adopts Jadex multi-Agent developing platform and modeling Agent, realizes collaborative filtering recommending in conjunction with Mahout machine learning storehouse, provides content recommendation by the concurrent cooperation of multi-Agent.
Further, in data mining process, Strategy Agent analyzes the preference of active user, and the self-similarity decision-making according to active user's preference adopts User CF algorithm or Item CF algorithm, and described data mining process is as follows:
1) Data Mining Agent receives the data mining request of Recommendation Agent;
2) Data Mining Agent asks Strategy Agent to carry out algorithm decision-making, calculate the self-similarity of all preference article of active user, self-similarity shows as the standard deviation sigma of all preferences of active user, setting decision-making threshold values, σ is less than decision-making threshold values, then self-similarity is comparatively large, and it is concentrated for namely thinking that the preference of active user compares, and adopts Item CF algorithm; When σ is greater than decision-making threshold values, then self-similarity is less, thinks that the preference of active user is not concentrated, and adopts User CF algorithm; Decision-making threshold values chooses setting according to the difference of recommended scene and data set;
3) when employing Item CF algorithm, DataMining Agent asks ItemCF Agent to adopt Item CF algorithm to carry out data mining, adopts Log LikeliHood to calculate the similarity of article to be recommended and active user's preference, and provides recommendation collection;
4) when employing User CF algorithm, Data Mining Agent asks User CF Agent to adopt User CF algorithm to excavate, select K the neighbor user the most similar to active user's preference, K is defaulted as 10, with the preference article of this K neighbor user for recommending article basis, adopt the similarity of Pearson correlation coefficient calculated recommendation article and active user's preference, provide to active user and recommend collection;
5) gained is recommended to assemble fruit and is returned to Recommender Agent via Data Mining Agent again;
6) Recommendation Agent triggers Search Agent calling data model layer, and the Item Number of the recommendation collection produced by Data Mining Agent is converted into more detailed Item Information;
7) details of the article recommending collection and Search Agent to return are integrated into the net result collection for showing by Recommender Agent.
The recommending data that recommended engine of the present invention is suitable for comprises any collaborative filtering that can adopt and carries out the data of recommending, require data through cleaning, each data tuple requires to have three at least: Customs Assigned Number, Item Number and preference value, preference value is through normalized.
In the commending system of current e-commerce field, classical way adopts the data digging method of sing on web daily record to carry out commercial product recommending, and also not based on the recommendation of the multi-Agent technology of artificial intelligence, main cause is:
First, personalized recommendation algorithm, such as collaborative filtering recommending, be all belong to information-theoretical mathematical algorithm based on correlation rule recommendation etc., just starting at the beginning of 21 century to obtain large-scale application in e-commerce field, wherein, the data mining technology emerging field especially that personalized recommendation is relevant, the basic object-oriented modeling adopting maturation, does not also consider and adopts multi-Agent technology modeling.Corresponding, multi-Agent technology is as one of distributed artificial intelligence (DAI) field latest technology, its main application at present concentrates on artificial intelligence field, does not yet have large-scale application to the rich experiences in existing e-commerce system.The how appropriate design Agent key that to be multi-Agent technology effectively run.
Secondly, existing Technologies of Recommendation System in E-Commerce also needs more intelligent framework to overcome more existing limitation: first, lack of wisdom, the data digging system of off-line is often selected to take single proposed algorithm to all users, such as collaborative filtering recommending, content-based recommendation or based on correlation rule recommendation etc., can not choose different proposed algorithms according to the preference of different user, this is difficult to the personalized recommendation demand meeting user.The second, lack adaptability, the preference of user in constantly change, and the commending system of dynamic conditioning Generalization bounds can not demonstrate the adaptability lacked environmental change.
The present invention is devoted in conjunction with these two technical fields of the data mining in distributed artificial intelligence and ecommerce just, a kind of personalized recommendation modeling method based on multi-Agent is proposed, intelligent by multi-Agent technology, social, study property and independence bring more personalized recommendation effect to Technologies of Recommendation System in E-Commerce.
Recommended engine of the present invention realizes personalized recommendation by the mode of multi-Agent, it is advantageous that:
(1) each several part has independence.By each functional module of recommended engine, as interactive module, data-mining module, algorithm decision-making module etc. are designed to the Agent with intelligence, this modeling personalized and design make each Agent have respective conviction, hope and plan, can independently complete respective task.
(2) there is better adaptability.Strategy Agent dynamically can choose proposed algorithm according to the feature of user preference, when the preference of user there occurs change, and also can adjustment algorithm adaptively.Compare traditional recommended engine and adopt single Generalization bounds and can not adjustment algorithm dynamically, this is progressive part of the present invention.
(3) there is good extensibility.Recommended engine of the present invention devises User CF Agent and Item CF Agent, holds different proposed algorithms separately, separately completes.Later stage can also expand more algorithm.Based on the modeling of multi-Agent, compare the recommended engine adopting object-oriented way modeling, the degree of coupling is lower, brings better extensibility.
(4) there is distributed nature.Multi-Agent cooperation recommendation can realize distributed concurrent, improves the performance of recommended engine, can provide real-time recommendation on line.
Accompanying drawing explanation
Fig. 1 is the layer architecture schematic diagram of recommended engine of the present invention.
Fig. 2 is the detailed design class figure of recommended engine of the present invention.
Fig. 3 is the precedence diagram that recommended engine of the present invention carries out collaborative filtering recommending calculating.
Fig. 4 is the data mining process constitutional diagram of recommended engine of the present invention.
Embodiment
Target of the present invention is applied in the personalized recommendation of ecommerce by the multi-Agent technology in distributed artificial intelligence.Commending system in traditional ecommerce has some limitations, and the present invention adopts multi-Agent technology, carries out personalizing modeling based on Agent, makes commending system more personalized.Adopting multi-Agent technology to develop personalized recommendation engine, is a kind of innovation.
Collaborative Filtering Recommendation Algorithm is the algorithm that a kind of user group's of utilization behavior and wisdom carry out recommending, this algorithm is widely used in e-commerce website, comprise User CF and Item CF two kinds of algorithms, the best results of recommending is carried out in practice discovery in conjunction with User CF and Item CF.The present invention, in conjunction with multi-Agent technology and collaborative filtering, realizes a kind of new personalized recommendation engine.
Specific embodiment of the invention is described below.As shown in Figure 1, recommended engine of the present invention is main employing layer architecture:
(1) Data Storage Layer data storage layer
Data storage layer is used for stores information, user profile and user preference information, here the data that Item Information is corresponding can be that any collaborative filtering that can adopt carries out the data of recommending, such as commodity data, cinematic data, news data, book data etc., here data are required through cleaning, each data tuple requires to have three at least: Customs Assigned Number, Item Number and preference value.Preference will through normalized, to ensure the accuracy of recommendation results.In this recommended engine, adopt MySQL data to store, and have employed the Indexing Mechanism that MySQL database provides, improve the query performance of data.
(2) Data Model Layer data model layer
Data model layer is based on data storage layer, and data model layer provides the data model (Model) encapsulating underlying database operation, for Application Layer provides transparent operation-interface, is optimized the inquiry of data simultaneously.This layer mainly comprises three Model:
A) Commodity Data Model: product data model, provides and retrieve product data, increases, the interface deleted and revise.
B) User Data Model: user data model, provides the interface retrieved user profile.
C) Preference Data Model: preference data model, provides the interface retrieved user preference.
(3) Application Layer application layer
Application layer comprises 8 Agent:
A) User Agent (User interface Agent): the input receiving user, the behavior of monitoring users, and show final recommendation results.
B) Recommendation Agent (adverse Agent): the process controlling whole data mining, the difference for user inputs to out the data of personalized recommendation; Recommendation Agent is coordination and control person, obtain the behavior of user, request Data Mining Agent adopts collaborative filtering to calculate raw recommendation collection, Search Agent is asked to search out the details of recommendation again, ask Recommendation Build Agent to construct the result set that can show again, and final result set is returned to User Agent.
C) Data Mining Agent (data mining Agent): perform data mining.Data Mining Agent asks Strategy Agent to carry out algorithm decision-making, and triggers User CF Agent or Item CF Agent carries out collaborative filtering calculating.
D) Strategy Agent (tactful Agent): execution algorithm decision-making; Strategy Agent calculates the self-similarity of active user's preference, if self-similarity greatly, adopts ItemCF, acquiescence adopts UserCF algorithm.
E) User CF Agent (the collaborative filtering Agent based on user): perform collaborative filtering (User CF, the User Collaborative Filtering) algorithm based on user;
F) Item CF Agent (the collaborative filtering Agent based on article): perform collaborative filtering (Item CF, the Item Collaborative Filtering) algorithm based on article;
G) Search Agent (search Agent): according to the result of data mining, is generally the Item Number set of recommendation, searches for the details of article by number in data store;
H) Recommendation Build Agent (recommending to show Agent): the net result of data mining is formatted as the result that can show.
The class detailed design of recommended engine, as Fig. 2, based on the application of Java, mainly contains 8 bags, and as shown in Figure 2, except Common bag and Model bag, other 6 each bags of bag are exactly an Agent structure based on BDI to the relation of parlor.Multi-Agent be the Goal trigger mechanism based on Service provided by Jadex BDI V3 alternately, each Goal is JavaPOJO, adopt Goal to be a kind of design of loose coupling as agency, bind different Goal to different Service based on dependence injection and reflex mechanism.
(1) Common bag: providing Application.xml, is the startup entrance of recommended engine.
(2) Model bag: define Data Model (data model), comprise Item Data Model (product data model), User Data Model (user data model), Rating Data Model (score data model), also have DataSource Factory (data source warehouse) simultaneously, create the link of MySQL.
(3) Recommender bag: define Recommender BDI (nominator BDI), RecommendServiceInterface (recommendation service interface) and Recommendation Goal (recommendation target), RecommendServiceInterface is the IRecommend Service in Fig. 2, Recommender bag is as Controller, transparent recommendation interface is provided, Recommendation Build BDI (recommending to show BDI) can be asked simultaneously, DataMiningBDI (data mining BDI) and Search BDI (search BDI) recommends to calculate.
(4) User bag: define User BDI (user BDI), User BDI can create the input that User Input GUI (user's tablet pattern user interface) obtains user when starting, and request Recommender BDI provides recommendation results.
(5) DataMining bag: this bag comprises DataMining BDI (data mining BDI), ItemCF BDI (the collaborative filtering BDI based on article), UserCF BDI (the collaborative filtering BDI based on user), three provides DataMiningService (data mining service), DataMining BDI can ask Strategy BDI (tactful BDI) to provide proposed algorithm decision-making, according to decision requests ItemCF BDI, or UserCF BDI, calculates and returns results.
(6) Search bag: define ISearch Service (search service interface) in this bag, SearchBDI (search BDI) and SearchGoal (search target), this BDI Agent understand the details of the article recommended in search database.
(7) Build bag: define IRecommendation Build Service (recommending to show service interface) in this bag, Recommendation Build BDI (recommending to show BDI) and Recommendation Build Goal (recommending to show target), this BDI gathers all result of calculation, and provides the recommendation results collection that may be used for showing.
Recommended engine carries out the precedence diagram of collaborative filtering recommending calculating as Fig. 3, in figure, each object represents a BDIAgent, the corresponding User Agent of UserBDI, the corresponding Recommendation Agent of RecommenderBDI, the corresponding DataMining Agent of DataMiningBDI, the corresponding RecommendationBuild Agent of SearchBDI corresponding Search Agent, RecommBuildBDI.Interaction sequences between each Agent is as follows:
(1) UserBDI carries out collaborative filtering recommending by the mode request RecommenderBDI of Dispatch Goal (invocation target).
(2), after RecommenderBDI receives request, the Intention (intention) in RecommendPlan, RecommendPlan and BDI is triggered.
(3) RecommenderBDI is by the mode request DataMiningBDI of DispatchSubGoal (calling sub-goal), carries out data mining.
(4) after the initial result set of excavation to be returned, then ask SearchBDI according to initial result set, provide the details of recommended article.
(5) RecommBuildBDI is asked to carry out each result set calculated being integrated into the recommended project that can show and return results collection.
(6) result set obtained is returned to UserBDI by last RecommenderBDI.
RecommenderBDI is effector in system and coordinator, is triggered the behavior of other Agent by the mode of Goal.
Data mining process is the core of recommended engine of the present invention, and the mode of excavation excavates in conjunction with the User CF algorithm of collaborative filtering and Item CF algorithm.The decision-making of excavating is the self-similarity calculating the current preference of user, if self-similarity greatly just adopts Item CF, the little just acquiescence of self-similarity adopts User CF.Self-similarity is by calculating the standard deviation of all scoring preferences of active user.
Data mining process constitutional diagram if Fig. 4 is recommended engine of the present invention:
1) Data Mining Agent receives the data mining request of Recommendation Agent;
2) Data Mining Agent asks Strategy Agent to carry out algorithm decision-making, calculates the self-similarity of all preference article of active user.Be generally a normalized scoring to the tolerance of article preference, as 0 ~ 5 point, self-similarity adopts the standard deviation calculating all preferences of user in this recommended engine, represents with σ.Setting decision-making threshold values, when σ is less than decision-making threshold values, then self-similarity is comparatively large, and it is concentrated for namely thinking that the preference of user compares, and adopts Item CF algorithm to have better recommendation effect.When σ is greater than decision-making threshold values, then self-similarity is less, and the preference of user is not concentrated, then adopt User CF algorithm can obtain better recommendation effect.
Decision-making threshold values needs to choose according to the difference of recommended scene and data set, can not be too large, can cause so much needing carrying out with Item CF algorithm the user that recommends, have employed User CF; In like manner, can not be too little, can cause so much needing to have employed Item CF algorithm with the user that User CF algorithm carries out recommending.In a word, choosing of decision-making threshold values needs after cleaning data set, makes a rational value.In the embodiment of this recommended engine, we recommend to carry out testing for film, the 1000000 film score data having selected MovieLens website to provide, and finally the decision-making threshold values of σ are decided to be 0.8 and can produce comparatively reasonably recommendation results collection;
3) when employing Item CF algorithm, DataMining Agent asks ItemCF Agent to adopt Item CF algorithm to carry out data mining, adopt LogLikeliHood to calculate similarity that article are recommended by institute, and provides recommendation and collect;
4) when employing User CF algorithm, Data Mining Agent asks User CF Agent to adopt User CF algorithm to excavate, adopt Pearson correlation coefficient calculate recommend article similarity, and select K the neighbours (K be defaulted as 10) the most similar to active user's preference, based on the preference article of this K neighbour, provide and recommend collection;
5) gained is recommended to assemble fruit and is returned to Recommender Agent via Data Mining Agent again;
6) Recommendation Agent triggers Search Agent calling data model layer, and the recommendation Item Number that Data Mining Agent produces is converted into more detailed Item Information;
7) details of recommending collection and Search Agent to return are integrated into the net result collection for showing by Recommender Agent.
Claims (5)
1. the personalized recommendation engine implementing method based on multi-Agent, recommended engine realizes personalized recommendation by data mining, it is characterized in that adopting the multi-Agent technology in distributed artificial intelligence field to realize personalized recommendation engine, described recommended engine is based on multi-agent modeling, design multiple Agent, comprise Agent as follows:
● User Agent: the input receiving user, and show final recommendation results;
● Recommendation Agent: the input according to user starts data mining, and control the process of whole data mining, the difference for user inputs to out the data of personalized recommendation;
● Data Mining Agent: perform data mining;
● Strategy Agent: execution algorithm decision-making, determines to adopt which kind of data mining algorithm;
● User CF Agent: perform User CF algorithm;
● Item CF Agent: perform Item CF algorithm;
● Search Agent: according to the result of data mining, searches for the details of the corresponding article of Result in data store;
● Recommendation Build Agent: the net result of data mining is formatted as the result that can show;
Adopt layer architecture in personalized recommendation engine, the superiors' application layer comprises above-mentioned Agent; Middle layer is data model layer, for encapsulating the operation to database, provides transparent database access operation; The bottom is data storage layer, stores the data needed for data mining.
2. a kind of personalized recommendation engine implementing method based on multi-Agent according to claim 1, it is characterized in that adopting Jadex multi-Agent developing platform and modeling Agent, realize collaborative filtering recommending in conjunction with Mahout machine learning storehouse, provide content recommendation by the concurrent cooperation of multi-Agent.
3. a kind of personalized recommendation engine implementing method based on multi-Agent according to claim 1, it is characterized in that in data mining process, Strategy Agent analyzes the preference of active user, self-similarity decision-making according to active user's preference adopts User CF algorithm or Item CF algorithm, and described data mining process is as follows:
1) Data Mining Agent receives the data mining request of Recommendation Agent;
2) Data Mining Agent asks Strategy Agent to carry out algorithm decision-making, calculate the self-similarity of all preference article of active user, self-similarity shows as the standard deviation sigma of all preferences of active user, setting decision-making threshold values, σ is less than decision-making threshold values, then self-similarity is comparatively large, and it is concentrated for namely thinking that the preference of active user compares, and adopts Item CF algorithm; When σ is greater than decision-making threshold values, then self-similarity is less, thinks that the preference of active user is not concentrated, and adopts User CF algorithm; Decision-making threshold values chooses setting according to the difference of recommended scene and data set;
3) when employing Item CF algorithm, DataMining Agent asks ItemCF Agent to adopt Item CF algorithm to carry out data mining, adopts Log LikeliHood to calculate the similarity of article to be recommended and active user's preference, and provides recommendation collection;
4) when employing User CF algorithm, Data Mining Agent asks User CF Agent to adopt User CF algorithm to excavate, select K the neighbor user the most similar to active user's preference, K is defaulted as 10, with the preference article of this K neighbor user for recommending article basis, adopt the similarity of Pearson correlation coefficient calculated recommendation article and active user's preference, provide to active user and recommend collection;
5) gained is recommended to assemble fruit and is returned to Recommender Agent via Data Mining Agent again;
6) Recommendation Agent triggers Search Agent calling data model layer, and the Item Number of the recommendation collection produced by Data Mining Agent is converted into more detailed Item Information;
7) details of the article recommending collection and Search Agent to return are integrated into the net result collection for showing by Recommender Agent.
4. a kind of personalized recommendation engine implementing method based on multi-Agent according to claim 3, it is characterized in that recommending data that recommended engine is suitable for comprises any collaborative filtering that can adopt and carries out the data of recommending, require data through cleaning, each data tuple requires to have three at least: Customs Assigned Number, Item Number and preference value, preference value is through normalized.
5. a kind of personalized recommendation engine implementing method based on multi-Agent according to claim 1 or 2 or 3, it is characterized in that the result of the data mining of Search Agent institute foundation is the Item Number set of recommending to concentrate article, in data store, search for the details of corresponding article by Item Number.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410749221.9A CN104462373A (en) | 2014-12-09 | 2014-12-09 | Personalized recommendation engine implementing method based on multiple Agents |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410749221.9A CN104462373A (en) | 2014-12-09 | 2014-12-09 | Personalized recommendation engine implementing method based on multiple Agents |
Publications (1)
Publication Number | Publication Date |
---|---|
CN104462373A true CN104462373A (en) | 2015-03-25 |
Family
ID=52908408
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410749221.9A Pending CN104462373A (en) | 2014-12-09 | 2014-12-09 | Personalized recommendation engine implementing method based on multiple Agents |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104462373A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106960039A (en) * | 2017-03-28 | 2017-07-18 | 西安电子科技大学 | Socialization search engine system based on whole sub- multi-Agent |
CN107067289A (en) * | 2016-10-28 | 2017-08-18 | 广东亿迅科技有限公司 | A kind of personal marketing commending system |
CN107169572A (en) * | 2016-12-23 | 2017-09-15 | 福州大学 | A kind of machine learning Service Assembly method based on Mahout |
CN107220721A (en) * | 2017-06-21 | 2017-09-29 | 苏州发飚智能科技有限公司 | The agreed-upon price method and system of booking rooms that a kind of Multi -Agent is supported |
EP3553676A4 (en) * | 2016-12-27 | 2019-11-06 | Huawei Technologies Co., Ltd. | Smart recommendation method and terminal |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050210285A1 (en) * | 2004-03-18 | 2005-09-22 | Microsoft Corporation | System and method for intelligent recommendation with experts for user trust decisions |
CN101271558A (en) * | 2008-05-16 | 2008-09-24 | 华东师范大学 | Multi-policy commercial product recommending system based on context information |
CN102208086A (en) * | 2010-03-31 | 2011-10-05 | 北京邮电大学 | Field-oriented personalized intelligent recommendation system and implementation method |
-
2014
- 2014-12-09 CN CN201410749221.9A patent/CN104462373A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050210285A1 (en) * | 2004-03-18 | 2005-09-22 | Microsoft Corporation | System and method for intelligent recommendation with experts for user trust decisions |
CN101271558A (en) * | 2008-05-16 | 2008-09-24 | 华东师范大学 | Multi-policy commercial product recommending system based on context information |
CN102208086A (en) * | 2010-03-31 | 2011-10-05 | 北京邮电大学 | Field-oriented personalized intelligent recommendation system and implementation method |
Non-Patent Citations (1)
Title |
---|
LGRCYANNY: "/lgrcyanny/Minion", 《HTTPS://GITHUB.COM/LGRCYANNY/MINION》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107067289A (en) * | 2016-10-28 | 2017-08-18 | 广东亿迅科技有限公司 | A kind of personal marketing commending system |
CN107169572A (en) * | 2016-12-23 | 2017-09-15 | 福州大学 | A kind of machine learning Service Assembly method based on Mahout |
EP3553676A4 (en) * | 2016-12-27 | 2019-11-06 | Huawei Technologies Co., Ltd. | Smart recommendation method and terminal |
CN106960039A (en) * | 2017-03-28 | 2017-07-18 | 西安电子科技大学 | Socialization search engine system based on whole sub- multi-Agent |
CN106960039B (en) * | 2017-03-28 | 2019-12-10 | 西安电子科技大学 | Social search engine system based on whole multi-Agent |
CN107220721A (en) * | 2017-06-21 | 2017-09-29 | 苏州发飚智能科技有限公司 | The agreed-upon price method and system of booking rooms that a kind of Multi -Agent is supported |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Sunhare et al. | Internet of things and data mining: An application oriented survey | |
Afsar et al. | Reinforcement learning based recommender systems: A survey | |
Felfernig et al. | An overview of recommender systems in the internet of things | |
Ortega et al. | Recommending items to group of users using matrix factorization based collaborative filtering | |
CN112989064B (en) | Recommendation method for aggregating knowledge graph neural network and self-adaptive attention | |
CN104462373A (en) | Personalized recommendation engine implementing method based on multiple Agents | |
Altulyan et al. | A survey on recommender systems for Internet of Things: techniques, applications and future directions | |
Pushpam et al. | Overview on data mining in social media | |
Zhu et al. | Query-based interactive recommendation by meta-path and adapted attention-gru | |
Wang et al. | Session-based graph convolutional ARMA filter recommendation model | |
Xu et al. | CLDG: Contrastive learning on dynamic graphs | |
Manjunath et al. | Combining heterogeneous classifiers for relational databases | |
Li et al. | Knowledge graph-based recommendation method for cold chain logistics | |
Li et al. | Multi-behavior enhanced heterogeneous graph convolutional networks recommendation algorithm based on feature-interaction | |
Sattar et al. | Graph neural network for context-aware recommendation | |
Sheng et al. | Personalized recommendation of location-based services using spatio-temporal-aware long and short term neural network | |
Zhang et al. | Extreme learning machines’ ensemble selection with GRASP | |
US20180096255A1 (en) | Architecture and method for providing insights in networks domain | |
Subramani et al. | Mining actionable knowledge using reordering based diversified actionable decision trees | |
Kalanat | An overview of actionable knowledge discovery techniques | |
Tu et al. | A new service recommendation method for agricultural industries in the fog-based Internet of Things environment using a hybrid meta-heuristic algorithm | |
Song et al. | Dgcn-rs: A dilated graph convolutional networks jointly modelling relation and semantic for multi-event forecasting | |
Pariserum Perumal et al. | FIRMACA-Fuzzy intelligent recommendation model using ant clustering algorithm for social networking | |
Zhou et al. | A factorisation-based recommendation model for customised products configuration design | |
Gosain et al. | Investigating structural metrics for understandability prediction of data warehouse multidimensional schemas using machine learning techniques |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150325 |
|
WD01 | Invention patent application deemed withdrawn after publication |